You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@iceberg.apache.org by gi...@apache.org on 2026/08/14 18:19:41 UTC

(iceberg-rust) branch gh-pages updated: deploy: e50a094cb119766b8371e402ba2dd40d085f829d

This is an automated email from the ASF dual-hosted git repository.

github-actions[bot] pushed a commit to branch gh-pages
in repository https://gitbox.apache.org/repos/asf/iceberg-rust.git


The following commit(s) were added to refs/heads/gh-pages by this push:
     new 49cc91e07 deploy: e50a094cb119766b8371e402ba2dd40d085f829d
49cc91e07 is described below

commit 49cc91e072056128a9252eeed509f4008a9ccdd1
Author: CTTY <CT...@users.noreply.github.com>
AuthorDate: Fri Aug 14 18:19:35 2026 +0000

    deploy: e50a094cb119766b8371e402ba2dd40d085f829d
---
 api/help.html                                      |    2 +-
 .../desc/{2396db35545a.js => 78716288c756.js}      |    2 +-
 .../entry/{08e6ff30364b.js => 47c7d3e27645.js}     |    2 +-
 .../entry/{7146aa56d62e.js => b691aab07b2c.js}     |    2 +-
 .../function/{1aaa1acc45ae.js => 4991e2ae7049.js}  |    2 +-
 .../path/{1445f87d5e24.js => b4dd74ae2986.js}      |    2 +-
 api/search.index/root.js                           |    2 +-
 .../type/{2d00f643fc8b.js => 5a27e3ee9475.js}      |    2 +-
 api/settings.html                                  |    2 +-
 api/src/iceberg/arrow/reader/pipeline.rs.html      | 3816 +++++++++++---------
 10 files changed, 2150 insertions(+), 1684 deletions(-)

diff --git a/api/help.html b/api/help.html
index fba61dabe..393e3c7ff 100644
--- a/api/help.html
+++ b/api/help.html
@@ -1 +1 @@
-<!DOCTYPE html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><meta name="generator" content="rustdoc"><meta name="description" content="Documentation for Rustdoc"><title>Help</title><script>if(window.location.protocol!=="file:")document.head.insertAdjacentHTML("beforeend","SourceSerif4-Regular-6b053e98.ttf.woff2,FiraSans-Italic-81dc35de.woff2,FiraSans-Regular-0fe48ade.woff2,FiraSans-MediumItalic-ccf7e434.woff2,FiraSans-M [...]
\ No newline at end of file
+<!DOCTYPE html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><meta name="generator" content="rustdoc"><meta name="description" content="Documentation for Rustdoc"><title>Help</title><script>if(window.location.protocol!=="file:")document.head.insertAdjacentHTML("beforeend","SourceSerif4-Regular-6b053e98.ttf.woff2,FiraSans-Italic-81dc35de.woff2,FiraSans-Regular-0fe48ade.woff2,FiraSans-MediumItalic-ccf7e434.woff2,FiraSans-M [...]
\ No newline at end of file
diff --git a/api/search.index/desc/2396db35545a.js b/api/search.index/desc/78716288c756.js
similarity index 99%
rename from api/search.index/desc/2396db35545a.js
rename to api/search.index/desc/78716288c756.js
index 53104a03d..359832c97 100644
--- a/api/search.index/desc/2396db35545a.js
+++ b/api/search.index/desc/78716288c756.js
@@ -1 +1 @@
-rd_("BkThe catalog type that this builder creates.AkThe associated writer type.B`The associated file writer type.AlReturn type of this visitor.0BkReturn type of this visitor on arrow field.BlReturn type of this visitor on arrow schema.BaLocal filesystem storage factory.B`HuggingFace Hub storage factory.B`HuggingFace Hub storage variant.CgMatch fields by their ID stored in Arrow field metadataAjContains the success value0CoOr predicate, for example, <code>a &gt; 10 OR b &lt; 20</code>.EnA [...]
\ No newline at end of file
+rd_("BkThe catalog type that this builder creates.AkThe associated writer type.B`The associated file writer type.AlReturn type of this visitor.0BkReturn type of this visitor on arrow field.BlReturn type of this visitor on arrow schema.BaLocal filesystem storage factory.B`HuggingFace Hub storage factory.B`HuggingFace Hub storage variant.CgMatch fields by their ID stored in Arrow field metadataAjContains the success value0CoOr predicate, for example, <code>a &gt; 10 OR b &lt; 20</code>.EnA [...]
\ No newline at end of file
diff --git a/api/search.index/entry/08e6ff30364b.js b/api/search.index/entry/47c7d3e27645.js
similarity index 84%
rename from api/search.index/entry/08e6ff30364b.js
rename to api/search.index/entry/47c7d3e27645.js
index 631819fd4..ec5cb52fd 100644
--- a/api/search.index/entry/08e6ff30364b.js
+++ b/api/search.index/entry/47c7d3e27645.js
@@ -1 +1 @@
-rd_("Aj[3920,18,146,1123,0,0,0,0]Ag[7946,5,7947,0,0,0,0,0]Aj[8306,5,8307,5307,0,0,0,0]An[3920,19,2316,7754,7332,0,0,0]000Aj[3920,5,2316,7300,0,0,0,0]Ai[3920,18,146,383,0,0,0,0]Ak[8105,18,8106,3843,0,0,0,0]Aj[3920,8,2316,5290,0,0,0,0]Aj[3920,7,2620,3088,0,0,0,0]Ad[8306,3,0,0,0,0,0,0]Ag[3920,7,7732,0,0,0,0,0]An[3920,14,2316,7754,7332,0,0,0]0An[3920,13,2316,7304,7081,0,0,0]An[3920,12,1132,3631,7982,0,0,0]Ba[3920,13,1132,3692,8355,7982,0,0]An[3920,13,2316,2812,2580,0,0,0]An[3920,13,2316,3926 [...]
\ No newline at end of file
+rd_("Aj[3920,18,146,1123,0,0,0,0]Ag[7946,5,7947,0,0,0,0,0]Aj[8306,5,8307,5307,0,0,0,0]An[3920,19,2316,7754,7332,0,0,0]000Aj[3920,5,2316,7300,0,0,0,0]Ai[3920,18,146,383,0,0,0,0]Ak[8105,18,8106,3843,0,0,0,0]Aj[3920,8,2316,5290,0,0,0,0]Aj[3920,7,2620,3088,0,0,0,0]Ad[8306,3,0,0,0,0,0,0]Ag[3920,7,7732,0,0,0,0,0]An[3920,14,2316,7754,7332,0,0,0]0An[3920,13,2316,7304,7081,0,0,0]An[3920,12,1132,3631,7982,0,0,0]Ba[3920,13,1132,3692,8355,7982,0,0]An[3920,13,2316,2812,2580,0,0,0]An[3920,13,2316,3926 [...]
\ No newline at end of file
diff --git a/api/search.index/entry/7146aa56d62e.js b/api/search.index/entry/b691aab07b2c.js
similarity index 92%
rename from api/search.index/entry/7146aa56d62e.js
rename to api/search.index/entry/b691aab07b2c.js
index 544526b11..ffcb31ce6 100644
--- a/api/search.index/entry/7146aa56d62e.js
+++ b/api/search.index/entry/b691aab07b2c.js
@@ -1 +1 @@
-rd_("An[3920,17,3921,3837,7188,0,0,0]An[3920,17,3754,3754,8135,0,0,0]An[3920,17,6789,6789,7778,0,0,0]An[3920,17,2316,4194,7067,0,0,0]An[3920,17,2316,4194,8337,0,0,0]An[3920,17,2620,3715,7874,0,0,0]0An[3920,15,1512,1512,7787,0,0,0]An[8306,15,8307,8307,8215,0,0,0]0An[8306,15,8307,8307,7202,0,0,0]An[3920,15,2620,3088,7191,0,0,0]3An[3920,15,3921,3047,3148,0,0,0]An[8173,15,3049,3049,3149,0,0,0]An[3920,15,1512,5297,4982,0,0,0]An[3920,15,1512,5297,7185,0,0,0]65An[3920,15,2316,7304,7047,0,0,0]An [...]
\ No newline at end of file
+rd_("An[3920,17,3921,3837,7188,0,0,0]An[3920,17,3754,3754,8135,0,0,0]An[3920,17,6789,6789,7778,0,0,0]An[3920,17,2316,4194,7067,0,0,0]An[3920,17,2316,4194,8337,0,0,0]An[3920,17,2620,3716,7874,0,0,0]0An[3920,15,1512,1512,7787,0,0,0]An[8306,15,8307,8307,8215,0,0,0]0An[8306,15,8307,8307,7202,0,0,0]An[3920,15,2620,3088,7191,0,0,0]3An[3920,15,3921,3048,3148,0,0,0]An[8173,15,3047,3047,3149,0,0,0]An[3920,15,1512,5297,4982,0,0,0]An[3920,15,1512,5297,7185,0,0,0]65An[3920,15,2316,7304,7047,0,0,0]An [...]
\ No newline at end of file
diff --git a/api/search.index/function/1aaa1acc45ae.js b/api/search.index/function/4991e2ae7049.js
similarity index 96%
rename from api/search.index/function/1aaa1acc45ae.js
rename to api/search.index/function/4991e2ae7049.js
index 18f127679..6bff18836 100644
--- a/api/search.index/function/1aaa1acc45ae.js
+++ b/api/search.index/function/4991e2ae7049.js
@@ -1 +1 @@
-rd_("Bf[\"{{{BIGb{BFLb}}{BIGb{BFLb}}}IOf}\",[]]Bf[\"{{{BIGb{CKCj}}{BIGb{CKCj}}}IOf}\",[]]Bf[\"{{{BIGb{CHCn}}{BIGb{CHCn}}}IOf}\",[]]Bf[\"{{{BIGb{BFMj}}{BIGb{BFMj}}}IOf}\",[]]Bf[\"{{{BIGb{BLN`}}{BIGb{BLN`}}}IOf}\",[]]Bf[\"{{{BIGb{CKEn}}{BIGb{CKEn}}}IOf}\",[]]Bf[\"{{{BIGb{CDAd}}{BIGb{CDAd}}}IOf}\",[]]Bf[\"{{{BIGb{BLNf}}{BIGb{BLNf}}}IOf}\",[]]Bf[\"{{{BIGb{BFOb}}{BIGb{BFOb}}}IOf}\",[]]Bf[\"{{{BIGb{CHBd}}{BIGb{CHBd}}}IOf}\",[]]Cg[\"{{{BIGb{{CLLl{c}}}}{BIGb{{CLLl{c}}}}}IOfBFNd}\",[\"T\"]]Cg[\"{ [...]
\ No newline at end of file
+rd_("Bf[\"{{{BIGb{BFLb}}{BIGb{BFLb}}}IOf}\",[]]Bf[\"{{{BIGb{CKCj}}{BIGb{CKCj}}}IOf}\",[]]Bf[\"{{{BIGb{CHCn}}{BIGb{CHCn}}}IOf}\",[]]Bf[\"{{{BIGb{BFMj}}{BIGb{BFMj}}}IOf}\",[]]Bf[\"{{{BIGb{BLN`}}{BIGb{BLN`}}}IOf}\",[]]Bf[\"{{{BIGb{CKEn}}{BIGb{CKEn}}}IOf}\",[]]Bf[\"{{{BIGb{CDAd}}{BIGb{CDAd}}}IOf}\",[]]Bf[\"{{{BIGb{BLNf}}{BIGb{BLNf}}}IOf}\",[]]Bf[\"{{{BIGb{BFOb}}{BIGb{BFOb}}}IOf}\",[]]Bf[\"{{{BIGb{CHBd}}{BIGb{CHBd}}}IOf}\",[]]Cg[\"{{{BIGb{{CLLl{c}}}}{BIGb{{CLLl{c}}}}}IOfBFNd}\",[\"T\"]]Cg[\"{ [...]
\ No newline at end of file
diff --git a/api/search.index/path/1445f87d5e24.js b/api/search.index/path/b4dd74ae2986.js
similarity index 95%
rename from api/search.index/path/1445f87d5e24.js
rename to api/search.index/path/b4dd74ae2986.js
index f1df301aa..201df3053 100644
--- a/api/search.index/path/1445f87d5e24.js
+++ b/api/search.index/path/b4dd74ae2986.js
@@ -1 +1 @@
-rd_("g[17,\"\"]000f[1,\"\"]0Aa[10,\"bnum::cast\"]Bn[15,\"iceberg_storage_opendal::OpenDalStorage\"]Eg[5,\"tower_http::follow_redirect::policy::or\",\"tower_http::follow_redirect::policy::or\"]1Ce[15,\"iceberg_storage_opendal::OpenDalStorageFactory\"]4Bb[2,\"iceberg::io::storage::config\"]m[2,\"iceberg\"]Ai[2,\"iceberg::encryption\"]2Am[2,\"iceberg_storage_opendal\"]8Ei[5,\"tower_http::follow_redirect::policy::and\",\"tower_http::follow_redirect::policy::and\"]A`[10,\"core::any\"]Ao[5,\"a [...]
\ No newline at end of file
+rd_("g[17,\"\"]000f[1,\"\"]0Aa[10,\"bnum::cast\"]Bn[15,\"iceberg_storage_opendal::OpenDalStorage\"]Eg[5,\"tower_http::follow_redirect::policy::or\",\"tower_http::follow_redirect::policy::or\"]1Ce[15,\"iceberg_storage_opendal::OpenDalStorageFactory\"]4Bb[2,\"iceberg::io::storage::config\"]m[2,\"iceberg\"]Ai[2,\"iceberg::encryption\"]2Am[2,\"iceberg_storage_opendal\"]8Ei[5,\"tower_http::follow_redirect::policy::and\",\"tower_http::follow_redirect::policy::and\"]A`[10,\"core::any\"]Ao[5,\"a [...]
\ No newline at end of file
diff --git a/api/search.index/root.js b/api/search.index/root.js
index aece50988..d4c2bbd6e 100644
--- a/api/search.index/root.js
+++ b/api/search.index/root.js
@@ -1 +1 @@
-rr_('{"normalizedName":{"I":"RQBAAAAEihNlgQEEBzoMtw7dDhUAQAAAIH0DZwTKBDTIBLkOFQBAAAAAuQ42ZRZMITUAQAAAAjwMcpMBfQNnBMoEVQJAAAAxBAc6DLcO3Q4AyAQBuA4yNjkeAB8AggGIEMUe7CAFAEAAABF/EKwbbpQBBQHAAAADtR4AqxthcwEBAaBgAAAbFmQ7MAAAAQAA5gADANQEAADqBAAAIQnkAAUBwwAAl7YgkvsfZHM1AEIAAAXpGWMKCgsKDAo9CgUAQAAAAjwMcpMBBABCdGQAAAZlZmF1bHR9G2TtBHYbAMHIefibO73CoFAAAB8cAAAAAQhOHAAA0gwaAAC5BQBRAAAGViFkLCEFAFYAAAZkIWRVISOBAoAQY9NgAWZwVQBDAAAGXiFkYSDpIBshHCFBIUUhNQBDAAAGWCFkphlUHcwgNCEjAAKgUAAAH9BhbvTwGgAAkgErBDcA+wJ [...]
\ No newline at end of file
+rr_('{"normalizedName":{"I":"RQBAAAAEihNlgQEEBzoMtw7dDhUAQAAAIH0DZwTKBDTIBLkOFQBAAAAAuQ42ZRZMITUAQAAAAjwMcpMBfQNnBMoEVQJAAAAxBAc6DLcO3Q4AyAQBuA4yNjkeAB8AggGIEMUe7CAFAEAAABF/EKwbbpQBBQHAAAADtR4AqxthcwEBAaBgAAAbFmQ7MAAAAQAA5gADANQEAADqBAAAIQnkAAUBwwAAl7YgkvsfZHM1AEIAAAXpGWMKCgsKDAo9CgUAQAAAAjwMcpMBBABCdGQAAAZlZmF1bHR9G2TtBHYbAMHIefibO73CoFAAAB8cAAAAAQhOHAAA0gwaAAC5BQBRAAAGViFkLCEFAFYAAAZkIWRVISOBAoAQY9NgAWZwVQBDAAAGXiFkYSDpIBshHCFBIUUhNQBDAAAGWCFkphlUHcwgNCEjAAKgUAAAH9BhbvTwGgAAkgErBDcA+wJ [...]
\ No newline at end of file
diff --git a/api/search.index/type/2d00f643fc8b.js b/api/search.index/type/5a27e3ee9475.js
similarity index 98%
rename from api/search.index/type/2d00f643fc8b.js
rename to api/search.index/type/5a27e3ee9475.js
index 482178f96..8e8572dbd 100644
--- a/api/search.index/type/2d00f643fc8b.js
+++ b/api/search.index/type/5a27e3ee9475.js
@@ -1 +1 @@
-rd_("Ec[\"AAAAAAAAAAABchsAAAIDIAAAfSAAAAAAAAAAAAHWCAAA\",\"AAAAAAAAAAABchsAAAIDIAAAfSAAAA==\"]Bg[\"AAAAAAAAAAAAAAAAAU4KAAABTwoAAA==\",\"\"]Og[\"AAAAAAAAAAAAOjAAAAEAAAAAABEAEAAAAF0QtBSzG7QbOBw5HKocqxyzHbQd5R7mHuse7B48Hz0fQh9DHwPTCAAA6wsAAIYOAAA7MAAAAQAAJgAOAHIEAAClDgAAXhACALUUAQASGgEA+RoBALUbAwA6HAMArBwDALUdAwDnHgMA7R4BAD4fAwBEHwEABNQIAADVCAAA7AsAAIcOAAAFcwQAAHQEAACSDgAApg4AAKcOAAA=\",\"\"]Gg[\"AAAAAAAAAAAAAAA6MAAAAQAAAAAAEgAQAAAAXxBgELYUEhoTGvka+hq3G7gbPBw9HK4crxy3Hbgd6R7qHkAfQR8B1QgAAAN [...]
\ No newline at end of file
+rd_("Ec[\"AAAAAAAAAAABchsAAAIDIAAAfSAAAAAAAAAAAAHWCAAA\",\"AAAAAAAAAAABchsAAAIDIAAAfSAAAA==\"]Bg[\"AAAAAAAAAAAAAAAAAU4KAAABTwoAAA==\",\"\"]Og[\"AAAAAAAAAAAAOjAAAAEAAAAAABEAEAAAAF0QtBSzG7QbOBw5HKocqxyzHbQd5R7mHuse7B48Hz0fQh9DHwPTCAAA6wsAAIYOAAA7MAAAAQAAJgAOAHIEAAClDgAAXhACALUUAQASGgEA+RoBALUbAwA6HAMArBwDALUdAwDnHgMA7R4BAD4fAwBEHwEABNQIAADVCAAA7AsAAIcOAAAFcwQAAHQEAACSDgAApg4AAKcOAAA=\",\"\"]Gg[\"AAAAAAAAAAAAAAA6MAAAAQAAAAAAEgAQAAAAXxBgELYUEhoTGvka+hq3G7gbPBw9HK4crxy3Hbgd6R7qHkAfQR8B1QgAAAN [...]
\ No newline at end of file
diff --git a/api/settings.html b/api/settings.html
index 94cbba246..df990645d 100644
--- a/api/settings.html
+++ b/api/settings.html
@@ -1 +1 @@
-<!DOCTYPE html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><meta name="generator" content="rustdoc"><meta name="description" content="Settings of Rustdoc"><title>Settings</title><script>if(window.location.protocol!=="file:")document.head.insertAdjacentHTML("beforeend","SourceSerif4-Regular-6b053e98.ttf.woff2,FiraSans-Italic-81dc35de.woff2,FiraSans-Regular-0fe48ade.woff2,FiraSans-MediumItalic-ccf7e434.woff2,FiraSans-Med [...]
\ No newline at end of file
+<!DOCTYPE html><html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width, initial-scale=1.0"><meta name="generator" content="rustdoc"><meta name="description" content="Settings of Rustdoc"><title>Settings</title><script>if(window.location.protocol!=="file:")document.head.insertAdjacentHTML("beforeend","SourceSerif4-Regular-6b053e98.ttf.woff2,FiraSans-Italic-81dc35de.woff2,FiraSans-Regular-0fe48ade.woff2,FiraSans-MediumItalic-ccf7e434.woff2,FiraSans-Med [...]
\ No newline at end of file
diff --git a/api/src/iceberg/arrow/reader/pipeline.rs.html b/api/src/iceberg/arrow/reader/pipeline.rs.html
index 243a65e42..30cf13460 100644
--- a/api/src/iceberg/arrow/reader/pipeline.rs.html
+++ b/api/src/iceberg/arrow/reader/pipeline.rs.html
@@ -310,1373 +310,1373 @@
 <a href=#310 id=310 data-nosnippet>310</a>            use_position_fallback, <span class="comment">// Whether to use position-based (true) or field-ID-based (false) projection
 <a href=#311 id=311 data-nosnippet>311</a>        </span>)<span class="question-mark">?</span>;
 <a href=#312 id=312 data-nosnippet>312</a>
-<a href=#313 id=313 data-nosnippet>313</a>        <span class="comment">// Union in the physical `_last_updated_sequence_number` column when we will
-<a href=#314 id=314 data-nosnippet>314</a>        // coalesce it. The metadata field id is not in the task schema, so it can't be
-<a href=#315 id=315 data-nosnippet>315</a>        // requested through `get_arrow_projection_mask` (which resolves ids against the
-<a href=#316 id=316 data-nosnippet>316</a>        // task schema); add its Parquet leaf directly.
-<a href=#317 id=317 data-nosnippet>317</a>        </span><span class="kw">if let </span><span class="prelude-val">Some</span>(leaf) = coalesce_last_updated_seq_leaf {
-<a href=#318 id=318 data-nosnippet>318</a>            <span class="kw">let </span>phys_mask =
-<a href=#319 id=319 data-nosnippet>319</a>                ProjectionMask::leaves(record_batch_stream_builder.parquet_schema(), <span class="macro">vec!</span>[leaf]);
-<a href=#320 id=320 data-nosnippet>320</a>            projection_mask.union(<span class="kw-2">&amp;</span>phys_mask);
-<a href=#321 id=321 data-nosnippet>321</a>        }
-<a href=#322 id=322 data-nosnippet>322</a>
-<a href=#323 id=323 data-nosnippet>323</a>        record_batch_stream_builder =
-<a href=#324 id=324 data-nosnippet>324</a>            record_batch_stream_builder.with_projection(projection_mask.clone());
-<a href=#325 id=325 data-nosnippet>325</a>
-<a href=#326 id=326 data-nosnippet>326</a>        <span class="comment">// RecordBatchTransformer performs any transformations required on the RecordBatches
-<a href=#327 id=327 data-nosnippet>327</a>        // that come back from the file, such as type promotion, default column insertion,
-<a href=#328 id=328 data-nosnippet>328</a>        // column re-ordering, partition constants, and virtual field addition (like _file)
-<a href=#329 id=329 data-nosnippet>329</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>record_batch_transformer_builder =
-<a href=#330 id=330 data-nosnippet>330</a>            RecordBatchTransformerBuilder::new(task.schema_ref(), task.project_field_ids());
-<a href=#331 id=331 data-nosnippet>331</a>
-<a href=#332 id=332 data-nosnippet>332</a>        <span class="comment">// Add the _file metadata column if it's in the projected fields
-<a href=#333 id=333 data-nosnippet>333</a>        </span><span class="kw">if </span>task.project_field_ids().contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_FILE) {
-<a href=#334 id=334 data-nosnippet>334</a>            <span class="kw">let </span>file_datum = Datum::string(task.data_file_path.clone());
-<a href=#335 id=335 data-nosnippet>335</a>            record_batch_transformer_builder =
-<a href=#336 id=336 data-nosnippet>336</a>                record_batch_transformer_builder.with_constant(RESERVED_FIELD_ID_FILE, file_datum);
-<a href=#337 id=337 data-nosnippet>337</a>        }
-<a href=#338 id=338 data-nosnippet>338</a>
-<a href=#339 id=339 data-nosnippet>339</a>        <span class="kw">if </span>task
-<a href=#340 id=340 data-nosnippet>340</a>            .project_field_ids()
-<a href=#341 id=341 data-nosnippet>341</a>            .contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_SPEC_ID)
-<a href=#342 id=342 data-nosnippet>342</a>        {
-<a href=#343 id=343 data-nosnippet>343</a>            <span class="kw">let </span>partition_spec = task
-<a href=#344 id=344 data-nosnippet>344</a>                .partition_spec
-<a href=#345 id=345 data-nosnippet>345</a>                .as_ref()
-<a href=#346 id=346 data-nosnippet>346</a>                .ok_or_else(|| Error::new(ErrorKind::Unexpected, <span class="string">"Partition spec is missing"</span>))<span class="question-mark">?</span>;
-<a href=#347 id=347 data-nosnippet>347</a>
-<a href=#348 id=348 data-nosnippet>348</a>            <span class="kw">let </span>spec_id_datum = Datum::int(partition_spec.spec_id());
-<a href=#349 id=349 data-nosnippet>349</a>            record_batch_transformer_builder = record_batch_transformer_builder
-<a href=#350 id=350 data-nosnippet>350</a>                .with_constant(RESERVED_FIELD_ID_SPEC_ID, spec_id_datum);
-<a href=#351 id=351 data-nosnippet>351</a>        }
-<a href=#352 id=352 data-nosnippet>352</a>
-<a href=#353 id=353 data-nosnippet>353</a>        <span class="kw">if </span>project_last_updated_seq {
-<a href=#354 id=354 data-nosnippet>354</a>            <span class="comment">// Materialize the column, gated on the data file's `first_row_id`. Java gates
-<a href=#355 id=355 data-nosnippet>355</a>            // it this way (`ValueReaders.lastUpdated` returns nulls when the base row id is
-<a href=#356 id=356 data-nosnippet>356</a>            // null); the spec itself only says the column is assigned the manifest entry's
-<a href=#357 id=357 data-nosnippet>357</a>            // sequence number on read.
-<a href=#358 id=358 data-nosnippet>358</a>            </span>record_batch_transformer_builder = <span class="kw">match </span>(task.first_row_id, task.data_sequence_number)
-<a href=#359 id=359 data-nosnippet>359</a>            {
-<a href=#360 id=360 data-nosnippet>360</a>                (<span class="prelude-val">Some</span>(<span class="kw">_</span>), <span class="prelude-val">Some</span>(seq)) =&gt; {
-<a href=#361 id=361 data-nosnippet>361</a>                    <span class="kw">let </span>datum = Datum::long(seq);
-<a href=#362 id=362 data-nosnippet>362</a>                    <span class="kw">if </span>coalesce_last_updated_seq_leaf.is_some() {
-<a href=#363 id=363 data-nosnippet>363</a>                        <span class="comment">// The file physically carries the column: read the per-row value,
-<a href=#364 id=364 data-nosnippet>364</a>                        // falling back to the data sequence number only where null.
-<a href=#365 id=365 data-nosnippet>365</a>                        </span>record_batch_transformer_builder.with_coalesced_last_updated_seq_column(
-<a href=#366 id=366 data-nosnippet>366</a>                            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#367 id=367 data-nosnippet>367</a>                            datum,
-<a href=#368 id=368 data-nosnippet>368</a>                        )
-<a href=#369 id=369 data-nosnippet>369</a>                    } <span class="kw">else if </span>last_updated_seq_present_by_name_only {
-<a href=#370 id=370 data-nosnippet>370</a>                        <span class="comment">// Present by name but without the embedded field id (name mapping /
-<a href=#371 id=371 data-nosnippet>371</a>                        // positional fallback). The transformer keys the source column by
-<a href=#372 id=372 data-nosnippet>372</a>                        // field id, so we can't thread it; no real writer produces this, so
-<a href=#373 id=373 data-nosnippet>373</a>                        // reject loudly rather than silently overwrite with the constant.
-<a href=#374 id=374 data-nosnippet>374</a>                        // Arm-local by design: only this arm reads the physical column, so
-<a href=#375 id=375 data-nosnippet>375</a>                        // only here can a name-only column defeat us. The `(None, _)` arm
-<a href=#376 id=376 data-nosnippet>376</a>                        // nulls the column without reading it, so it needs no such guard.
-<a href=#377 id=377 data-nosnippet>377</a>                        </span><span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
-<a href=#378 id=378 data-nosnippet>378</a>                            ErrorKind::FeatureUnsupported,
-<a href=#379 id=379 data-nosnippet>379</a>                            <span class="string">"Reading a physically-stored _last_updated_sequence_number column \
-<a href=#380 id=380 data-nosnippet>380</a>                             without an embedded field id is not supported"</span>,
-<a href=#381 id=381 data-nosnippet>381</a>                        ));
-<a href=#382 id=382 data-nosnippet>382</a>                    } <span class="kw">else </span>{
-<a href=#383 id=383 data-nosnippet>383</a>                        <span class="comment">// Column absent: derive it from the data sequence number.
-<a href=#384 id=384 data-nosnippet>384</a>                        </span>record_batch_transformer_builder
-<a href=#385 id=385 data-nosnippet>385</a>                            .with_constant(RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER, datum)
-<a href=#386 id=386 data-nosnippet>386</a>                    }
-<a href=#387 id=387 data-nosnippet>387</a>                }
-<a href=#388 id=388 data-nosnippet>388</a>                <span class="comment">// Null first_row_id (v1/v2, or a pre-upgrade v3 snapshot): the column is null.
-<a href=#389 id=389 data-nosnippet>389</a>                </span>(<span class="prelude-val">None</span>, <span class="kw">_</span>) =&gt; record_batch_transformer_builder
-<a href=#390 id=390 data-nosnippet>390</a>                    .with_null_metadata_column(RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER)<span class="question-mark">?</span>,
-<a href=#391 id=391 data-nosnippet>391</a>                <span class="comment">// first_row_id present but no data sequence number: after manifest
-<a href=#392 id=392 data-nosnippet>392</a>                // inheritance a committed entry always has one, so this is a malformed
-<a href=#393 id=393 data-nosnippet>393</a>                // manifest rather than a legitimate null.
-<a href=#394 id=394 data-nosnippet>394</a>                </span>(<span class="prelude-val">Some</span>(<span class="kw">_</span>), <span class="prelude-val">None</span>) =&gt; {
-<a href=#395 id=395 data-nosnippet>395</a>                    <span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
-<a href=#396 id=396 data-nosnippet>396</a>                        ErrorKind::DataInvalid,
-<a href=#397 id=397 data-nosnippet>397</a>                        <span class="macro">format!</span>(
-<a href=#398 id=398 data-nosnippet>398</a>                            <span class="string">"Data file {} has a first_row_id but no data sequence number"</span>,
-<a href=#399 id=399 data-nosnippet>399</a>                            task.data_file_path
-<a href=#400 id=400 data-nosnippet>400</a>                        ),
-<a href=#401 id=401 data-nosnippet>401</a>                    ));
-<a href=#402 id=402 data-nosnippet>402</a>                }
-<a href=#403 id=403 data-nosnippet>403</a>            };
-<a href=#404 id=404 data-nosnippet>404</a>        }
-<a href=#405 id=405 data-nosnippet>405</a>
-<a href=#406 id=406 data-nosnippet>406</a>        <span class="kw">if let </span>(<span class="prelude-val">Some</span>(partition_spec), <span class="prelude-val">Some</span>(partition_data)) =
-<a href=#407 id=407 data-nosnippet>407</a>            (task.partition_spec.clone(), task.partition.clone())
-<a href=#408 id=408 data-nosnippet>408</a>        {
-<a href=#409 id=409 data-nosnippet>409</a>            record_batch_transformer_builder =
-<a href=#410 id=410 data-nosnippet>410</a>                record_batch_transformer_builder.with_partition(partition_spec, partition_data)<span class="question-mark">?</span>;
-<a href=#411 id=411 data-nosnippet>411</a>        }
-<a href=#412 id=412 data-nosnippet>412</a>
-<a href=#413 id=413 data-nosnippet>413</a>        <span class="kw">if </span>project_pos {
-<a href=#414 id=414 data-nosnippet>414</a>            record_batch_transformer_builder =
-<a href=#415 id=415 data-nosnippet>415</a>                record_batch_transformer_builder.with_virtual_field(RESERVED_FIELD_ID_POS);
-<a href=#416 id=416 data-nosnippet>416</a>        }
-<a href=#417 id=417 data-nosnippet>417</a>
-<a href=#418 id=418 data-nosnippet>418</a>        <span class="comment">// Add the _partition metadata struct column if it's in the projected fields.
-<a href=#419 id=419 data-nosnippet>419</a>        // Computed lazily here at read time from the unified partition type + task's spec + data.
-<a href=#420 id=420 data-nosnippet>420</a>        </span><span class="kw">if </span>task
-<a href=#421 id=421 data-nosnippet>421</a>            .project_field_ids()
-<a href=#422 id=422 data-nosnippet>422</a>            .contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_PARTITION)
-<a href=#423 id=423 data-nosnippet>423</a>            &amp;&amp; <span class="kw">let </span><span class="prelude-val">Some</span>(unified_type) = <span class="kw-2">&amp;</span>task.unified_partition_type
-<a href=#424 id=424 data-nosnippet>424</a>        {
-<a href=#425 id=425 data-nosnippet>425</a>            <span class="kw">let </span>(spec, partition_data) = <span class="kw">match </span>(<span class="kw-2">&amp;</span>task.partition_spec, <span class="kw-2">&amp;</span>task.partition) {
-<a href=#426 id=426 data-nosnippet>426</a>                (<span class="prelude-val">Some</span>(spec), <span class="prelude-val">Some</span>(data)) =&gt; (spec.clone(), data.clone()),
-<a href=#427 id=427 data-nosnippet>427</a>                <span class="comment">// A missing spec/data is only acceptable when there are no partition
-<a href=#428 id=428 data-nosnippet>428</a>                // fields to fill (unpartitioned table). If the unified type has fields
-<a href=#429 id=429 data-nosnippet>429</a>                // but we lack a spec or data, the task is inconsistent and we cannot
-<a href=#430 id=430 data-nosnippet>430</a>                // build the _partition column.
-<a href=#431 id=431 data-nosnippet>431</a>                </span><span class="kw">_ if </span>unified_type.fields().is_empty() =&gt; {
-<a href=#432 id=432 data-nosnippet>432</a>                    (Arc::new(PartitionSpec::unpartition_spec()), Struct::empty())
-<a href=#433 id=433 data-nosnippet>433</a>                }
-<a href=#434 id=434 data-nosnippet>434</a>                <span class="kw">_ </span>=&gt; {
-<a href=#435 id=435 data-nosnippet>435</a>                    <span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
-<a href=#436 id=436 data-nosnippet>436</a>                        ErrorKind::Unexpected,
-<a href=#437 id=437 data-nosnippet>437</a>                        <span class="string">"cannot build _partition column: unified partition type has fields \
-<a href=#438 id=438 data-nosnippet>438</a>                         but the scan task is missing its partition spec or data"</span>,
-<a href=#439 id=439 data-nosnippet>439</a>                    ));
-<a href=#440 id=440 data-nosnippet>440</a>                }
-<a href=#441 id=441 data-nosnippet>441</a>            };
-<a href=#442 id=442 data-nosnippet>442</a>            <span class="kw">let </span>constant = build_partition_constant(unified_type, <span class="kw-2">&amp;</span>spec, <span class="kw-2">&amp;</span>partition_data)<span class="question-mark">?</span>;
-<a href=#443 id=443 data-nosnippet>443</a>            record_batch_transformer_builder =
-<a href=#444 id=444 data-nosnippet>444</a>                record_batch_transformer_builder.with_partition_constant(constant);
-<a href=#445 id=445 data-nosnippet>445</a>        }
-<a href=#446 id=446 data-nosnippet>446</a>
-<a href=#447 id=447 data-nosnippet>447</a>        <span class="kw">let </span><span class="kw-2">mut </span>record_batch_transformer = record_batch_transformer_builder.build();
-<a href=#448 id=448 data-nosnippet>448</a>
-<a href=#449 id=449 data-nosnippet>449</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(batch_size) = <span class="self">self</span>.batch_size {
-<a href=#450 id=450 data-nosnippet>450</a>            record_batch_stream_builder = record_batch_stream_builder.with_batch_size(batch_size);
-<a href=#451 id=451 data-nosnippet>451</a>        }
-<a href=#452 id=452 data-nosnippet>452</a>
-<a href=#453 id=453 data-nosnippet>453</a>        <span class="kw">let </span>delete_filter = delete_filter_rx.<span class="kw">await</span>.unwrap()<span class="question-mark">?</span>;
-<a href=#454 id=454 data-nosnippet>454</a>        <span class="kw">let </span>delete_predicate = delete_filter.build_equality_delete_predicate(<span class="kw-2">&amp;</span>task).<span class="kw">await</span><span class="question-mark">?</span>;
-<a href=#455 id=455 data-nosnippet>455</a>
-<a href=#456 id=456 data-nosnippet>456</a>        <span class="comment">// In addition to the optional predicate supplied in the `FileScanTask`,
-<a href=#457 id=457 data-nosnippet>457</a>        // we also have an optional predicate resulting from equality delete files.
-<a href=#458 id=458 data-nosnippet>458</a>        // If both are present, we logical-AND them together to form a single filter
-<a href=#459 id=459 data-nosnippet>459</a>        // predicate that we can pass to the `RecordBatchStreamBuilder`.
-<a href=#460 id=460 data-nosnippet>460</a>        </span><span class="kw">let </span>final_predicate = <span class="kw">match </span>(<span class="kw-2">&amp;</span>task.predicate, delete_predicate) {
-<a href=#461 id=461 data-nosnippet>461</a>            (<span class="prelude-val">None</span>, <span class="prelude-val">None</span>) =&gt; <span class="prelude-val">None</span>,
-<a href=#462 id=462 data-nosnippet>462</a>            (<span class="prelude-val">Some</span>(predicate), <span class="prelude-val">None</span>) =&gt; <span class="prelude-val">Some</span>(predicate.clone()),
-<a href=#463 id=463 data-nosnippet>463</a>            (<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="kw-2">ref </span>predicate)) =&gt; <span class="prelude-val">Some</span>(predicate.clone()),
-<a href=#464 id=464 data-nosnippet>464</a>            (<span class="prelude-val">Some</span>(filter_predicate), <span class="prelude-val">Some</span>(delete_predicate)) =&gt; {
-<a href=#465 id=465 data-nosnippet>465</a>                <span class="prelude-val">Some</span>(filter_predicate.clone().and(delete_predicate))
-<a href=#466 id=466 data-nosnippet>466</a>            }
-<a href=#467 id=467 data-nosnippet>467</a>        };
-<a href=#468 id=468 data-nosnippet>468</a>
-<a href=#469 id=469 data-nosnippet>469</a>        <span class="comment">// There are three possible sources for potential lists of selected RowGroup indices,
-<a href=#470 id=470 data-nosnippet>470</a>        // and two for `RowSelection`s.
-<a href=#471 id=471 data-nosnippet>471</a>        // Selected RowGroup index lists can come from three sources:
-<a href=#472 id=472 data-nosnippet>472</a>        //   * When task.start and task.length specify a byte range (file splitting);
-<a href=#473 id=473 data-nosnippet>473</a>        //   * When there are equality delete files that are applicable;
-<a href=#474 id=474 data-nosnippet>474</a>        //   * When there is a scan predicate and row_group_filtering_enabled = true.
-<a href=#475 id=475 data-nosnippet>475</a>        // `RowSelection`s can be created in either or both of the following cases:
-<a href=#476 id=476 data-nosnippet>476</a>        //   * When there are positional delete files that are applicable;
-<a href=#477 id=477 data-nosnippet>477</a>        //   * When there is a scan predicate and row_selection_enabled = true
-<a href=#478 id=478 data-nosnippet>478</a>        // Note that row group filtering from predicates only happens when
-<a href=#479 id=479 data-nosnippet>479</a>        // there is a scan predicate AND row_group_filtering_enabled = true,
-<a href=#480 id=480 data-nosnippet>480</a>        // but we perform row selection filtering if there are applicable
-<a href=#481 id=481 data-nosnippet>481</a>        // equality delete files OR (there is a scan predicate AND row_selection_enabled),
-<a href=#482 id=482 data-nosnippet>482</a>        // since the only implemented method of applying positional deletes is
-<a href=#483 id=483 data-nosnippet>483</a>        // by using a `RowSelection`.
-<a href=#484 id=484 data-nosnippet>484</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>selected_row_group_indices = <span class="prelude-val">None</span>;
-<a href=#485 id=485 data-nosnippet>485</a>        <span class="kw">let </span><span class="kw-2">mut </span>row_selection = <span class="prelude-val">None</span>;
+<a href=#313 id=313 data-nosnippet>313</a>        <span class="comment">// A metadata-only projection leaves `project_field_ids_without_metadata` empty,
+<a href=#314 id=314 data-nosnippet>314</a>        // which `get_arrow_projection_mask` maps to "read all columns" (so `COUNT(*)` still
+<a href=#315 id=315 data-nosnippet>315</a>        // gets a row count). Downgrade that to "read no data columns" when a row-count
+<a href=#316 id=316 data-nosnippet>316</a>        // source exists independently of the data columns: the RowNumber virtual column
+<a href=#317 id=317 data-nosnippet>317</a>        // (installed above under `project_pos`) or a physical metadata leaf unioned in
+<a href=#318 id=318 data-nosnippet>318</a>        // below. Pure-constant / `COUNT(*)` projections have neither and must keep reading
+<a href=#319 id=319 data-nosnippet>319</a>        // all columns to preserve the row count. Any future physical metadata leaf (e.g. a
+<a href=#320 id=320 data-nosnippet>320</a>        // `_row_id` read path) is likewise a row source.
+<a href=#321 id=321 data-nosnippet>321</a>        //
+<a href=#322 id=322 data-nosnippet>322</a>        // This runs BEFORE the union so the physical leaf is added onto a `none` base,
+<a href=#323 id=323 data-nosnippet>323</a>        // pruning the read to just that leaf (`union` with an `all` base stays `all`).
+<a href=#324 id=324 data-nosnippet>324</a>        </span><span class="kw">if </span>project_field_ids_without_metadata.is_empty()
+<a href=#325 id=325 data-nosnippet>325</a>            &amp;&amp; (project_pos || coalesce_last_updated_seq_leaf.is_some())
+<a href=#326 id=326 data-nosnippet>326</a>        {
+<a href=#327 id=327 data-nosnippet>327</a>            projection_mask =
+<a href=#328 id=328 data-nosnippet>328</a>                ProjectionMask::none(record_batch_stream_builder.parquet_schema().num_columns());
+<a href=#329 id=329 data-nosnippet>329</a>        }
+<a href=#330 id=330 data-nosnippet>330</a>
+<a href=#331 id=331 data-nosnippet>331</a>        <span class="comment">// Union in the physical `_last_updated_sequence_number` column when we will
+<a href=#332 id=332 data-nosnippet>332</a>        // coalesce it. The metadata field id is not in the task schema, so it can't be
+<a href=#333 id=333 data-nosnippet>333</a>        // requested through `get_arrow_projection_mask` (which resolves ids against the
+<a href=#334 id=334 data-nosnippet>334</a>        // task schema); add its Parquet leaf directly.
+<a href=#335 id=335 data-nosnippet>335</a>        </span><span class="kw">if let </span><span class="prelude-val">Some</span>(leaf) = coalesce_last_updated_seq_leaf {
+<a href=#336 id=336 data-nosnippet>336</a>            <span class="kw">let </span>phys_mask =
+<a href=#337 id=337 data-nosnippet>337</a>                ProjectionMask::leaves(record_batch_stream_builder.parquet_schema(), <span class="macro">vec!</span>[leaf]);
+<a href=#338 id=338 data-nosnippet>338</a>            projection_mask.union(<span class="kw-2">&amp;</span>phys_mask);
+<a href=#339 id=339 data-nosnippet>339</a>        }
+<a href=#340 id=340 data-nosnippet>340</a>
+<a href=#341 id=341 data-nosnippet>341</a>        record_batch_stream_builder =
+<a href=#342 id=342 data-nosnippet>342</a>            record_batch_stream_builder.with_projection(projection_mask.clone());
+<a href=#343 id=343 data-nosnippet>343</a>
+<a href=#344 id=344 data-nosnippet>344</a>        <span class="comment">// RecordBatchTransformer performs any transformations required on the RecordBatches
+<a href=#345 id=345 data-nosnippet>345</a>        // that come back from the file, such as type promotion, default column insertion,
+<a href=#346 id=346 data-nosnippet>346</a>        // column re-ordering, partition constants, and virtual field addition (like _file)
+<a href=#347 id=347 data-nosnippet>347</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>record_batch_transformer_builder =
+<a href=#348 id=348 data-nosnippet>348</a>            RecordBatchTransformerBuilder::new(task.schema_ref(), task.project_field_ids());
+<a href=#349 id=349 data-nosnippet>349</a>
+<a href=#350 id=350 data-nosnippet>350</a>        <span class="comment">// Add the _file metadata column if it's in the projected fields
+<a href=#351 id=351 data-nosnippet>351</a>        </span><span class="kw">if </span>task.project_field_ids().contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_FILE) {
+<a href=#352 id=352 data-nosnippet>352</a>            <span class="kw">let </span>file_datum = Datum::string(task.data_file_path.clone());
+<a href=#353 id=353 data-nosnippet>353</a>            record_batch_transformer_builder =
+<a href=#354 id=354 data-nosnippet>354</a>                record_batch_transformer_builder.with_constant(RESERVED_FIELD_ID_FILE, file_datum);
+<a href=#355 id=355 data-nosnippet>355</a>        }
+<a href=#356 id=356 data-nosnippet>356</a>
+<a href=#357 id=357 data-nosnippet>357</a>        <span class="kw">if </span>task
+<a href=#358 id=358 data-nosnippet>358</a>            .project_field_ids()
+<a href=#359 id=359 data-nosnippet>359</a>            .contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_SPEC_ID)
+<a href=#360 id=360 data-nosnippet>360</a>        {
+<a href=#361 id=361 data-nosnippet>361</a>            <span class="kw">let </span>partition_spec = task
+<a href=#362 id=362 data-nosnippet>362</a>                .partition_spec
+<a href=#363 id=363 data-nosnippet>363</a>                .as_ref()
+<a href=#364 id=364 data-nosnippet>364</a>                .ok_or_else(|| Error::new(ErrorKind::Unexpected, <span class="string">"Partition spec is missing"</span>))<span class="question-mark">?</span>;
+<a href=#365 id=365 data-nosnippet>365</a>
+<a href=#366 id=366 data-nosnippet>366</a>            <span class="kw">let </span>spec_id_datum = Datum::int(partition_spec.spec_id());
+<a href=#367 id=367 data-nosnippet>367</a>            record_batch_transformer_builder = record_batch_transformer_builder
+<a href=#368 id=368 data-nosnippet>368</a>                .with_constant(RESERVED_FIELD_ID_SPEC_ID, spec_id_datum);
+<a href=#369 id=369 data-nosnippet>369</a>        }
+<a href=#370 id=370 data-nosnippet>370</a>
+<a href=#371 id=371 data-nosnippet>371</a>        <span class="kw">if </span>project_last_updated_seq {
+<a href=#372 id=372 data-nosnippet>372</a>            <span class="comment">// Materialize the column, gated on the data file's `first_row_id`. Java gates
+<a href=#373 id=373 data-nosnippet>373</a>            // it this way (`ValueReaders.lastUpdated` returns nulls when the base row id is
+<a href=#374 id=374 data-nosnippet>374</a>            // null); the spec itself only says the column is assigned the manifest entry's
+<a href=#375 id=375 data-nosnippet>375</a>            // sequence number on read.
+<a href=#376 id=376 data-nosnippet>376</a>            </span>record_batch_transformer_builder = <span class="kw">match </span>(task.first_row_id, task.data_sequence_number)
+<a href=#377 id=377 data-nosnippet>377</a>            {
+<a href=#378 id=378 data-nosnippet>378</a>                (<span class="prelude-val">Some</span>(<span class="kw">_</span>), <span class="prelude-val">Some</span>(seq)) =&gt; {
+<a href=#379 id=379 data-nosnippet>379</a>                    <span class="kw">let </span>datum = Datum::long(seq);
+<a href=#380 id=380 data-nosnippet>380</a>                    <span class="kw">if </span>coalesce_last_updated_seq_leaf.is_some() {
+<a href=#381 id=381 data-nosnippet>381</a>                        <span class="comment">// The file physically carries the column: read the per-row value,
+<a href=#382 id=382 data-nosnippet>382</a>                        // falling back to the data sequence number only where null.
+<a href=#383 id=383 data-nosnippet>383</a>                        </span>record_batch_transformer_builder.with_coalesced_last_updated_seq_column(
+<a href=#384 id=384 data-nosnippet>384</a>                            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#385 id=385 data-nosnippet>385</a>                            datum,
+<a href=#386 id=386 data-nosnippet>386</a>                        )
+<a href=#387 id=387 data-nosnippet>387</a>                    } <span class="kw">else if </span>last_updated_seq_present_by_name_only {
+<a href=#388 id=388 data-nosnippet>388</a>                        <span class="comment">// Present by name but without the embedded field id (name mapping /
+<a href=#389 id=389 data-nosnippet>389</a>                        // positional fallback). The transformer keys the source column by
+<a href=#390 id=390 data-nosnippet>390</a>                        // field id, so we can't thread it; no real writer produces this, so
+<a href=#391 id=391 data-nosnippet>391</a>                        // reject loudly rather than silently overwrite with the constant.
+<a href=#392 id=392 data-nosnippet>392</a>                        // Arm-local by design: only this arm reads the physical column, so
+<a href=#393 id=393 data-nosnippet>393</a>                        // only here can a name-only column defeat us. The `(None, _)` arm
+<a href=#394 id=394 data-nosnippet>394</a>                        // nulls the column without reading it, so it needs no such guard.
+<a href=#395 id=395 data-nosnippet>395</a>                        </span><span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
+<a href=#396 id=396 data-nosnippet>396</a>                            ErrorKind::FeatureUnsupported,
+<a href=#397 id=397 data-nosnippet>397</a>                            <span class="string">"Reading a physically-stored _last_updated_sequence_number column \
+<a href=#398 id=398 data-nosnippet>398</a>                             without an embedded field id is not supported"</span>,
+<a href=#399 id=399 data-nosnippet>399</a>                        ));
+<a href=#400 id=400 data-nosnippet>400</a>                    } <span class="kw">else </span>{
+<a href=#401 id=401 data-nosnippet>401</a>                        <span class="comment">// Column absent: derive it from the data sequence number.
+<a href=#402 id=402 data-nosnippet>402</a>                        </span>record_batch_transformer_builder
+<a href=#403 id=403 data-nosnippet>403</a>                            .with_constant(RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER, datum)
+<a href=#404 id=404 data-nosnippet>404</a>                    }
+<a href=#405 id=405 data-nosnippet>405</a>                }
+<a href=#406 id=406 data-nosnippet>406</a>                <span class="comment">// Null first_row_id (v1/v2, or a pre-upgrade v3 snapshot): the column is null.
+<a href=#407 id=407 data-nosnippet>407</a>                </span>(<span class="prelude-val">None</span>, <span class="kw">_</span>) =&gt; record_batch_transformer_builder
+<a href=#408 id=408 data-nosnippet>408</a>                    .with_null_metadata_column(RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER)<span class="question-mark">?</span>,
+<a href=#409 id=409 data-nosnippet>409</a>                <span class="comment">// first_row_id present but no data sequence number: after manifest
+<a href=#410 id=410 data-nosnippet>410</a>                // inheritance a committed entry always has one, so this is a malformed
+<a href=#411 id=411 data-nosnippet>411</a>                // manifest rather than a legitimate null.
+<a href=#412 id=412 data-nosnippet>412</a>                </span>(<span class="prelude-val">Some</span>(<span class="kw">_</span>), <span class="prelude-val">None</span>) =&gt; {
+<a href=#413 id=413 data-nosnippet>413</a>                    <span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
+<a href=#414 id=414 data-nosnippet>414</a>                        ErrorKind::DataInvalid,
+<a href=#415 id=415 data-nosnippet>415</a>                        <span class="macro">format!</span>(
+<a href=#416 id=416 data-nosnippet>416</a>                            <span class="string">"Data file {} has a first_row_id but no data sequence number"</span>,
+<a href=#417 id=417 data-nosnippet>417</a>                            task.data_file_path
+<a href=#418 id=418 data-nosnippet>418</a>                        ),
+<a href=#419 id=419 data-nosnippet>419</a>                    ));
+<a href=#420 id=420 data-nosnippet>420</a>                }
+<a href=#421 id=421 data-nosnippet>421</a>            };
+<a href=#422 id=422 data-nosnippet>422</a>        }
+<a href=#423 id=423 data-nosnippet>423</a>
+<a href=#424 id=424 data-nosnippet>424</a>        <span class="kw">if let </span>(<span class="prelude-val">Some</span>(partition_spec), <span class="prelude-val">Some</span>(partition_data)) =
+<a href=#425 id=425 data-nosnippet>425</a>            (task.partition_spec.clone(), task.partition.clone())
+<a href=#426 id=426 data-nosnippet>426</a>        {
+<a href=#427 id=427 data-nosnippet>427</a>            record_batch_transformer_builder =
+<a href=#428 id=428 data-nosnippet>428</a>                record_batch_transformer_builder.with_partition(partition_spec, partition_data)<span class="question-mark">?</span>;
+<a href=#429 id=429 data-nosnippet>429</a>        }
+<a href=#430 id=430 data-nosnippet>430</a>
+<a href=#431 id=431 data-nosnippet>431</a>        <span class="kw">if </span>project_pos {
+<a href=#432 id=432 data-nosnippet>432</a>            record_batch_transformer_builder =
+<a href=#433 id=433 data-nosnippet>433</a>                record_batch_transformer_builder.with_virtual_field(RESERVED_FIELD_ID_POS);
+<a href=#434 id=434 data-nosnippet>434</a>        }
+<a href=#435 id=435 data-nosnippet>435</a>
+<a href=#436 id=436 data-nosnippet>436</a>        <span class="comment">// Add the _partition metadata struct column if it's in the projected fields.
+<a href=#437 id=437 data-nosnippet>437</a>        // Computed lazily here at read time from the unified partition type + task's spec + data.
+<a href=#438 id=438 data-nosnippet>438</a>        </span><span class="kw">if </span>task
+<a href=#439 id=439 data-nosnippet>439</a>            .project_field_ids()
+<a href=#440 id=440 data-nosnippet>440</a>            .contains(<span class="kw-2">&amp;</span>RESERVED_FIELD_ID_PARTITION)
+<a href=#441 id=441 data-nosnippet>441</a>            &amp;&amp; <span class="kw">let </span><span class="prelude-val">Some</span>(unified_type) = <span class="kw-2">&amp;</span>task.unified_partition_type
+<a href=#442 id=442 data-nosnippet>442</a>        {
+<a href=#443 id=443 data-nosnippet>443</a>            <span class="kw">let </span>(spec, partition_data) = <span class="kw">match </span>(<span class="kw-2">&amp;</span>task.partition_spec, <span class="kw-2">&amp;</span>task.partition) {
+<a href=#444 id=444 data-nosnippet>444</a>                (<span class="prelude-val">Some</span>(spec), <span class="prelude-val">Some</span>(data)) =&gt; (spec.clone(), data.clone()),
+<a href=#445 id=445 data-nosnippet>445</a>                <span class="comment">// A missing spec/data is only acceptable when there are no partition
+<a href=#446 id=446 data-nosnippet>446</a>                // fields to fill (unpartitioned table). If the unified type has fields
+<a href=#447 id=447 data-nosnippet>447</a>                // but we lack a spec or data, the task is inconsistent and we cannot
+<a href=#448 id=448 data-nosnippet>448</a>                // build the _partition column.
+<a href=#449 id=449 data-nosnippet>449</a>                </span><span class="kw">_ if </span>unified_type.fields().is_empty() =&gt; {
+<a href=#450 id=450 data-nosnippet>450</a>                    (Arc::new(PartitionSpec::unpartition_spec()), Struct::empty())
+<a href=#451 id=451 data-nosnippet>451</a>                }
+<a href=#452 id=452 data-nosnippet>452</a>                <span class="kw">_ </span>=&gt; {
+<a href=#453 id=453 data-nosnippet>453</a>                    <span class="kw">return </span><span class="prelude-val">Err</span>(Error::new(
+<a href=#454 id=454 data-nosnippet>454</a>                        ErrorKind::Unexpected,
+<a href=#455 id=455 data-nosnippet>455</a>                        <span class="string">"cannot build _partition column: unified partition type has fields \
+<a href=#456 id=456 data-nosnippet>456</a>                         but the scan task is missing its partition spec or data"</span>,
+<a href=#457 id=457 data-nosnippet>457</a>                    ));
+<a href=#458 id=458 data-nosnippet>458</a>                }
+<a href=#459 id=459 data-nosnippet>459</a>            };
+<a href=#460 id=460 data-nosnippet>460</a>            <span class="kw">let </span>constant = build_partition_constant(unified_type, <span class="kw-2">&amp;</span>spec, <span class="kw-2">&amp;</span>partition_data)<span class="question-mark">?</span>;
+<a href=#461 id=461 data-nosnippet>461</a>            record_batch_transformer_builder =
+<a href=#462 id=462 data-nosnippet>462</a>                record_batch_transformer_builder.with_partition_constant(constant);
+<a href=#463 id=463 data-nosnippet>463</a>        }
+<a href=#464 id=464 data-nosnippet>464</a>
+<a href=#465 id=465 data-nosnippet>465</a>        <span class="kw">let </span><span class="kw-2">mut </span>record_batch_transformer = record_batch_transformer_builder.build();
+<a href=#466 id=466 data-nosnippet>466</a>
+<a href=#467 id=467 data-nosnippet>467</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(batch_size) = <span class="self">self</span>.batch_size {
+<a href=#468 id=468 data-nosnippet>468</a>            record_batch_stream_builder = record_batch_stream_builder.with_batch_size(batch_size);
+<a href=#469 id=469 data-nosnippet>469</a>        }
+<a href=#470 id=470 data-nosnippet>470</a>
+<a href=#471 id=471 data-nosnippet>471</a>        <span class="kw">let </span>delete_filter = delete_filter_rx.<span class="kw">await</span>.unwrap()<span class="question-mark">?</span>;
+<a href=#472 id=472 data-nosnippet>472</a>        <span class="kw">let </span>delete_predicate = delete_filter.build_equality_delete_predicate(<span class="kw-2">&amp;</span>task).<span class="kw">await</span><span class="question-mark">?</span>;
+<a href=#473 id=473 data-nosnippet>473</a>
+<a href=#474 id=474 data-nosnippet>474</a>        <span class="comment">// In addition to the optional predicate supplied in the `FileScanTask`,
+<a href=#475 id=475 data-nosnippet>475</a>        // we also have an optional predicate resulting from equality delete files.
+<a href=#476 id=476 data-nosnippet>476</a>        // If both are present, we logical-AND them together to form a single filter
+<a href=#477 id=477 data-nosnippet>477</a>        // predicate that we can pass to the `RecordBatchStreamBuilder`.
+<a href=#478 id=478 data-nosnippet>478</a>        </span><span class="kw">let </span>final_predicate = <span class="kw">match </span>(<span class="kw-2">&amp;</span>task.predicate, delete_predicate) {
+<a href=#479 id=479 data-nosnippet>479</a>            (<span class="prelude-val">None</span>, <span class="prelude-val">None</span>) =&gt; <span class="prelude-val">None</span>,
+<a href=#480 id=480 data-nosnippet>480</a>            (<span class="prelude-val">Some</span>(predicate), <span class="prelude-val">None</span>) =&gt; <span class="prelude-val">Some</span>(predicate.clone()),
+<a href=#481 id=481 data-nosnippet>481</a>            (<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="kw-2">ref </span>predicate)) =&gt; <span class="prelude-val">Some</span>(predicate.clone()),
+<a href=#482 id=482 data-nosnippet>482</a>            (<span class="prelude-val">Some</span>(filter_predicate), <span class="prelude-val">Some</span>(delete_predicate)) =&gt; {
+<a href=#483 id=483 data-nosnippet>483</a>                <span class="prelude-val">Some</span>(filter_predicate.clone().and(delete_predicate))
+<a href=#484 id=484 data-nosnippet>484</a>            }
+<a href=#485 id=485 data-nosnippet>485</a>        };
 <a href=#486 id=486 data-nosnippet>486</a>
-<a href=#487 id=487 data-nosnippet>487</a>        <span class="comment">// Filter row groups based on byte range from task.start and task.length.
-<a href=#488 id=488 data-nosnippet>488</a>        // If both start and length are 0, read the entire file (backwards compatibility).
-<a href=#489 id=489 data-nosnippet>489</a>        </span><span class="kw">if </span>task.start != <span class="number">0 </span>|| task.length != <span class="number">0 </span>{
-<a href=#490 id=490 data-nosnippet>490</a>            <span class="kw">let </span>byte_range_filtered_row_groups = ArrowReader::filter_row_groups_by_byte_range(
-<a href=#491 id=491 data-nosnippet>491</a>                record_batch_stream_builder.metadata(),
-<a href=#492 id=492 data-nosnippet>492</a>                task.start,
-<a href=#493 id=493 data-nosnippet>493</a>                task.length,
-<a href=#494 id=494 data-nosnippet>494</a>            )<span class="question-mark">?</span>;
-<a href=#495 id=495 data-nosnippet>495</a>            selected_row_group_indices = <span class="prelude-val">Some</span>(byte_range_filtered_row_groups);
-<a href=#496 id=496 data-nosnippet>496</a>        }
-<a href=#497 id=497 data-nosnippet>497</a>
-<a href=#498 id=498 data-nosnippet>498</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(predicate) = final_predicate {
-<a href=#499 id=499 data-nosnippet>499</a>            <span class="kw">let </span>(iceberg_field_ids, field_id_map) = ArrowReader::build_field_id_set_and_map(
-<a href=#500 id=500 data-nosnippet>500</a>                record_batch_stream_builder.parquet_schema(),
-<a href=#501 id=501 data-nosnippet>501</a>                record_batch_stream_builder.schema(),
-<a href=#502 id=502 data-nosnippet>502</a>                <span class="kw-2">&amp;</span>predicate,
-<a href=#503 id=503 data-nosnippet>503</a>                use_position_fallback,
-<a href=#504 id=504 data-nosnippet>504</a>            )<span class="question-mark">?</span>;
-<a href=#505 id=505 data-nosnippet>505</a>
-<a href=#506 id=506 data-nosnippet>506</a>            <span class="kw">let </span>row_filter = ArrowReader::get_row_filter(
-<a href=#507 id=507 data-nosnippet>507</a>                <span class="kw-2">&amp;</span>predicate,
-<a href=#508 id=508 data-nosnippet>508</a>                record_batch_stream_builder.parquet_schema(),
-<a href=#509 id=509 data-nosnippet>509</a>                <span class="kw-2">&amp;</span>iceberg_field_ids,
-<a href=#510 id=510 data-nosnippet>510</a>                <span class="kw-2">&amp;</span>field_id_map,
-<a href=#511 id=511 data-nosnippet>511</a>            )<span class="question-mark">?</span>;
-<a href=#512 id=512 data-nosnippet>512</a>            record_batch_stream_builder = record_batch_stream_builder.with_row_filter(row_filter);
-<a href=#513 id=513 data-nosnippet>513</a>
-<a href=#514 id=514 data-nosnippet>514</a>            <span class="kw">if </span><span class="self">self</span>.row_group_filtering_enabled {
-<a href=#515 id=515 data-nosnippet>515</a>                <span class="kw">let </span>predicate_filtered_row_groups = ArrowReader::get_selected_row_group_indices(
-<a href=#516 id=516 data-nosnippet>516</a>                    <span class="kw-2">&amp;</span>predicate,
-<a href=#517 id=517 data-nosnippet>517</a>                    record_batch_stream_builder.metadata(),
-<a href=#518 id=518 data-nosnippet>518</a>                    <span class="kw-2">&amp;</span>field_id_map,
-<a href=#519 id=519 data-nosnippet>519</a>                    <span class="kw-2">&amp;</span>task.schema,
-<a href=#520 id=520 data-nosnippet>520</a>                )<span class="question-mark">?</span>;
-<a href=#521 id=521 data-nosnippet>521</a>
-<a href=#522 id=522 data-nosnippet>522</a>                <span class="comment">// Merge predicate-based filtering with byte range filtering (if present)
-<a href=#523 id=523 data-nosnippet>523</a>                // by taking the intersection of both filters
-<a href=#524 id=524 data-nosnippet>524</a>                </span>selected_row_group_indices = <span class="kw">match </span>selected_row_group_indices {
-<a href=#525 id=525 data-nosnippet>525</a>                    <span class="prelude-val">Some</span>(byte_range_filtered) =&gt; {
-<a href=#526 id=526 data-nosnippet>526</a>                        <span class="comment">// Keep only row groups that are in both filters
-<a href=#527 id=527 data-nosnippet>527</a>                        </span><span class="kw">let </span>intersection: Vec&lt;usize&gt; = byte_range_filtered
-<a href=#528 id=528 data-nosnippet>528</a>                            .into_iter()
-<a href=#529 id=529 data-nosnippet>529</a>                            .filter(|idx| predicate_filtered_row_groups.contains(idx))
-<a href=#530 id=530 data-nosnippet>530</a>                            .collect();
-<a href=#531 id=531 data-nosnippet>531</a>                        <span class="prelude-val">Some</span>(intersection)
-<a href=#532 id=532 data-nosnippet>532</a>                    }
-<a href=#533 id=533 data-nosnippet>533</a>                    <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Some</span>(predicate_filtered_row_groups),
-<a href=#534 id=534 data-nosnippet>534</a>                };
-<a href=#535 id=535 data-nosnippet>535</a>            }
-<a href=#536 id=536 data-nosnippet>536</a>
-<a href=#537 id=537 data-nosnippet>537</a>            <span class="kw">if </span><span class="self">self</span>.row_selection_enabled {
-<a href=#538 id=538 data-nosnippet>538</a>                row_selection = ArrowReader::get_row_selection_for_filter_predicate(
-<a href=#539 id=539 data-nosnippet>539</a>                    <span class="kw-2">&amp;</span>predicate,
-<a href=#540 id=540 data-nosnippet>540</a>                    record_batch_stream_builder.metadata(),
-<a href=#541 id=541 data-nosnippet>541</a>                    <span class="kw-2">&amp;</span>selected_row_group_indices,
-<a href=#542 id=542 data-nosnippet>542</a>                    <span class="kw-2">&amp;</span>field_id_map,
-<a href=#543 id=543 data-nosnippet>543</a>                    <span class="kw-2">&amp;</span>task.schema,
-<a href=#544 id=544 data-nosnippet>544</a>                )<span class="question-mark">?</span>;
-<a href=#545 id=545 data-nosnippet>545</a>            }
-<a href=#546 id=546 data-nosnippet>546</a>        }
-<a href=#547 id=547 data-nosnippet>547</a>
-<a href=#548 id=548 data-nosnippet>548</a>        <span class="kw">let </span>positional_delete_indexes = delete_filter.get_delete_vector(<span class="kw-2">&amp;</span>task);
-<a href=#549 id=549 data-nosnippet>549</a>
-<a href=#550 id=550 data-nosnippet>550</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(positional_delete_indexes) = positional_delete_indexes {
-<a href=#551 id=551 data-nosnippet>551</a>            <span class="kw">let </span>delete_row_selection = {
-<a href=#552 id=552 data-nosnippet>552</a>                <span class="kw">let </span>positional_delete_indexes = positional_delete_indexes.lock().unwrap();
-<a href=#553 id=553 data-nosnippet>553</a>
-<a href=#554 id=554 data-nosnippet>554</a>                ArrowReader::build_deletes_row_selection(
-<a href=#555 id=555 data-nosnippet>555</a>                    record_batch_stream_builder.metadata().row_groups(),
-<a href=#556 id=556 data-nosnippet>556</a>                    <span class="kw-2">&amp;</span>selected_row_group_indices,
-<a href=#557 id=557 data-nosnippet>557</a>                    <span class="kw-2">&amp;</span>positional_delete_indexes,
-<a href=#558 id=558 data-nosnippet>558</a>                )
-<a href=#559 id=559 data-nosnippet>559</a>            }<span class="question-mark">?</span>;
-<a href=#560 id=560 data-nosnippet>560</a>
-<a href=#561 id=561 data-nosnippet>561</a>            <span class="comment">// merge the row selection from the delete files with the row selection
-<a href=#562 id=562 data-nosnippet>562</a>            // from the filter predicate, if there is one from the filter predicate
-<a href=#563 id=563 data-nosnippet>563</a>            </span>row_selection = <span class="kw">match </span>row_selection {
-<a href=#564 id=564 data-nosnippet>564</a>                <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Some</span>(delete_row_selection),
-<a href=#565 id=565 data-nosnippet>565</a>                <span class="prelude-val">Some</span>(filter_row_selection) =&gt; {
-<a href=#566 id=566 data-nosnippet>566</a>                    <span class="prelude-val">Some</span>(filter_row_selection.intersection(<span class="kw-2">&amp;</span>delete_row_selection))
-<a href=#567 id=567 data-nosnippet>567</a>                }
-<a href=#568 id=568 data-nosnippet>568</a>            };
-<a href=#569 id=569 data-nosnippet>569</a>        }
-<a href=#570 id=570 data-nosnippet>570</a>
-<a href=#571 id=571 data-nosnippet>571</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(row_selection) = row_selection {
-<a href=#572 id=572 data-nosnippet>572</a>            record_batch_stream_builder =
-<a href=#573 id=573 data-nosnippet>573</a>                record_batch_stream_builder.with_row_selection(row_selection);
-<a href=#574 id=574 data-nosnippet>574</a>        }
-<a href=#575 id=575 data-nosnippet>575</a>
-<a href=#576 id=576 data-nosnippet>576</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(selected_row_group_indices) = selected_row_group_indices {
-<a href=#577 id=577 data-nosnippet>577</a>            record_batch_stream_builder =
-<a href=#578 id=578 data-nosnippet>578</a>                record_batch_stream_builder.with_row_groups(selected_row_group_indices);
-<a href=#579 id=579 data-nosnippet>579</a>        }
-<a href=#580 id=580 data-nosnippet>580</a>
-<a href=#581 id=581 data-nosnippet>581</a>        <span class="comment">// Build the batch stream and send all the RecordBatches that it generates
-<a href=#582 id=582 data-nosnippet>582</a>        // to the requester.
-<a href=#583 id=583 data-nosnippet>583</a>        </span><span class="kw">let </span>record_batch_stream =
-<a href=#584 id=584 data-nosnippet>584</a>            record_batch_stream_builder
-<a href=#585 id=585 data-nosnippet>585</a>                .build()<span class="question-mark">?
-<a href=#586 id=586 data-nosnippet>586</a>                </span>.map(<span class="kw">move </span>|batch| <span class="kw">match </span>batch {
-<a href=#587 id=587 data-nosnippet>587</a>                    <span class="prelude-val">Ok</span>(batch) =&gt; {
-<a href=#588 id=588 data-nosnippet>588</a>                        <span class="comment">// Process the record batch (type promotion, column reordering, virtual fields, etc.)
-<a href=#589 id=589 data-nosnippet>589</a>                        </span>record_batch_transformer.process_record_batch(batch)
-<a href=#590 id=590 data-nosnippet>590</a>                    }
-<a href=#591 id=591 data-nosnippet>591</a>                    <span class="prelude-val">Err</span>(err) =&gt; <span class="prelude-val">Err</span>(err.into()),
-<a href=#592 id=592 data-nosnippet>592</a>                });
+<a href=#487 id=487 data-nosnippet>487</a>        <span class="comment">// There are three possible sources for potential lists of selected RowGroup indices,
+<a href=#488 id=488 data-nosnippet>488</a>        // and two for `RowSelection`s.
+<a href=#489 id=489 data-nosnippet>489</a>        // Selected RowGroup index lists can come from three sources:
+<a href=#490 id=490 data-nosnippet>490</a>        //   * When task.start and task.length specify a byte range (file splitting);
+<a href=#491 id=491 data-nosnippet>491</a>        //   * When there are equality delete files that are applicable;
+<a href=#492 id=492 data-nosnippet>492</a>        //   * When there is a scan predicate and row_group_filtering_enabled = true.
+<a href=#493 id=493 data-nosnippet>493</a>        // `RowSelection`s can be created in either or both of the following cases:
+<a href=#494 id=494 data-nosnippet>494</a>        //   * When there are positional delete files that are applicable;
+<a href=#495 id=495 data-nosnippet>495</a>        //   * When there is a scan predicate and row_selection_enabled = true
+<a href=#496 id=496 data-nosnippet>496</a>        // Note that row group filtering from predicates only happens when
+<a href=#497 id=497 data-nosnippet>497</a>        // there is a scan predicate AND row_group_filtering_enabled = true,
+<a href=#498 id=498 data-nosnippet>498</a>        // but we perform row selection filtering if there are applicable
+<a href=#499 id=499 data-nosnippet>499</a>        // equality delete files OR (there is a scan predicate AND row_selection_enabled),
+<a href=#500 id=500 data-nosnippet>500</a>        // since the only implemented method of applying positional deletes is
+<a href=#501 id=501 data-nosnippet>501</a>        // by using a `RowSelection`.
+<a href=#502 id=502 data-nosnippet>502</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>selected_row_group_indices = <span class="prelude-val">None</span>;
+<a href=#503 id=503 data-nosnippet>503</a>        <span class="kw">let </span><span class="kw-2">mut </span>row_selection = <span class="prelude-val">None</span>;
+<a href=#504 id=504 data-nosnippet>504</a>
+<a href=#505 id=505 data-nosnippet>505</a>        <span class="comment">// Filter row groups based on byte range from task.start and task.length.
+<a href=#506 id=506 data-nosnippet>506</a>        // If both start and length are 0, read the entire file (backwards compatibility).
+<a href=#507 id=507 data-nosnippet>507</a>        </span><span class="kw">if </span>task.start != <span class="number">0 </span>|| task.length != <span class="number">0 </span>{
+<a href=#508 id=508 data-nosnippet>508</a>            <span class="kw">let </span>byte_range_filtered_row_groups = ArrowReader::filter_row_groups_by_byte_range(
+<a href=#509 id=509 data-nosnippet>509</a>                record_batch_stream_builder.metadata(),
+<a href=#510 id=510 data-nosnippet>510</a>                task.start,
+<a href=#511 id=511 data-nosnippet>511</a>                task.length,
+<a href=#512 id=512 data-nosnippet>512</a>            )<span class="question-mark">?</span>;
+<a href=#513 id=513 data-nosnippet>513</a>            selected_row_group_indices = <span class="prelude-val">Some</span>(byte_range_filtered_row_groups);
+<a href=#514 id=514 data-nosnippet>514</a>        }
+<a href=#515 id=515 data-nosnippet>515</a>
+<a href=#516 id=516 data-nosnippet>516</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(predicate) = final_predicate {
+<a href=#517 id=517 data-nosnippet>517</a>            <span class="kw">let </span>(iceberg_field_ids, field_id_map) = ArrowReader::build_field_id_set_and_map(
+<a href=#518 id=518 data-nosnippet>518</a>                record_batch_stream_builder.parquet_schema(),
+<a href=#519 id=519 data-nosnippet>519</a>                record_batch_stream_builder.schema(),
+<a href=#520 id=520 data-nosnippet>520</a>                <span class="kw-2">&amp;</span>predicate,
+<a href=#521 id=521 data-nosnippet>521</a>                use_position_fallback,
+<a href=#522 id=522 data-nosnippet>522</a>            )<span class="question-mark">?</span>;
+<a href=#523 id=523 data-nosnippet>523</a>
+<a href=#524 id=524 data-nosnippet>524</a>            <span class="kw">let </span>row_filter = ArrowReader::get_row_filter(
+<a href=#525 id=525 data-nosnippet>525</a>                <span class="kw-2">&amp;</span>predicate,
+<a href=#526 id=526 data-nosnippet>526</a>                record_batch_stream_builder.parquet_schema(),
+<a href=#527 id=527 data-nosnippet>527</a>                <span class="kw-2">&amp;</span>iceberg_field_ids,
+<a href=#528 id=528 data-nosnippet>528</a>                <span class="kw-2">&amp;</span>field_id_map,
+<a href=#529 id=529 data-nosnippet>529</a>            )<span class="question-mark">?</span>;
+<a href=#530 id=530 data-nosnippet>530</a>            record_batch_stream_builder = record_batch_stream_builder.with_row_filter(row_filter);
+<a href=#531 id=531 data-nosnippet>531</a>
+<a href=#532 id=532 data-nosnippet>532</a>            <span class="kw">if </span><span class="self">self</span>.row_group_filtering_enabled {
+<a href=#533 id=533 data-nosnippet>533</a>                <span class="kw">let </span>predicate_filtered_row_groups = ArrowReader::get_selected_row_group_indices(
+<a href=#534 id=534 data-nosnippet>534</a>                    <span class="kw-2">&amp;</span>predicate,
+<a href=#535 id=535 data-nosnippet>535</a>                    record_batch_stream_builder.metadata(),
+<a href=#536 id=536 data-nosnippet>536</a>                    <span class="kw-2">&amp;</span>field_id_map,
+<a href=#537 id=537 data-nosnippet>537</a>                    <span class="kw-2">&amp;</span>task.schema,
+<a href=#538 id=538 data-nosnippet>538</a>                )<span class="question-mark">?</span>;
+<a href=#539 id=539 data-nosnippet>539</a>
+<a href=#540 id=540 data-nosnippet>540</a>                <span class="comment">// Merge predicate-based filtering with byte range filtering (if present)
+<a href=#541 id=541 data-nosnippet>541</a>                // by taking the intersection of both filters
+<a href=#542 id=542 data-nosnippet>542</a>                </span>selected_row_group_indices = <span class="kw">match </span>selected_row_group_indices {
+<a href=#543 id=543 data-nosnippet>543</a>                    <span class="prelude-val">Some</span>(byte_range_filtered) =&gt; {
+<a href=#544 id=544 data-nosnippet>544</a>                        <span class="comment">// Keep only row groups that are in both filters
+<a href=#545 id=545 data-nosnippet>545</a>                        </span><span class="kw">let </span>intersection: Vec&lt;usize&gt; = byte_range_filtered
+<a href=#546 id=546 data-nosnippet>546</a>                            .into_iter()
+<a href=#547 id=547 data-nosnippet>547</a>                            .filter(|idx| predicate_filtered_row_groups.contains(idx))
+<a href=#548 id=548 data-nosnippet>548</a>                            .collect();
+<a href=#549 id=549 data-nosnippet>549</a>                        <span class="prelude-val">Some</span>(intersection)
+<a href=#550 id=550 data-nosnippet>550</a>                    }
+<a href=#551 id=551 data-nosnippet>551</a>                    <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Some</span>(predicate_filtered_row_groups),
+<a href=#552 id=552 data-nosnippet>552</a>                };
+<a href=#553 id=553 data-nosnippet>553</a>            }
+<a href=#554 id=554 data-nosnippet>554</a>
+<a href=#555 id=555 data-nosnippet>555</a>            <span class="kw">if </span><span class="self">self</span>.row_selection_enabled {
+<a href=#556 id=556 data-nosnippet>556</a>                row_selection = ArrowReader::get_row_selection_for_filter_predicate(
+<a href=#557 id=557 data-nosnippet>557</a>                    <span class="kw-2">&amp;</span>predicate,
+<a href=#558 id=558 data-nosnippet>558</a>                    record_batch_stream_builder.metadata(),
+<a href=#559 id=559 data-nosnippet>559</a>                    <span class="kw-2">&amp;</span>selected_row_group_indices,
+<a href=#560 id=560 data-nosnippet>560</a>                    <span class="kw-2">&amp;</span>field_id_map,
+<a href=#561 id=561 data-nosnippet>561</a>                    <span class="kw-2">&amp;</span>task.schema,
+<a href=#562 id=562 data-nosnippet>562</a>                )<span class="question-mark">?</span>;
+<a href=#563 id=563 data-nosnippet>563</a>            }
+<a href=#564 id=564 data-nosnippet>564</a>        }
+<a href=#565 id=565 data-nosnippet>565</a>
+<a href=#566 id=566 data-nosnippet>566</a>        <span class="kw">let </span>positional_delete_indexes = delete_filter.get_delete_vector(<span class="kw-2">&amp;</span>task);
+<a href=#567 id=567 data-nosnippet>567</a>
+<a href=#568 id=568 data-nosnippet>568</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(positional_delete_indexes) = positional_delete_indexes {
+<a href=#569 id=569 data-nosnippet>569</a>            <span class="kw">let </span>delete_row_selection = {
+<a href=#570 id=570 data-nosnippet>570</a>                <span class="kw">let </span>positional_delete_indexes = positional_delete_indexes.lock().unwrap();
+<a href=#571 id=571 data-nosnippet>571</a>
+<a href=#572 id=572 data-nosnippet>572</a>                ArrowReader::build_deletes_row_selection(
+<a href=#573 id=573 data-nosnippet>573</a>                    record_batch_stream_builder.metadata().row_groups(),
+<a href=#574 id=574 data-nosnippet>574</a>                    <span class="kw-2">&amp;</span>selected_row_group_indices,
+<a href=#575 id=575 data-nosnippet>575</a>                    <span class="kw-2">&amp;</span>positional_delete_indexes,
+<a href=#576 id=576 data-nosnippet>576</a>                )
+<a href=#577 id=577 data-nosnippet>577</a>            }<span class="question-mark">?</span>;
+<a href=#578 id=578 data-nosnippet>578</a>
+<a href=#579 id=579 data-nosnippet>579</a>            <span class="comment">// merge the row selection from the delete files with the row selection
+<a href=#580 id=580 data-nosnippet>580</a>            // from the filter predicate, if there is one from the filter predicate
+<a href=#581 id=581 data-nosnippet>581</a>            </span>row_selection = <span class="kw">match </span>row_selection {
+<a href=#582 id=582 data-nosnippet>582</a>                <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Some</span>(delete_row_selection),
+<a href=#583 id=583 data-nosnippet>583</a>                <span class="prelude-val">Some</span>(filter_row_selection) =&gt; {
+<a href=#584 id=584 data-nosnippet>584</a>                    <span class="prelude-val">Some</span>(filter_row_selection.intersection(<span class="kw-2">&amp;</span>delete_row_selection))
+<a href=#585 id=585 data-nosnippet>585</a>                }
+<a href=#586 id=586 data-nosnippet>586</a>            };
+<a href=#587 id=587 data-nosnippet>587</a>        }
+<a href=#588 id=588 data-nosnippet>588</a>
+<a href=#589 id=589 data-nosnippet>589</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(row_selection) = row_selection {
+<a href=#590 id=590 data-nosnippet>590</a>            record_batch_stream_builder =
+<a href=#591 id=591 data-nosnippet>591</a>                record_batch_stream_builder.with_row_selection(row_selection);
+<a href=#592 id=592 data-nosnippet>592</a>        }
 <a href=#593 id=593 data-nosnippet>593</a>
-<a href=#594 id=594 data-nosnippet>594</a>        <span class="prelude-val">Ok</span>(Box::pin(record_batch_stream) <span class="kw">as </span>ArrowRecordBatchStream)
-<a href=#595 id=595 data-nosnippet>595</a>    }
-<a href=#596 id=596 data-nosnippet>596</a>}
-<a href=#597 id=597 data-nosnippet>597</a>
-<a href=#598 id=598 data-nosnippet>598</a><span class="kw">impl </span>ArrowReader {
-<a href=#599 id=599 data-nosnippet>599</a>    <span class="doccomment">/// Opens a Parquet file and loads its metadata, wrapping the reader with
-<a href=#600 id=600 data-nosnippet>600</a>    /// [`CountingFileRead`] so all I/O is accumulated into `bytes_read`.
-<a href=#601 id=601 data-nosnippet>601</a>    </span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">async fn </span>open_parquet_file(
-<a href=#602 id=602 data-nosnippet>602</a>        data_file_path: <span class="kw-2">&amp;</span>str,
-<a href=#603 id=603 data-nosnippet>603</a>        file_io: <span class="kw-2">&amp;</span>FileIO,
-<a href=#604 id=604 data-nosnippet>604</a>        file_size_in_bytes: u64,
-<a href=#605 id=605 data-nosnippet>605</a>        parquet_read_options: ParquetReadOptions,
-<a href=#606 id=606 data-nosnippet>606</a>        bytes_read: <span class="kw-2">&amp;</span>Arc&lt;AtomicU64&gt;,
-<a href=#607 id=607 data-nosnippet>607</a>        key_metadata: <span class="prelude-ty">Option</span>&lt;<span class="kw-2">&amp;</span>[u8]&gt;,
-<a href=#608 id=608 data-nosnippet>608</a>    ) -&gt; <span class="prelude-ty">Result</span>&lt;(ArrowFileReader, ArrowReaderMetadata)&gt; {
-<a href=#609 id=609 data-nosnippet>609</a>        <span class="kw">let </span>parquet_file = file_io.new_input(data_file_path)<span class="question-mark">?</span>;
-<a href=#610 id=610 data-nosnippet>610</a>        <span class="kw">let </span>counting_reader =
-<a href=#611 id=611 data-nosnippet>611</a>            CountingFileRead::new(parquet_file.reader().<span class="kw">await</span><span class="question-mark">?</span>, Arc::clone(bytes_read));
-<a href=#612 id=612 data-nosnippet>612</a>        <span class="self">Self</span>::build_parquet_reader(
-<a href=#613 id=613 data-nosnippet>613</a>            Box::new(counting_reader),
-<a href=#614 id=614 data-nosnippet>614</a>            file_size_in_bytes,
-<a href=#615 id=615 data-nosnippet>615</a>            parquet_read_options,
-<a href=#616 id=616 data-nosnippet>616</a>            key_metadata,
-<a href=#617 id=617 data-nosnippet>617</a>        )
-<a href=#618 id=618 data-nosnippet>618</a>        .<span class="kw">await
-<a href=#619 id=619 data-nosnippet>619</a>    </span>}
-<a href=#620 id=620 data-nosnippet>620</a>
-<a href=#621 id=621 data-nosnippet>621</a>    <span class="kw">async fn </span>build_parquet_reader(
-<a href=#622 id=622 data-nosnippet>622</a>        parquet_reader: Box&lt;<span class="kw">dyn </span>FileRead&gt;,
-<a href=#623 id=623 data-nosnippet>623</a>        file_size_in_bytes: u64,
-<a href=#624 id=624 data-nosnippet>624</a>        parquet_read_options: ParquetReadOptions,
+<a href=#594 id=594 data-nosnippet>594</a>        <span class="kw">if let </span><span class="prelude-val">Some</span>(selected_row_group_indices) = selected_row_group_indices {
+<a href=#595 id=595 data-nosnippet>595</a>            record_batch_stream_builder =
+<a href=#596 id=596 data-nosnippet>596</a>                record_batch_stream_builder.with_row_groups(selected_row_group_indices);
+<a href=#597 id=597 data-nosnippet>597</a>        }
+<a href=#598 id=598 data-nosnippet>598</a>
+<a href=#599 id=599 data-nosnippet>599</a>        <span class="comment">// Build the batch stream and send all the RecordBatches that it generates
+<a href=#600 id=600 data-nosnippet>600</a>        // to the requester.
+<a href=#601 id=601 data-nosnippet>601</a>        </span><span class="kw">let </span>record_batch_stream =
+<a href=#602 id=602 data-nosnippet>602</a>            record_batch_stream_builder
+<a href=#603 id=603 data-nosnippet>603</a>                .build()<span class="question-mark">?
+<a href=#604 id=604 data-nosnippet>604</a>                </span>.map(<span class="kw">move </span>|batch| <span class="kw">match </span>batch {
+<a href=#605 id=605 data-nosnippet>605</a>                    <span class="prelude-val">Ok</span>(batch) =&gt; {
+<a href=#606 id=606 data-nosnippet>606</a>                        <span class="comment">// Process the record batch (type promotion, column reordering, virtual fields, etc.)
+<a href=#607 id=607 data-nosnippet>607</a>                        </span>record_batch_transformer.process_record_batch(batch)
+<a href=#608 id=608 data-nosnippet>608</a>                    }
+<a href=#609 id=609 data-nosnippet>609</a>                    <span class="prelude-val">Err</span>(err) =&gt; <span class="prelude-val">Err</span>(err.into()),
+<a href=#610 id=610 data-nosnippet>610</a>                });
+<a href=#611 id=611 data-nosnippet>611</a>
+<a href=#612 id=612 data-nosnippet>612</a>        <span class="prelude-val">Ok</span>(Box::pin(record_batch_stream) <span class="kw">as </span>ArrowRecordBatchStream)
+<a href=#613 id=613 data-nosnippet>613</a>    }
+<a href=#614 id=614 data-nosnippet>614</a>}
+<a href=#615 id=615 data-nosnippet>615</a>
+<a href=#616 id=616 data-nosnippet>616</a><span class="kw">impl </span>ArrowReader {
+<a href=#617 id=617 data-nosnippet>617</a>    <span class="doccomment">/// Opens a Parquet file and loads its metadata, wrapping the reader with
+<a href=#618 id=618 data-nosnippet>618</a>    /// [`CountingFileRead`] so all I/O is accumulated into `bytes_read`.
+<a href=#619 id=619 data-nosnippet>619</a>    </span><span class="kw">pub</span>(<span class="kw">crate</span>) <span class="kw">async fn </span>open_parquet_file(
+<a href=#620 id=620 data-nosnippet>620</a>        data_file_path: <span class="kw-2">&amp;</span>str,
+<a href=#621 id=621 data-nosnippet>621</a>        file_io: <span class="kw-2">&amp;</span>FileIO,
+<a href=#622 id=622 data-nosnippet>622</a>        file_size_in_bytes: u64,
+<a href=#623 id=623 data-nosnippet>623</a>        parquet_read_options: ParquetReadOptions,
+<a href=#624 id=624 data-nosnippet>624</a>        bytes_read: <span class="kw-2">&amp;</span>Arc&lt;AtomicU64&gt;,
 <a href=#625 id=625 data-nosnippet>625</a>        key_metadata: <span class="prelude-ty">Option</span>&lt;<span class="kw-2">&amp;</span>[u8]&gt;,
 <a href=#626 id=626 data-nosnippet>626</a>    ) -&gt; <span class="prelude-ty">Result</span>&lt;(ArrowFileReader, ArrowReaderMetadata)&gt; {
-<a href=#627 id=627 data-nosnippet>627</a>        <span class="kw">let </span><span class="kw-2">mut </span>reader = ArrowFileReader::new(
-<a href=#628 id=628 data-nosnippet>628</a>            FileMetadata {
-<a href=#629 id=629 data-nosnippet>629</a>                size: file_size_in_bytes,
-<a href=#630 id=630 data-nosnippet>630</a>            },
-<a href=#631 id=631 data-nosnippet>631</a>            parquet_reader,
-<a href=#632 id=632 data-nosnippet>632</a>        )
-<a href=#633 id=633 data-nosnippet>633</a>        .with_parquet_read_options(parquet_read_options);
-<a href=#634 id=634 data-nosnippet>634</a>
-<a href=#635 id=635 data-nosnippet>635</a>        <span class="kw">let </span>arrow_reader_options = <span class="self">Self</span>::build_arrow_reader_options(key_metadata)<span class="question-mark">?</span>;
-<a href=#636 id=636 data-nosnippet>636</a>
-<a href=#637 id=637 data-nosnippet>637</a>        <span class="kw">let </span>arrow_metadata = ArrowReaderMetadata::load_async(<span class="kw-2">&amp;mut </span>reader, arrow_reader_options)
-<a href=#638 id=638 data-nosnippet>638</a>            .<span class="kw">await
-<a href=#639 id=639 data-nosnippet>639</a>            </span>.map_err(|e| {
-<a href=#640 id=640 data-nosnippet>640</a>                Error::new(ErrorKind::Unexpected, <span class="string">"Failed to load Parquet metadata"</span>).with_source(e)
-<a href=#641 id=641 data-nosnippet>641</a>            })<span class="question-mark">?</span>;
-<a href=#642 id=642 data-nosnippet>642</a>
-<a href=#643 id=643 data-nosnippet>643</a>        <span class="prelude-val">Ok</span>((reader, arrow_metadata))
-<a href=#644 id=644 data-nosnippet>644</a>    }
-<a href=#645 id=645 data-nosnippet>645</a>
-<a href=#646 id=646 data-nosnippet>646</a>    <span class="doccomment">/// Builds `ArrowReaderOptions`, adding `FileDecryptionProperties` when
-<a href=#647 id=647 data-nosnippet>647</a>    /// key metadata is present for Parquet Modular Encryption.
-<a href=#648 id=648 data-nosnippet>648</a>    </span><span class="kw">fn </span>build_arrow_reader_options(key_metadata: <span class="prelude-ty">Option</span>&lt;<span class="kw-2">&amp;</span>[u8]&gt;) -&gt; <span class="prelude-ty">Result</span>&lt;ArrowReaderOptions&gt; {
-<a href=#649 id=649 data-nosnippet>649</a>        <span class="kw">match </span>key_metadata {
-<a href=#650 id=650 data-nosnippet>650</a>            <span class="prelude-val">Some</span>(km) =&gt; {
-<a href=#651 id=651 data-nosnippet>651</a>                <span class="kw">let </span>standard_key_metadata = StandardKeyMetadata::decode(km)<span class="question-mark">?</span>;
-<a href=#652 id=652 data-nosnippet>652</a>                <span class="kw">let </span><span class="kw-2">mut </span>builder = FileDecryptionProperties::builder(
-<a href=#653 id=653 data-nosnippet>653</a>                    standard_key_metadata.encryption_key().as_bytes().to_vec(),
-<a href=#654 id=654 data-nosnippet>654</a>                );
-<a href=#655 id=655 data-nosnippet>655</a>                <span class="kw">if let </span><span class="prelude-val">Some</span>(aad) = standard_key_metadata.aad_prefix() {
-<a href=#656 id=656 data-nosnippet>656</a>                    builder = builder.with_aad_prefix(aad.to_vec());
-<a href=#657 id=657 data-nosnippet>657</a>                }
-<a href=#658 id=658 data-nosnippet>658</a>                <span class="kw">let </span>decryption_properties = builder.build().map_err(|e| {
-<a href=#659 id=659 data-nosnippet>659</a>                    Error::new(
-<a href=#660 id=660 data-nosnippet>660</a>                        ErrorKind::Unexpected,
-<a href=#661 id=661 data-nosnippet>661</a>                        <span class="string">"Failed to build Parquet file decryption properties"</span>,
-<a href=#662 id=662 data-nosnippet>662</a>                    )
-<a href=#663 id=663 data-nosnippet>663</a>                    .with_source(e)
-<a href=#664 id=664 data-nosnippet>664</a>                })<span class="question-mark">?</span>;
-<a href=#665 id=665 data-nosnippet>665</a>                <span class="prelude-val">Ok</span>(
-<a href=#666 id=666 data-nosnippet>666</a>                    ArrowReaderOptions::new()
-<a href=#667 id=667 data-nosnippet>667</a>                        .with_file_decryption_properties(decryption_properties),
-<a href=#668 id=668 data-nosnippet>668</a>                )
-<a href=#669 id=669 data-nosnippet>669</a>            }
-<a href=#670 id=670 data-nosnippet>670</a>            <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Ok</span>(ArrowReaderOptions::default()),
-<a href=#671 id=671 data-nosnippet>671</a>        }
-<a href=#672 id=672 data-nosnippet>672</a>    }
-<a href=#673 id=673 data-nosnippet>673</a>}
-<a href=#674 id=674 data-nosnippet>674</a>
-<a href=#675 id=675 data-nosnippet>675</a><span class="attr">#[cfg(test)]
-<a href=#676 id=676 data-nosnippet>676</a></span><span class="kw">mod </span>tests {
-<a href=#677 id=677 data-nosnippet>677</a>    <span class="kw">use </span>std::collections::HashMap;
-<a href=#678 id=678 data-nosnippet>678</a>    <span class="kw">use </span>std::fs::File;
-<a href=#679 id=679 data-nosnippet>679</a>    <span class="kw">use </span>std::sync::Arc;
-<a href=#680 id=680 data-nosnippet>680</a>
-<a href=#681 id=681 data-nosnippet>681</a>    <span class="kw">use </span>arrow_array::cast::AsArray;
-<a href=#682 id=682 data-nosnippet>682</a>    <span class="kw">use </span>arrow_array::{Array, ArrayRef, Int32Array, Int64Array, RecordBatch};
-<a href=#683 id=683 data-nosnippet>683</a>    <span class="kw">use </span>arrow_schema::{DataType, Field, Schema <span class="kw">as </span>ArrowSchema};
-<a href=#684 id=684 data-nosnippet>684</a>    <span class="kw">use </span>futures::TryStreamExt;
-<a href=#685 id=685 data-nosnippet>685</a>    <span class="kw">use </span>parquet::arrow::{ArrowWriter, PARQUET_FIELD_ID_META_KEY};
-<a href=#686 id=686 data-nosnippet>686</a>    <span class="kw">use </span>parquet::basic::Compression;
-<a href=#687 id=687 data-nosnippet>687</a>    <span class="kw">use </span>parquet::file::properties::WriterProperties;
-<a href=#688 id=688 data-nosnippet>688</a>    <span class="kw">use </span>tempfile::TempDir;
-<a href=#689 id=689 data-nosnippet>689</a>
-<a href=#690 id=690 data-nosnippet>690</a>    <span class="kw">use </span><span class="kw">crate</span>::Runtime;
-<a href=#691 id=691 data-nosnippet>691</a>    <span class="kw">use </span><span class="kw">crate</span>::arrow::ArrowReaderBuilder;
-<a href=#692 id=692 data-nosnippet>692</a>    <span class="kw">use </span><span class="kw">crate</span>::arrow::test_utils::write_encrypted_parquet;
-<a href=#693 id=693 data-nosnippet>693</a>    <span class="kw">use </span><span class="kw">crate</span>::io::FileIO;
-<a href=#694 id=694 data-nosnippet>694</a>    <span class="kw">use </span><span class="kw">crate</span>::scan::{FileScanTask, FileScanTaskStream};
-<a href=#695 id=695 data-nosnippet>695</a>    <span class="kw">use </span><span class="kw">crate</span>::spec::{DataFileFormat, NestedField, PrimitiveType, Schema, SchemaRef, Type};
-<a href=#696 id=696 data-nosnippet>696</a>
-<a href=#697 id=697 data-nosnippet>697</a>    <span class="comment">// INT96 encoding: [nanos_low_u32, nanos_high_u32, julian_day_u32]
-<a href=#698 id=698 data-nosnippet>698</a>    // Julian day 2_440_588 = Unix epoch (1970-01-01)
-<a href=#699 id=699 data-nosnippet>699</a>    </span><span class="kw">const </span>UNIX_EPOCH_JULIAN: i64 = <span class="number">2_440_588</span>;
-<a href=#700 id=700 data-nosnippet>700</a>    <span class="kw">const </span>MICROS_PER_DAY: i64 = <span class="number">86_400_000_000</span>;
-<a href=#701 id=701 data-nosnippet>701</a>    <span class="comment">// Noon on 3333-01-01 (Julian day 2_953_529) — outside the i64 nanosecond range (~1677-2262).
-<a href=#702 id=702 data-nosnippet>702</a>    </span><span class="kw">const </span>INT96_TEST_NANOS_WITHIN_DAY: u64 = <span class="number">43_200_000_000_000</span>;
-<a href=#703 id=703 data-nosnippet>703</a>    <span class="kw">const </span>INT96_TEST_JULIAN_DAY: u32 = <span class="number">2_953_529</span>;
-<a href=#704 id=704 data-nosnippet>704</a>
-<a href=#705 id=705 data-nosnippet>705</a>    <span class="kw">fn </span>make_int96_test_value() -&gt; (parquet::data_type::Int96, i64) {
-<a href=#706 id=706 data-nosnippet>706</a>        <span class="kw">let </span><span class="kw-2">mut </span>val = parquet::data_type::Int96::new();
-<a href=#707 id=707 data-nosnippet>707</a>        val.set_data(
-<a href=#708 id=708 data-nosnippet>708</a>            (INT96_TEST_NANOS_WITHIN_DAY &amp; <span class="number">0xFFFFFFFF</span>) <span class="kw">as </span>u32,
-<a href=#709 id=709 data-nosnippet>709</a>            (INT96_TEST_NANOS_WITHIN_DAY &gt;&gt; <span class="number">32</span>) <span class="kw">as </span>u32,
-<a href=#710 id=710 data-nosnippet>710</a>            INT96_TEST_JULIAN_DAY,
-<a href=#711 id=711 data-nosnippet>711</a>        );
-<a href=#712 id=712 data-nosnippet>712</a>        <span class="kw">let </span>expected_micros = (INT96_TEST_JULIAN_DAY <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY
-<a href=#713 id=713 data-nosnippet>713</a>            + (INT96_TEST_NANOS_WITHIN_DAY / <span class="number">1_000</span>) <span class="kw">as </span>i64;
-<a href=#714 id=714 data-nosnippet>714</a>        (val, expected_micros)
-<a href=#715 id=715 data-nosnippet>715</a>    }
-<a href=#716 id=716 data-nosnippet>716</a>
-<a href=#717 id=717 data-nosnippet>717</a>    <span class="kw">async fn </span>read_int96_batches(
-<a href=#718 id=718 data-nosnippet>718</a>        file_path: <span class="kw-2">&amp;</span>str,
-<a href=#719 id=719 data-nosnippet>719</a>        schema: SchemaRef,
-<a href=#720 id=720 data-nosnippet>720</a>        project_field_ids: Vec&lt;i32&gt;,
-<a href=#721 id=721 data-nosnippet>721</a>    ) -&gt; Vec&lt;RecordBatch&gt; {
-<a href=#722 id=722 data-nosnippet>722</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#723 id=723 data-nosnippet>723</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#724 id=724 data-nosnippet>724</a>
-<a href=#725 id=725 data-nosnippet>725</a>        <span class="kw">let </span>file_size = std::fs::metadata(file_path).unwrap().len();
-<a href=#726 id=726 data-nosnippet>726</a>        <span class="kw">let </span>task = FileScanTask::builder()
-<a href=#727 id=727 data-nosnippet>727</a>            .with_file_size_in_bytes(file_size)
-<a href=#728 id=728 data-nosnippet>728</a>            .with_start(<span class="number">0</span>)
-<a href=#729 id=729 data-nosnippet>729</a>            .with_length(file_size)
-<a href=#730 id=730 data-nosnippet>730</a>            .with_data_file_path(file_path.to_string())
-<a href=#731 id=731 data-nosnippet>731</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#732 id=732 data-nosnippet>732</a>            .with_schema(schema)
-<a href=#733 id=733 data-nosnippet>733</a>            .with_project_field_ids(project_field_ids)
-<a href=#734 id=734 data-nosnippet>734</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#735 id=735 data-nosnippet>735</a>            .build();
-<a href=#736 id=736 data-nosnippet>736</a>
-<a href=#737 id=737 data-nosnippet>737</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#738 id=738 data-nosnippet>738</a>        reader
-<a href=#739 id=739 data-nosnippet>739</a>            .read(tasks)
-<a href=#740 id=740 data-nosnippet>740</a>            .unwrap()
-<a href=#741 id=741 data-nosnippet>741</a>            .stream()
-<a href=#742 id=742 data-nosnippet>742</a>            .try_collect()
-<a href=#743 id=743 data-nosnippet>743</a>            .<span class="kw">await
-<a href=#744 id=744 data-nosnippet>744</a>            </span>.unwrap()
-<a href=#745 id=745 data-nosnippet>745</a>    }
+<a href=#627 id=627 data-nosnippet>627</a>        <span class="kw">let </span>parquet_file = file_io.new_input(data_file_path)<span class="question-mark">?</span>;
+<a href=#628 id=628 data-nosnippet>628</a>        <span class="kw">let </span>counting_reader =
+<a href=#629 id=629 data-nosnippet>629</a>            CountingFileRead::new(parquet_file.reader().<span class="kw">await</span><span class="question-mark">?</span>, Arc::clone(bytes_read));
+<a href=#630 id=630 data-nosnippet>630</a>        <span class="self">Self</span>::build_parquet_reader(
+<a href=#631 id=631 data-nosnippet>631</a>            Box::new(counting_reader),
+<a href=#632 id=632 data-nosnippet>632</a>            file_size_in_bytes,
+<a href=#633 id=633 data-nosnippet>633</a>            parquet_read_options,
+<a href=#634 id=634 data-nosnippet>634</a>            key_metadata,
+<a href=#635 id=635 data-nosnippet>635</a>        )
+<a href=#636 id=636 data-nosnippet>636</a>        .<span class="kw">await
+<a href=#637 id=637 data-nosnippet>637</a>    </span>}
+<a href=#638 id=638 data-nosnippet>638</a>
+<a href=#639 id=639 data-nosnippet>639</a>    <span class="kw">async fn </span>build_parquet_reader(
+<a href=#640 id=640 data-nosnippet>640</a>        parquet_reader: Box&lt;<span class="kw">dyn </span>FileRead&gt;,
+<a href=#641 id=641 data-nosnippet>641</a>        file_size_in_bytes: u64,
+<a href=#642 id=642 data-nosnippet>642</a>        parquet_read_options: ParquetReadOptions,
+<a href=#643 id=643 data-nosnippet>643</a>        key_metadata: <span class="prelude-ty">Option</span>&lt;<span class="kw-2">&amp;</span>[u8]&gt;,
+<a href=#644 id=644 data-nosnippet>644</a>    ) -&gt; <span class="prelude-ty">Result</span>&lt;(ArrowFileReader, ArrowReaderMetadata)&gt; {
+<a href=#645 id=645 data-nosnippet>645</a>        <span class="kw">let </span><span class="kw-2">mut </span>reader = ArrowFileReader::new(
+<a href=#646 id=646 data-nosnippet>646</a>            FileMetadata {
+<a href=#647 id=647 data-nosnippet>647</a>                size: file_size_in_bytes,
+<a href=#648 id=648 data-nosnippet>648</a>            },
+<a href=#649 id=649 data-nosnippet>649</a>            parquet_reader,
+<a href=#650 id=650 data-nosnippet>650</a>        )
+<a href=#651 id=651 data-nosnippet>651</a>        .with_parquet_read_options(parquet_read_options);
+<a href=#652 id=652 data-nosnippet>652</a>
+<a href=#653 id=653 data-nosnippet>653</a>        <span class="kw">let </span>arrow_reader_options = <span class="self">Self</span>::build_arrow_reader_options(key_metadata)<span class="question-mark">?</span>;
+<a href=#654 id=654 data-nosnippet>654</a>
+<a href=#655 id=655 data-nosnippet>655</a>        <span class="kw">let </span>arrow_metadata = ArrowReaderMetadata::load_async(<span class="kw-2">&amp;mut </span>reader, arrow_reader_options)
+<a href=#656 id=656 data-nosnippet>656</a>            .<span class="kw">await
+<a href=#657 id=657 data-nosnippet>657</a>            </span>.map_err(|e| {
+<a href=#658 id=658 data-nosnippet>658</a>                Error::new(ErrorKind::Unexpected, <span class="string">"Failed to load Parquet metadata"</span>).with_source(e)
+<a href=#659 id=659 data-nosnippet>659</a>            })<span class="question-mark">?</span>;
+<a href=#660 id=660 data-nosnippet>660</a>
+<a href=#661 id=661 data-nosnippet>661</a>        <span class="prelude-val">Ok</span>((reader, arrow_metadata))
+<a href=#662 id=662 data-nosnippet>662</a>    }
+<a href=#663 id=663 data-nosnippet>663</a>
+<a href=#664 id=664 data-nosnippet>664</a>    <span class="doccomment">/// Builds `ArrowReaderOptions`, adding `FileDecryptionProperties` when
+<a href=#665 id=665 data-nosnippet>665</a>    /// key metadata is present for Parquet Modular Encryption.
+<a href=#666 id=666 data-nosnippet>666</a>    </span><span class="kw">fn </span>build_arrow_reader_options(key_metadata: <span class="prelude-ty">Option</span>&lt;<span class="kw-2">&amp;</span>[u8]&gt;) -&gt; <span class="prelude-ty">Result</span>&lt;ArrowReaderOptions&gt; {
+<a href=#667 id=667 data-nosnippet>667</a>        <span class="kw">match </span>key_metadata {
+<a href=#668 id=668 data-nosnippet>668</a>            <span class="prelude-val">Some</span>(km) =&gt; {
+<a href=#669 id=669 data-nosnippet>669</a>                <span class="kw">let </span>standard_key_metadata = StandardKeyMetadata::decode(km)<span class="question-mark">?</span>;
+<a href=#670 id=670 data-nosnippet>670</a>                <span class="kw">let </span><span class="kw-2">mut </span>builder = FileDecryptionProperties::builder(
+<a href=#671 id=671 data-nosnippet>671</a>                    standard_key_metadata.encryption_key().as_bytes().to_vec(),
+<a href=#672 id=672 data-nosnippet>672</a>                );
+<a href=#673 id=673 data-nosnippet>673</a>                <span class="kw">if let </span><span class="prelude-val">Some</span>(aad) = standard_key_metadata.aad_prefix() {
+<a href=#674 id=674 data-nosnippet>674</a>                    builder = builder.with_aad_prefix(aad.to_vec());
+<a href=#675 id=675 data-nosnippet>675</a>                }
+<a href=#676 id=676 data-nosnippet>676</a>                <span class="kw">let </span>decryption_properties = builder.build().map_err(|e| {
+<a href=#677 id=677 data-nosnippet>677</a>                    Error::new(
+<a href=#678 id=678 data-nosnippet>678</a>                        ErrorKind::Unexpected,
+<a href=#679 id=679 data-nosnippet>679</a>                        <span class="string">"Failed to build Parquet file decryption properties"</span>,
+<a href=#680 id=680 data-nosnippet>680</a>                    )
+<a href=#681 id=681 data-nosnippet>681</a>                    .with_source(e)
+<a href=#682 id=682 data-nosnippet>682</a>                })<span class="question-mark">?</span>;
+<a href=#683 id=683 data-nosnippet>683</a>                <span class="prelude-val">Ok</span>(
+<a href=#684 id=684 data-nosnippet>684</a>                    ArrowReaderOptions::new()
+<a href=#685 id=685 data-nosnippet>685</a>                        .with_file_decryption_properties(decryption_properties),
+<a href=#686 id=686 data-nosnippet>686</a>                )
+<a href=#687 id=687 data-nosnippet>687</a>            }
+<a href=#688 id=688 data-nosnippet>688</a>            <span class="prelude-val">None </span>=&gt; <span class="prelude-val">Ok</span>(ArrowReaderOptions::default()),
+<a href=#689 id=689 data-nosnippet>689</a>        }
+<a href=#690 id=690 data-nosnippet>690</a>    }
+<a href=#691 id=691 data-nosnippet>691</a>}
+<a href=#692 id=692 data-nosnippet>692</a>
+<a href=#693 id=693 data-nosnippet>693</a><span class="attr">#[cfg(test)]
+<a href=#694 id=694 data-nosnippet>694</a></span><span class="kw">mod </span>tests {
+<a href=#695 id=695 data-nosnippet>695</a>    <span class="kw">use </span>std::collections::HashMap;
+<a href=#696 id=696 data-nosnippet>696</a>    <span class="kw">use </span>std::fs::File;
+<a href=#697 id=697 data-nosnippet>697</a>    <span class="kw">use </span>std::sync::Arc;
+<a href=#698 id=698 data-nosnippet>698</a>
+<a href=#699 id=699 data-nosnippet>699</a>    <span class="kw">use </span>arrow_array::cast::AsArray;
+<a href=#700 id=700 data-nosnippet>700</a>    <span class="kw">use </span>arrow_array::{Array, ArrayRef, Int32Array, Int64Array, RecordBatch, StringArray};
+<a href=#701 id=701 data-nosnippet>701</a>    <span class="kw">use </span>arrow_cast::cast;
+<a href=#702 id=702 data-nosnippet>702</a>    <span class="kw">use </span>arrow_schema::{DataType, Field, Schema <span class="kw">as </span>ArrowSchema};
+<a href=#703 id=703 data-nosnippet>703</a>    <span class="kw">use </span>futures::TryStreamExt;
+<a href=#704 id=704 data-nosnippet>704</a>    <span class="kw">use </span>parquet::arrow::{ArrowWriter, PARQUET_FIELD_ID_META_KEY};
+<a href=#705 id=705 data-nosnippet>705</a>    <span class="kw">use </span>parquet::basic::Compression;
+<a href=#706 id=706 data-nosnippet>706</a>    <span class="kw">use </span>parquet::file::properties::WriterProperties;
+<a href=#707 id=707 data-nosnippet>707</a>    <span class="kw">use </span>tempfile::TempDir;
+<a href=#708 id=708 data-nosnippet>708</a>
+<a href=#709 id=709 data-nosnippet>709</a>    <span class="kw">use </span><span class="kw">crate</span>::Runtime;
+<a href=#710 id=710 data-nosnippet>710</a>    <span class="kw">use </span><span class="kw">crate</span>::arrow::ArrowReaderBuilder;
+<a href=#711 id=711 data-nosnippet>711</a>    <span class="kw">use </span><span class="kw">crate</span>::arrow::test_utils::write_encrypted_parquet;
+<a href=#712 id=712 data-nosnippet>712</a>    <span class="kw">use </span><span class="kw">crate</span>::io::FileIO;
+<a href=#713 id=713 data-nosnippet>713</a>    <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#714 id=714 data-nosnippet>714</a>        RESERVED_COL_NAME_POS, RESERVED_FIELD_ID_FILE, RESERVED_FIELD_ID_POS,
+<a href=#715 id=715 data-nosnippet>715</a>    };
+<a href=#716 id=716 data-nosnippet>716</a>    <span class="kw">use </span><span class="kw">crate</span>::scan::{FileScanTask, FileScanTaskStream};
+<a href=#717 id=717 data-nosnippet>717</a>    <span class="kw">use </span><span class="kw">crate</span>::spec::{DataFileFormat, NestedField, PrimitiveType, Schema, SchemaRef, Type};
+<a href=#718 id=718 data-nosnippet>718</a>
+<a href=#719 id=719 data-nosnippet>719</a>    <span class="comment">// INT96 encoding: [nanos_low_u32, nanos_high_u32, julian_day_u32]
+<a href=#720 id=720 data-nosnippet>720</a>    // Julian day 2_440_588 = Unix epoch (1970-01-01)
+<a href=#721 id=721 data-nosnippet>721</a>    </span><span class="kw">const </span>UNIX_EPOCH_JULIAN: i64 = <span class="number">2_440_588</span>;
+<a href=#722 id=722 data-nosnippet>722</a>    <span class="kw">const </span>MICROS_PER_DAY: i64 = <span class="number">86_400_000_000</span>;
+<a href=#723 id=723 data-nosnippet>723</a>    <span class="comment">// Noon on 3333-01-01 (Julian day 2_953_529) — outside the i64 nanosecond range (~1677-2262).
+<a href=#724 id=724 data-nosnippet>724</a>    </span><span class="kw">const </span>INT96_TEST_NANOS_WITHIN_DAY: u64 = <span class="number">43_200_000_000_000</span>;
+<a href=#725 id=725 data-nosnippet>725</a>    <span class="kw">const </span>INT96_TEST_JULIAN_DAY: u32 = <span class="number">2_953_529</span>;
+<a href=#726 id=726 data-nosnippet>726</a>
+<a href=#727 id=727 data-nosnippet>727</a>    <span class="kw">fn </span>make_int96_test_value() -&gt; (parquet::data_type::Int96, i64) {
+<a href=#728 id=728 data-nosnippet>728</a>        <span class="kw">let </span><span class="kw-2">mut </span>val = parquet::data_type::Int96::new();
+<a href=#729 id=729 data-nosnippet>729</a>        val.set_data(
+<a href=#730 id=730 data-nosnippet>730</a>            (INT96_TEST_NANOS_WITHIN_DAY &amp; <span class="number">0xFFFFFFFF</span>) <span class="kw">as </span>u32,
+<a href=#731 id=731 data-nosnippet>731</a>            (INT96_TEST_NANOS_WITHIN_DAY &gt;&gt; <span class="number">32</span>) <span class="kw">as </span>u32,
+<a href=#732 id=732 data-nosnippet>732</a>            INT96_TEST_JULIAN_DAY,
+<a href=#733 id=733 data-nosnippet>733</a>        );
+<a href=#734 id=734 data-nosnippet>734</a>        <span class="kw">let </span>expected_micros = (INT96_TEST_JULIAN_DAY <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY
+<a href=#735 id=735 data-nosnippet>735</a>            + (INT96_TEST_NANOS_WITHIN_DAY / <span class="number">1_000</span>) <span class="kw">as </span>i64;
+<a href=#736 id=736 data-nosnippet>736</a>        (val, expected_micros)
+<a href=#737 id=737 data-nosnippet>737</a>    }
+<a href=#738 id=738 data-nosnippet>738</a>
+<a href=#739 id=739 data-nosnippet>739</a>    <span class="kw">async fn </span>read_int96_batches(
+<a href=#740 id=740 data-nosnippet>740</a>        file_path: <span class="kw-2">&amp;</span>str,
+<a href=#741 id=741 data-nosnippet>741</a>        schema: SchemaRef,
+<a href=#742 id=742 data-nosnippet>742</a>        project_field_ids: Vec&lt;i32&gt;,
+<a href=#743 id=743 data-nosnippet>743</a>    ) -&gt; Vec&lt;RecordBatch&gt; {
+<a href=#744 id=744 data-nosnippet>744</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#745 id=745 data-nosnippet>745</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
 <a href=#746 id=746 data-nosnippet>746</a>
-<a href=#747 id=747 data-nosnippet>747</a>    <span class="comment">// ArrowWriter cannot write INT96, so we use SerializedFileWriter directly.
-<a href=#748 id=748 data-nosnippet>748</a>    </span><span class="kw">fn </span>write_int96_parquet_file(
-<a href=#749 id=749 data-nosnippet>749</a>        table_location: <span class="kw-2">&amp;</span>str,
-<a href=#750 id=750 data-nosnippet>750</a>        filename: <span class="kw-2">&amp;</span>str,
-<a href=#751 id=751 data-nosnippet>751</a>        with_field_ids: bool,
-<a href=#752 id=752 data-nosnippet>752</a>    ) -&gt; (String, Vec&lt;i64&gt;) {
-<a href=#753 id=753 data-nosnippet>753</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
-<a href=#754 id=754 data-nosnippet>754</a>        <span class="kw">use </span>parquet::data_type::{Int32Type, Int96, Int96Type};
-<a href=#755 id=755 data-nosnippet>755</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
-<a href=#756 id=756 data-nosnippet>756</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
-<a href=#757 id=757 data-nosnippet>757</a>
-<a href=#758 id=758 data-nosnippet>758</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/{filename}"</span>);
-<a href=#759 id=759 data-nosnippet>759</a>
-<a href=#760 id=760 data-nosnippet>760</a>        <span class="kw">let </span><span class="kw-2">mut </span>ts_builder = SchemaType::primitive_type_builder(<span class="string">"ts"</span>, PhysicalType::INT96)
-<a href=#761 id=761 data-nosnippet>761</a>            .with_repetition(Repetition::OPTIONAL);
-<a href=#762 id=762 data-nosnippet>762</a>        <span class="kw">let </span><span class="kw-2">mut </span>id_builder = SchemaType::primitive_type_builder(<span class="string">"id"</span>, PhysicalType::INT32)
-<a href=#763 id=763 data-nosnippet>763</a>            .with_repetition(Repetition::REQUIRED);
-<a href=#764 id=764 data-nosnippet>764</a>
-<a href=#765 id=765 data-nosnippet>765</a>        <span class="kw">if </span>with_field_ids {
-<a href=#766 id=766 data-nosnippet>766</a>            ts_builder = ts_builder.with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>));
-<a href=#767 id=767 data-nosnippet>767</a>            id_builder = id_builder.with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>));
-<a href=#768 id=768 data-nosnippet>768</a>        }
-<a href=#769 id=769 data-nosnippet>769</a>
-<a href=#770 id=770 data-nosnippet>770</a>        <span class="kw">let </span>schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
-<a href=#771 id=771 data-nosnippet>771</a>            .with_fields(<span class="macro">vec!</span>[
-<a href=#772 id=772 data-nosnippet>772</a>                Arc::new(ts_builder.build().unwrap()),
-<a href=#773 id=773 data-nosnippet>773</a>                Arc::new(id_builder.build().unwrap()),
-<a href=#774 id=774 data-nosnippet>774</a>            ])
-<a href=#775 id=775 data-nosnippet>775</a>            .build()
-<a href=#776 id=776 data-nosnippet>776</a>            .unwrap();
-<a href=#777 id=777 data-nosnippet>777</a>
-<a href=#778 id=778 data-nosnippet>778</a>        <span class="comment">// Dates outside the i64 nanosecond range (~1677-2262) overflow without coercion.
-<a href=#779 id=779 data-nosnippet>779</a>        </span><span class="kw">const </span>NOON_NANOS: u64 = INT96_TEST_NANOS_WITHIN_DAY;
-<a href=#780 id=780 data-nosnippet>780</a>        <span class="kw">const </span>JULIAN_3333: u32 = INT96_TEST_JULIAN_DAY;
-<a href=#781 id=781 data-nosnippet>781</a>        <span class="kw">const </span>JULIAN_2100: u32 = <span class="number">2_488_070</span>;
-<a href=#782 id=782 data-nosnippet>782</a>
-<a href=#783 id=783 data-nosnippet>783</a>        <span class="kw">let </span>test_data: Vec&lt;(u32, u32, u32, i64)&gt; = <span class="macro">vec!</span>[
-<a href=#784 id=784 data-nosnippet>784</a>            <span class="comment">// 3333-01-01 00:00:00
-<a href=#785 id=785 data-nosnippet>785</a>            </span>(
-<a href=#786 id=786 data-nosnippet>786</a>                <span class="number">0</span>,
-<a href=#787 id=787 data-nosnippet>787</a>                <span class="number">0</span>,
-<a href=#788 id=788 data-nosnippet>788</a>                JULIAN_3333,
-<a href=#789 id=789 data-nosnippet>789</a>                (JULIAN_3333 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY,
-<a href=#790 id=790 data-nosnippet>790</a>            ),
-<a href=#791 id=791 data-nosnippet>791</a>            <span class="comment">// 3333-01-01 12:00:00
-<a href=#792 id=792 data-nosnippet>792</a>            </span>(
-<a href=#793 id=793 data-nosnippet>793</a>                (NOON_NANOS &amp; <span class="number">0xFFFFFFFF</span>) <span class="kw">as </span>u32,
-<a href=#794 id=794 data-nosnippet>794</a>                (NOON_NANOS &gt;&gt; <span class="number">32</span>) <span class="kw">as </span>u32,
-<a href=#795 id=795 data-nosnippet>795</a>                JULIAN_3333,
-<a href=#796 id=796 data-nosnippet>796</a>                (JULIAN_3333 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY
-<a href=#797 id=797 data-nosnippet>797</a>                    + (NOON_NANOS / <span class="number">1_000</span>) <span class="kw">as </span>i64,
-<a href=#798 id=798 data-nosnippet>798</a>            ),
-<a href=#799 id=799 data-nosnippet>799</a>            <span class="comment">// 2100-01-01 00:00:00
-<a href=#800 id=800 data-nosnippet>800</a>            </span>(
-<a href=#801 id=801 data-nosnippet>801</a>                <span class="number">0</span>,
-<a href=#802 id=802 data-nosnippet>802</a>                <span class="number">0</span>,
-<a href=#803 id=803 data-nosnippet>803</a>                JULIAN_2100,
-<a href=#804 id=804 data-nosnippet>804</a>                (JULIAN_2100 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY,
-<a href=#805 id=805 data-nosnippet>805</a>            ),
-<a href=#806 id=806 data-nosnippet>806</a>        ];
-<a href=#807 id=807 data-nosnippet>807</a>
-<a href=#808 id=808 data-nosnippet>808</a>        <span class="kw">let </span>int96_values: Vec&lt;Int96&gt; = test_data
-<a href=#809 id=809 data-nosnippet>809</a>            .iter()
-<a href=#810 id=810 data-nosnippet>810</a>            .map(|(lo, hi, day, <span class="kw">_</span>)| {
-<a href=#811 id=811 data-nosnippet>811</a>                <span class="kw">let </span><span class="kw-2">mut </span>v = Int96::new();
-<a href=#812 id=812 data-nosnippet>812</a>                v.set_data(<span class="kw-2">*</span>lo, <span class="kw-2">*</span>hi, <span class="kw-2">*</span>day);
-<a href=#813 id=813 data-nosnippet>813</a>                v
-<a href=#814 id=814 data-nosnippet>814</a>            })
-<a href=#815 id=815 data-nosnippet>815</a>            .collect();
-<a href=#816 id=816 data-nosnippet>816</a>
-<a href=#817 id=817 data-nosnippet>817</a>        <span class="kw">let </span>id_values: Vec&lt;i32&gt; = (<span class="number">0</span>..test_data.len() <span class="kw">as </span>i32).collect();
-<a href=#818 id=818 data-nosnippet>818</a>        <span class="kw">let </span>expected_micros: Vec&lt;i64&gt; = test_data.iter().map(|(<span class="kw">_</span>, <span class="kw">_</span>, <span class="kw">_</span>, m)| <span class="kw-2">*</span>m).collect();
-<a href=#819 id=819 data-nosnippet>819</a>
-<a href=#820 id=820 data-nosnippet>820</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
-<a href=#821 id=821 data-nosnippet>821</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#822 id=822 data-nosnippet>822</a>            SerializedFileWriter::new(file, Arc::new(schema), Default::default()).unwrap();
-<a href=#823 id=823 data-nosnippet>823</a>
-<a href=#824 id=824 data-nosnippet>824</a>        <span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
-<a href=#825 id=825 data-nosnippet>825</a>        {
-<a href=#826 id=826 data-nosnippet>826</a>            <span class="comment">// def=1: ts is OPTIONAL and present. No repetition levels (top-level columns).
-<a href=#827 id=827 data-nosnippet>827</a>            </span><span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#828 id=828 data-nosnippet>828</a>            col.typed::&lt;Int96Type&gt;()
-<a href=#829 id=829 data-nosnippet>829</a>                .write_batch(<span class="kw-2">&amp;</span>int96_values, <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span><span class="macro">vec!</span>[<span class="number">1</span>; test_data.len()]), <span class="prelude-val">None</span>)
-<a href=#830 id=830 data-nosnippet>830</a>                .unwrap();
-<a href=#831 id=831 data-nosnippet>831</a>            col.close().unwrap();
-<a href=#832 id=832 data-nosnippet>832</a>        }
-<a href=#833 id=833 data-nosnippet>833</a>        {
-<a href=#834 id=834 data-nosnippet>834</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#835 id=835 data-nosnippet>835</a>            col.typed::&lt;Int32Type&gt;()
-<a href=#836 id=836 data-nosnippet>836</a>                .write_batch(<span class="kw-2">&amp;</span>id_values, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>)
-<a href=#837 id=837 data-nosnippet>837</a>                .unwrap();
-<a href=#838 id=838 data-nosnippet>838</a>            col.close().unwrap();
-<a href=#839 id=839 data-nosnippet>839</a>        }
-<a href=#840 id=840 data-nosnippet>840</a>        row_group.close().unwrap();
-<a href=#841 id=841 data-nosnippet>841</a>        writer.close().unwrap();
-<a href=#842 id=842 data-nosnippet>842</a>
-<a href=#843 id=843 data-nosnippet>843</a>        (file_path, expected_micros)
-<a href=#844 id=844 data-nosnippet>844</a>    }
+<a href=#747 id=747 data-nosnippet>747</a>        <span class="kw">let </span>file_size = std::fs::metadata(file_path).unwrap().len();
+<a href=#748 id=748 data-nosnippet>748</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#749 id=749 data-nosnippet>749</a>            .with_file_size_in_bytes(file_size)
+<a href=#750 id=750 data-nosnippet>750</a>            .with_start(<span class="number">0</span>)
+<a href=#751 id=751 data-nosnippet>751</a>            .with_length(file_size)
+<a href=#752 id=752 data-nosnippet>752</a>            .with_data_file_path(file_path.to_string())
+<a href=#753 id=753 data-nosnippet>753</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#754 id=754 data-nosnippet>754</a>            .with_schema(schema)
+<a href=#755 id=755 data-nosnippet>755</a>            .with_project_field_ids(project_field_ids)
+<a href=#756 id=756 data-nosnippet>756</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#757 id=757 data-nosnippet>757</a>            .build();
+<a href=#758 id=758 data-nosnippet>758</a>
+<a href=#759 id=759 data-nosnippet>759</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#760 id=760 data-nosnippet>760</a>        reader
+<a href=#761 id=761 data-nosnippet>761</a>            .read(tasks)
+<a href=#762 id=762 data-nosnippet>762</a>            .unwrap()
+<a href=#763 id=763 data-nosnippet>763</a>            .stream()
+<a href=#764 id=764 data-nosnippet>764</a>            .try_collect()
+<a href=#765 id=765 data-nosnippet>765</a>            .<span class="kw">await
+<a href=#766 id=766 data-nosnippet>766</a>            </span>.unwrap()
+<a href=#767 id=767 data-nosnippet>767</a>    }
+<a href=#768 id=768 data-nosnippet>768</a>
+<a href=#769 id=769 data-nosnippet>769</a>    <span class="comment">// ArrowWriter cannot write INT96, so we use SerializedFileWriter directly.
+<a href=#770 id=770 data-nosnippet>770</a>    </span><span class="kw">fn </span>write_int96_parquet_file(
+<a href=#771 id=771 data-nosnippet>771</a>        table_location: <span class="kw-2">&amp;</span>str,
+<a href=#772 id=772 data-nosnippet>772</a>        filename: <span class="kw-2">&amp;</span>str,
+<a href=#773 id=773 data-nosnippet>773</a>        with_field_ids: bool,
+<a href=#774 id=774 data-nosnippet>774</a>    ) -&gt; (String, Vec&lt;i64&gt;) {
+<a href=#775 id=775 data-nosnippet>775</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
+<a href=#776 id=776 data-nosnippet>776</a>        <span class="kw">use </span>parquet::data_type::{Int32Type, Int96, Int96Type};
+<a href=#777 id=777 data-nosnippet>777</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
+<a href=#778 id=778 data-nosnippet>778</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
+<a href=#779 id=779 data-nosnippet>779</a>
+<a href=#780 id=780 data-nosnippet>780</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/{filename}"</span>);
+<a href=#781 id=781 data-nosnippet>781</a>
+<a href=#782 id=782 data-nosnippet>782</a>        <span class="kw">let </span><span class="kw-2">mut </span>ts_builder = SchemaType::primitive_type_builder(<span class="string">"ts"</span>, PhysicalType::INT96)
+<a href=#783 id=783 data-nosnippet>783</a>            .with_repetition(Repetition::OPTIONAL);
+<a href=#784 id=784 data-nosnippet>784</a>        <span class="kw">let </span><span class="kw-2">mut </span>id_builder = SchemaType::primitive_type_builder(<span class="string">"id"</span>, PhysicalType::INT32)
+<a href=#785 id=785 data-nosnippet>785</a>            .with_repetition(Repetition::REQUIRED);
+<a href=#786 id=786 data-nosnippet>786</a>
+<a href=#787 id=787 data-nosnippet>787</a>        <span class="kw">if </span>with_field_ids {
+<a href=#788 id=788 data-nosnippet>788</a>            ts_builder = ts_builder.with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>));
+<a href=#789 id=789 data-nosnippet>789</a>            id_builder = id_builder.with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>));
+<a href=#790 id=790 data-nosnippet>790</a>        }
+<a href=#791 id=791 data-nosnippet>791</a>
+<a href=#792 id=792 data-nosnippet>792</a>        <span class="kw">let </span>schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
+<a href=#793 id=793 data-nosnippet>793</a>            .with_fields(<span class="macro">vec!</span>[
+<a href=#794 id=794 data-nosnippet>794</a>                Arc::new(ts_builder.build().unwrap()),
+<a href=#795 id=795 data-nosnippet>795</a>                Arc::new(id_builder.build().unwrap()),
+<a href=#796 id=796 data-nosnippet>796</a>            ])
+<a href=#797 id=797 data-nosnippet>797</a>            .build()
+<a href=#798 id=798 data-nosnippet>798</a>            .unwrap();
+<a href=#799 id=799 data-nosnippet>799</a>
+<a href=#800 id=800 data-nosnippet>800</a>        <span class="comment">// Dates outside the i64 nanosecond range (~1677-2262) overflow without coercion.
+<a href=#801 id=801 data-nosnippet>801</a>        </span><span class="kw">const </span>NOON_NANOS: u64 = INT96_TEST_NANOS_WITHIN_DAY;
+<a href=#802 id=802 data-nosnippet>802</a>        <span class="kw">const </span>JULIAN_3333: u32 = INT96_TEST_JULIAN_DAY;
+<a href=#803 id=803 data-nosnippet>803</a>        <span class="kw">const </span>JULIAN_2100: u32 = <span class="number">2_488_070</span>;
+<a href=#804 id=804 data-nosnippet>804</a>
+<a href=#805 id=805 data-nosnippet>805</a>        <span class="kw">let </span>test_data: Vec&lt;(u32, u32, u32, i64)&gt; = <span class="macro">vec!</span>[
+<a href=#806 id=806 data-nosnippet>806</a>            <span class="comment">// 3333-01-01 00:00:00
+<a href=#807 id=807 data-nosnippet>807</a>            </span>(
+<a href=#808 id=808 data-nosnippet>808</a>                <span class="number">0</span>,
+<a href=#809 id=809 data-nosnippet>809</a>                <span class="number">0</span>,
+<a href=#810 id=810 data-nosnippet>810</a>                JULIAN_3333,
+<a href=#811 id=811 data-nosnippet>811</a>                (JULIAN_3333 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY,
+<a href=#812 id=812 data-nosnippet>812</a>            ),
+<a href=#813 id=813 data-nosnippet>813</a>            <span class="comment">// 3333-01-01 12:00:00
+<a href=#814 id=814 data-nosnippet>814</a>            </span>(
+<a href=#815 id=815 data-nosnippet>815</a>                (NOON_NANOS &amp; <span class="number">0xFFFFFFFF</span>) <span class="kw">as </span>u32,
+<a href=#816 id=816 data-nosnippet>816</a>                (NOON_NANOS &gt;&gt; <span class="number">32</span>) <span class="kw">as </span>u32,
+<a href=#817 id=817 data-nosnippet>817</a>                JULIAN_3333,
+<a href=#818 id=818 data-nosnippet>818</a>                (JULIAN_3333 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY
+<a href=#819 id=819 data-nosnippet>819</a>                    + (NOON_NANOS / <span class="number">1_000</span>) <span class="kw">as </span>i64,
+<a href=#820 id=820 data-nosnippet>820</a>            ),
+<a href=#821 id=821 data-nosnippet>821</a>            <span class="comment">// 2100-01-01 00:00:00
+<a href=#822 id=822 data-nosnippet>822</a>            </span>(
+<a href=#823 id=823 data-nosnippet>823</a>                <span class="number">0</span>,
+<a href=#824 id=824 data-nosnippet>824</a>                <span class="number">0</span>,
+<a href=#825 id=825 data-nosnippet>825</a>                JULIAN_2100,
+<a href=#826 id=826 data-nosnippet>826</a>                (JULIAN_2100 <span class="kw">as </span>i64 - UNIX_EPOCH_JULIAN) * MICROS_PER_DAY,
+<a href=#827 id=827 data-nosnippet>827</a>            ),
+<a href=#828 id=828 data-nosnippet>828</a>        ];
+<a href=#829 id=829 data-nosnippet>829</a>
+<a href=#830 id=830 data-nosnippet>830</a>        <span class="kw">let </span>int96_values: Vec&lt;Int96&gt; = test_data
+<a href=#831 id=831 data-nosnippet>831</a>            .iter()
+<a href=#832 id=832 data-nosnippet>832</a>            .map(|(lo, hi, day, <span class="kw">_</span>)| {
+<a href=#833 id=833 data-nosnippet>833</a>                <span class="kw">let </span><span class="kw-2">mut </span>v = Int96::new();
+<a href=#834 id=834 data-nosnippet>834</a>                v.set_data(<span class="kw-2">*</span>lo, <span class="kw-2">*</span>hi, <span class="kw-2">*</span>day);
+<a href=#835 id=835 data-nosnippet>835</a>                v
+<a href=#836 id=836 data-nosnippet>836</a>            })
+<a href=#837 id=837 data-nosnippet>837</a>            .collect();
+<a href=#838 id=838 data-nosnippet>838</a>
+<a href=#839 id=839 data-nosnippet>839</a>        <span class="kw">let </span>id_values: Vec&lt;i32&gt; = (<span class="number">0</span>..test_data.len() <span class="kw">as </span>i32).collect();
+<a href=#840 id=840 data-nosnippet>840</a>        <span class="kw">let </span>expected_micros: Vec&lt;i64&gt; = test_data.iter().map(|(<span class="kw">_</span>, <span class="kw">_</span>, <span class="kw">_</span>, m)| <span class="kw-2">*</span>m).collect();
+<a href=#841 id=841 data-nosnippet>841</a>
+<a href=#842 id=842 data-nosnippet>842</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
+<a href=#843 id=843 data-nosnippet>843</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#844 id=844 data-nosnippet>844</a>            SerializedFileWriter::new(file, Arc::new(schema), Default::default()).unwrap();
 <a href=#845 id=845 data-nosnippet>845</a>
-<a href=#846 id=846 data-nosnippet>846</a>    <span class="kw">async fn </span>assert_int96_read_matches(
-<a href=#847 id=847 data-nosnippet>847</a>        file_path: <span class="kw-2">&amp;</span>str,
-<a href=#848 id=848 data-nosnippet>848</a>        schema: SchemaRef,
-<a href=#849 id=849 data-nosnippet>849</a>        project_field_ids: Vec&lt;i32&gt;,
-<a href=#850 id=850 data-nosnippet>850</a>        expected_micros: <span class="kw-2">&amp;</span>[i64],
-<a href=#851 id=851 data-nosnippet>851</a>    ) {
-<a href=#852 id=852 data-nosnippet>852</a>        <span class="kw">use </span>arrow_array::TimestampMicrosecondArray;
-<a href=#853 id=853 data-nosnippet>853</a>
-<a href=#854 id=854 data-nosnippet>854</a>        <span class="kw">let </span>batches = read_int96_batches(file_path, schema, project_field_ids).<span class="kw">await</span>;
-<a href=#855 id=855 data-nosnippet>855</a>
-<a href=#856 id=856 data-nosnippet>856</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
-<a href=#857 id=857 data-nosnippet>857</a>        <span class="kw">let </span>ts_array = batches[<span class="number">0</span>]
-<a href=#858 id=858 data-nosnippet>858</a>            .column(<span class="number">0</span>)
-<a href=#859 id=859 data-nosnippet>859</a>            .as_any()
-<a href=#860 id=860 data-nosnippet>860</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
-<a href=#861 id=861 data-nosnippet>861</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray"</span>);
-<a href=#862 id=862 data-nosnippet>862</a>
-<a href=#863 id=863 data-nosnippet>863</a>        <span class="kw">for </span>(i, expected) <span class="kw">in </span>expected_micros.iter().enumerate() {
-<a href=#864 id=864 data-nosnippet>864</a>            <span class="macro">assert_eq!</span>(
-<a href=#865 id=865 data-nosnippet>865</a>                ts_array.value(i),
-<a href=#866 id=866 data-nosnippet>866</a>                <span class="kw-2">*</span>expected,
-<a href=#867 id=867 data-nosnippet>867</a>                <span class="string">"Row {i}: got {}, expected {expected}"</span>,
-<a href=#868 id=868 data-nosnippet>868</a>                ts_array.value(i)
-<a href=#869 id=869 data-nosnippet>869</a>            );
-<a href=#870 id=870 data-nosnippet>870</a>        }
-<a href=#871 id=871 data-nosnippet>871</a>    }
-<a href=#872 id=872 data-nosnippet>872</a>
-<a href=#873 id=873 data-nosnippet>873</a>    <span class="doccomment">/// Writes a single-column Parquet file encrypted with `encryption_key`, then reads it
-<a href=#874 id=874 data-nosnippet>874</a>    /// back through `ArrowReader` and asserts the round-tripped values. The key length
-<a href=#875 id=875 data-nosnippet>875</a>    /// selects the AES-GCM variant in arrow-rs (16 -&gt; AES-128, 32 -&gt; AES-256).
-<a href=#876 id=876 data-nosnippet>876</a>    </span><span class="kw">async fn </span>assert_encrypted_parquet_roundtrip(encryption_key: <span class="kw-2">&amp;</span>[u8]) {
-<a href=#877 id=877 data-nosnippet>877</a>        <span class="kw">let </span>aad_prefix = <span class="string">b"aad_prefix"</span>;
-<a href=#878 id=878 data-nosnippet>878</a>
-<a href=#879 id=879 data-nosnippet>879</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#880 id=880 data-nosnippet>880</a>            Schema::builder()
-<a href=#881 id=881 data-nosnippet>881</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#882 id=882 data-nosnippet>882</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#883 id=883 data-nosnippet>883</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#884 id=884 data-nosnippet>884</a>                ])
-<a href=#885 id=885 data-nosnippet>885</a>                .build()
-<a href=#886 id=886 data-nosnippet>886</a>                .unwrap(),
-<a href=#887 id=887 data-nosnippet>887</a>        );
-<a href=#888 id=888 data-nosnippet>888</a>
-<a href=#889 id=889 data-nosnippet>889</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#890 id=890 data-nosnippet>890</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#891 id=891 data-nosnippet>891</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#892 id=892 data-nosnippet>892</a>                <span class="string">"1"</span>.to_string(),
-<a href=#893 id=893 data-nosnippet>893</a>            )])),
-<a href=#894 id=894 data-nosnippet>894</a>        ]));
-<a href=#895 id=895 data-nosnippet>895</a>
-<a href=#896 id=896 data-nosnippet>896</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#897 id=897 data-nosnippet>897</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#898 id=898 data-nosnippet>898</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#899 id=899 data-nosnippet>899</a>
-<a href=#900 id=900 data-nosnippet>900</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">10</span>, <span class="number">20</span>, <span class="number">30</span>])) <span class="kw">as </span>ArrayRef;
-<a href=#901 id=901 data-nosnippet>901</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
-<a href=#902 id=902 data-nosnippet>902</a>
-<a href=#903 id=903 data-nosnippet>903</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted.parquet"</span>);
-<a href=#904 id=904 data-nosnippet>904</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">Some</span>(aad_prefix));
-<a href=#905 id=905 data-nosnippet>905</a>
-<a href=#906 id=906 data-nosnippet>906</a>        <span class="kw">let </span>key_metadata = <span class="kw">crate</span>::encryption::StandardKeyMetadata::try_new(encryption_key)
-<a href=#907 id=907 data-nosnippet>907</a>            .unwrap()
-<a href=#908 id=908 data-nosnippet>908</a>            .with_aad_prefix(aad_prefix)
-<a href=#909 id=909 data-nosnippet>909</a>            .encode()
-<a href=#910 id=910 data-nosnippet>910</a>            .unwrap();
-<a href=#911 id=911 data-nosnippet>911</a>
-<a href=#912 id=912 data-nosnippet>912</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#913 id=913 data-nosnippet>913</a>
-<a href=#914 id=914 data-nosnippet>914</a>        <span class="kw">let </span>task = FileScanTask::builder()
-<a href=#915 id=915 data-nosnippet>915</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
-<a href=#916 id=916 data-nosnippet>916</a>            .with_start(<span class="number">0</span>)
-<a href=#917 id=917 data-nosnippet>917</a>            .with_length(<span class="number">0</span>)
-<a href=#918 id=918 data-nosnippet>918</a>            .with_data_file_path(file_path)
-<a href=#919 id=919 data-nosnippet>919</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#920 id=920 data-nosnippet>920</a>            .with_schema(schema)
-<a href=#921 id=921 data-nosnippet>921</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#922 id=922 data-nosnippet>922</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#923 id=923 data-nosnippet>923</a>            .with_key_metadata(<span class="prelude-val">Some</span>(key_metadata))
-<a href=#924 id=924 data-nosnippet>924</a>            .build();
-<a href=#925 id=925 data-nosnippet>925</a>
-<a href=#926 id=926 data-nosnippet>926</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#927 id=927 data-nosnippet>927</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#928 id=928 data-nosnippet>928</a>            .read(tasks)
+<a href=#846 id=846 data-nosnippet>846</a>        <span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
+<a href=#847 id=847 data-nosnippet>847</a>        {
+<a href=#848 id=848 data-nosnippet>848</a>            <span class="comment">// def=1: ts is OPTIONAL and present. No repetition levels (top-level columns).
+<a href=#849 id=849 data-nosnippet>849</a>            </span><span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#850 id=850 data-nosnippet>850</a>            col.typed::&lt;Int96Type&gt;()
+<a href=#851 id=851 data-nosnippet>851</a>                .write_batch(<span class="kw-2">&amp;</span>int96_values, <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span><span class="macro">vec!</span>[<span class="number">1</span>; test_data.len()]), <span class="prelude-val">None</span>)
+<a href=#852 id=852 data-nosnippet>852</a>                .unwrap();
+<a href=#853 id=853 data-nosnippet>853</a>            col.close().unwrap();
+<a href=#854 id=854 data-nosnippet>854</a>        }
+<a href=#855 id=855 data-nosnippet>855</a>        {
+<a href=#856 id=856 data-nosnippet>856</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#857 id=857 data-nosnippet>857</a>            col.typed::&lt;Int32Type&gt;()
+<a href=#858 id=858 data-nosnippet>858</a>                .write_batch(<span class="kw-2">&amp;</span>id_values, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>)
+<a href=#859 id=859 data-nosnippet>859</a>                .unwrap();
+<a href=#860 id=860 data-nosnippet>860</a>            col.close().unwrap();
+<a href=#861 id=861 data-nosnippet>861</a>        }
+<a href=#862 id=862 data-nosnippet>862</a>        row_group.close().unwrap();
+<a href=#863 id=863 data-nosnippet>863</a>        writer.close().unwrap();
+<a href=#864 id=864 data-nosnippet>864</a>
+<a href=#865 id=865 data-nosnippet>865</a>        (file_path, expected_micros)
+<a href=#866 id=866 data-nosnippet>866</a>    }
+<a href=#867 id=867 data-nosnippet>867</a>
+<a href=#868 id=868 data-nosnippet>868</a>    <span class="kw">async fn </span>assert_int96_read_matches(
+<a href=#869 id=869 data-nosnippet>869</a>        file_path: <span class="kw-2">&amp;</span>str,
+<a href=#870 id=870 data-nosnippet>870</a>        schema: SchemaRef,
+<a href=#871 id=871 data-nosnippet>871</a>        project_field_ids: Vec&lt;i32&gt;,
+<a href=#872 id=872 data-nosnippet>872</a>        expected_micros: <span class="kw-2">&amp;</span>[i64],
+<a href=#873 id=873 data-nosnippet>873</a>    ) {
+<a href=#874 id=874 data-nosnippet>874</a>        <span class="kw">use </span>arrow_array::TimestampMicrosecondArray;
+<a href=#875 id=875 data-nosnippet>875</a>
+<a href=#876 id=876 data-nosnippet>876</a>        <span class="kw">let </span>batches = read_int96_batches(file_path, schema, project_field_ids).<span class="kw">await</span>;
+<a href=#877 id=877 data-nosnippet>877</a>
+<a href=#878 id=878 data-nosnippet>878</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
+<a href=#879 id=879 data-nosnippet>879</a>        <span class="kw">let </span>ts_array = batches[<span class="number">0</span>]
+<a href=#880 id=880 data-nosnippet>880</a>            .column(<span class="number">0</span>)
+<a href=#881 id=881 data-nosnippet>881</a>            .as_any()
+<a href=#882 id=882 data-nosnippet>882</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
+<a href=#883 id=883 data-nosnippet>883</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray"</span>);
+<a href=#884 id=884 data-nosnippet>884</a>
+<a href=#885 id=885 data-nosnippet>885</a>        <span class="kw">for </span>(i, expected) <span class="kw">in </span>expected_micros.iter().enumerate() {
+<a href=#886 id=886 data-nosnippet>886</a>            <span class="macro">assert_eq!</span>(
+<a href=#887 id=887 data-nosnippet>887</a>                ts_array.value(i),
+<a href=#888 id=888 data-nosnippet>888</a>                <span class="kw-2">*</span>expected,
+<a href=#889 id=889 data-nosnippet>889</a>                <span class="string">"Row {i}: got {}, expected {expected}"</span>,
+<a href=#890 id=890 data-nosnippet>890</a>                ts_array.value(i)
+<a href=#891 id=891 data-nosnippet>891</a>            );
+<a href=#892 id=892 data-nosnippet>892</a>        }
+<a href=#893 id=893 data-nosnippet>893</a>    }
+<a href=#894 id=894 data-nosnippet>894</a>
+<a href=#895 id=895 data-nosnippet>895</a>    <span class="doccomment">/// Writes a single-column Parquet file encrypted with `encryption_key`, then reads it
+<a href=#896 id=896 data-nosnippet>896</a>    /// back through `ArrowReader` and asserts the round-tripped values. The key length
+<a href=#897 id=897 data-nosnippet>897</a>    /// selects the AES-GCM variant in arrow-rs (16 -&gt; AES-128, 32 -&gt; AES-256).
+<a href=#898 id=898 data-nosnippet>898</a>    </span><span class="kw">async fn </span>assert_encrypted_parquet_roundtrip(encryption_key: <span class="kw-2">&amp;</span>[u8]) {
+<a href=#899 id=899 data-nosnippet>899</a>        <span class="kw">let </span>aad_prefix = <span class="string">b"aad_prefix"</span>;
+<a href=#900 id=900 data-nosnippet>900</a>
+<a href=#901 id=901 data-nosnippet>901</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#902 id=902 data-nosnippet>902</a>            Schema::builder()
+<a href=#903 id=903 data-nosnippet>903</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#904 id=904 data-nosnippet>904</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#905 id=905 data-nosnippet>905</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#906 id=906 data-nosnippet>906</a>                ])
+<a href=#907 id=907 data-nosnippet>907</a>                .build()
+<a href=#908 id=908 data-nosnippet>908</a>                .unwrap(),
+<a href=#909 id=909 data-nosnippet>909</a>        );
+<a href=#910 id=910 data-nosnippet>910</a>
+<a href=#911 id=911 data-nosnippet>911</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#912 id=912 data-nosnippet>912</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#913 id=913 data-nosnippet>913</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#914 id=914 data-nosnippet>914</a>                <span class="string">"1"</span>.to_string(),
+<a href=#915 id=915 data-nosnippet>915</a>            )])),
+<a href=#916 id=916 data-nosnippet>916</a>        ]));
+<a href=#917 id=917 data-nosnippet>917</a>
+<a href=#918 id=918 data-nosnippet>918</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#919 id=919 data-nosnippet>919</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#920 id=920 data-nosnippet>920</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#921 id=921 data-nosnippet>921</a>
+<a href=#922 id=922 data-nosnippet>922</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">10</span>, <span class="number">20</span>, <span class="number">30</span>])) <span class="kw">as </span>ArrayRef;
+<a href=#923 id=923 data-nosnippet>923</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
+<a href=#924 id=924 data-nosnippet>924</a>
+<a href=#925 id=925 data-nosnippet>925</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted.parquet"</span>);
+<a href=#926 id=926 data-nosnippet>926</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">Some</span>(aad_prefix));
+<a href=#927 id=927 data-nosnippet>927</a>
+<a href=#928 id=928 data-nosnippet>928</a>        <span class="kw">let </span>key_metadata = <span class="kw">crate</span>::encryption::StandardKeyMetadata::try_new(encryption_key)
 <a href=#929 id=929 data-nosnippet>929</a>            .unwrap()
-<a href=#930 id=930 data-nosnippet>930</a>            .stream()
-<a href=#931 id=931 data-nosnippet>931</a>            .try_collect()
-<a href=#932 id=932 data-nosnippet>932</a>            .<span class="kw">await
-<a href=#933 id=933 data-nosnippet>933</a>            </span>.unwrap();
-<a href=#934 id=934 data-nosnippet>934</a>
-<a href=#935 id=935 data-nosnippet>935</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
-<a href=#936 id=936 data-nosnippet>936</a>        <span class="kw">let </span>ids = batches[<span class="number">0</span>]
-<a href=#937 id=937 data-nosnippet>937</a>            .column(<span class="number">0</span>)
-<a href=#938 id=938 data-nosnippet>938</a>            .as_any()
-<a href=#939 id=939 data-nosnippet>939</a>            .downcast_ref::&lt;Int32Array&gt;()
-<a href=#940 id=940 data-nosnippet>940</a>            .unwrap();
-<a href=#941 id=941 data-nosnippet>941</a>        <span class="macro">assert_eq!</span>(ids.values(), <span class="kw-2">&amp;</span>[<span class="number">10</span>, <span class="number">20</span>, <span class="number">30</span>]);
-<a href=#942 id=942 data-nosnippet>942</a>    }
-<a href=#943 id=943 data-nosnippet>943</a>
-<a href=#944 id=944 data-nosnippet>944</a>    <span class="attr">#[tokio::test]
-<a href=#945 id=945 data-nosnippet>945</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_aes_128() {
-<a href=#946 id=946 data-nosnippet>946</a>        assert_encrypted_parquet_roundtrip(<span class="string">b"0123456789abcdef"</span>).<span class="kw">await</span>;
-<a href=#947 id=947 data-nosnippet>947</a>    }
-<a href=#948 id=948 data-nosnippet>948</a>
-<a href=#949 id=949 data-nosnippet>949</a>    <span class="attr">#[tokio::test]
-<a href=#950 id=950 data-nosnippet>950</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_aes_256() {
-<a href=#951 id=951 data-nosnippet>951</a>        assert_encrypted_parquet_roundtrip(<span class="string">b"0123456789abcdef0123456789abcdef"</span>).<span class="kw">await</span>;
-<a href=#952 id=952 data-nosnippet>952</a>    }
-<a href=#953 id=953 data-nosnippet>953</a>
-<a href=#954 id=954 data-nosnippet>954</a>    <span class="attr">#[tokio::test]
-<a href=#955 id=955 data-nosnippet>955</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_without_key_metadata_fails() {
-<a href=#956 id=956 data-nosnippet>956</a>        <span class="kw">let </span>encryption_key = <span class="string">b"0123456789abcdef"</span>;
-<a href=#957 id=957 data-nosnippet>957</a>
-<a href=#958 id=958 data-nosnippet>958</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#959 id=959 data-nosnippet>959</a>            Schema::builder()
-<a href=#960 id=960 data-nosnippet>960</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#961 id=961 data-nosnippet>961</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#962 id=962 data-nosnippet>962</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#963 id=963 data-nosnippet>963</a>                ])
-<a href=#964 id=964 data-nosnippet>964</a>                .build()
-<a href=#965 id=965 data-nosnippet>965</a>                .unwrap(),
-<a href=#966 id=966 data-nosnippet>966</a>        );
-<a href=#967 id=967 data-nosnippet>967</a>
-<a href=#968 id=968 data-nosnippet>968</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#969 id=969 data-nosnippet>969</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#970 id=970 data-nosnippet>970</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#971 id=971 data-nosnippet>971</a>                <span class="string">"1"</span>.to_string(),
-<a href=#972 id=972 data-nosnippet>972</a>            )])),
-<a href=#973 id=973 data-nosnippet>973</a>        ]));
-<a href=#974 id=974 data-nosnippet>974</a>
-<a href=#975 id=975 data-nosnippet>975</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#976 id=976 data-nosnippet>976</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#977 id=977 data-nosnippet>977</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#978 id=978 data-nosnippet>978</a>
-<a href=#979 id=979 data-nosnippet>979</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>])) <span class="kw">as </span>ArrayRef;
-<a href=#980 id=980 data-nosnippet>980</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
-<a href=#981 id=981 data-nosnippet>981</a>
-<a href=#982 id=982 data-nosnippet>982</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted_no_key.parquet"</span>);
-<a href=#983 id=983 data-nosnippet>983</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">None</span>);
-<a href=#984 id=984 data-nosnippet>984</a>
-<a href=#985 id=985 data-nosnippet>985</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#986 id=986 data-nosnippet>986</a>
-<a href=#987 id=987 data-nosnippet>987</a>        <span class="kw">let </span>task = FileScanTask::builder()
-<a href=#988 id=988 data-nosnippet>988</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
-<a href=#989 id=989 data-nosnippet>989</a>            .with_start(<span class="number">0</span>)
-<a href=#990 id=990 data-nosnippet>990</a>            .with_length(<span class="number">0</span>)
-<a href=#991 id=991 data-nosnippet>991</a>            .with_data_file_path(file_path)
-<a href=#992 id=992 data-nosnippet>992</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#993 id=993 data-nosnippet>993</a>            .with_schema(schema)
-<a href=#994 id=994 data-nosnippet>994</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#995 id=995 data-nosnippet>995</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#996 id=996 data-nosnippet>996</a>            .build();
-<a href=#997 id=997 data-nosnippet>997</a>
-<a href=#998 id=998 data-nosnippet>998</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#999 id=999 data-nosnippet>999</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
-<a href=#1000 id=1000 data-nosnippet>1000</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
-<a href=#1001 id=1001 data-nosnippet>1001</a>
-<a href=#1002 id=1002 data-nosnippet>1002</a>        <span class="kw">let </span>err = result.unwrap_err();
-<a href=#1003 id=1003 data-nosnippet>1003</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::Unexpected);
-<a href=#1004 id=1004 data-nosnippet>1004</a>        <span class="kw">let </span>err_str = <span class="macro">format!</span>(<span class="string">"{err}"</span>);
-<a href=#1005 id=1005 data-nosnippet>1005</a>        <span class="macro">assert!</span>(
-<a href=#1006 id=1006 data-nosnippet>1006</a>            err_str.contains(<span class="string">"encrypted footer"</span>),
-<a href=#1007 id=1007 data-nosnippet>1007</a>            <span class="string">"Expected error about encrypted footer, got: {err_str}"
-<a href=#1008 id=1008 data-nosnippet>1008</a>        </span>);
-<a href=#1009 id=1009 data-nosnippet>1009</a>        <span class="macro">assert!</span>(
-<a href=#1010 id=1010 data-nosnippet>1010</a>            err_str.contains(<span class="string">"decryption properties were not provided"</span>),
-<a href=#1011 id=1011 data-nosnippet>1011</a>            <span class="string">"Expected error about missing decryption properties, got: {err_str}"
-<a href=#1012 id=1012 data-nosnippet>1012</a>        </span>);
-<a href=#1013 id=1013 data-nosnippet>1013</a>    }
-<a href=#1014 id=1014 data-nosnippet>1014</a>
-<a href=#1015 id=1015 data-nosnippet>1015</a>    <span class="doccomment">/// Writes a plain (unencrypted) single-column Int32 "id" parquet file with the
-<a href=#1016 id=1016 data-nosnippet>1016</a>    /// given extra Arrow fields/columns appended, returning the file path.
-<a href=#1017 id=1017 data-nosnippet>1017</a>    </span><span class="kw">fn </span>write_plain_parquet(
-<a href=#1018 id=1018 data-nosnippet>1018</a>        dir: <span class="kw-2">&amp;</span>str,
-<a href=#1019 id=1019 data-nosnippet>1019</a>        name: <span class="kw-2">&amp;</span>str,
-<a href=#1020 id=1020 data-nosnippet>1020</a>        extra_fields: Vec&lt;Field&gt;,
-<a href=#1021 id=1021 data-nosnippet>1021</a>        extra_columns: Vec&lt;ArrayRef&gt;,
-<a href=#1022 id=1022 data-nosnippet>1022</a>    ) -&gt; String {
-<a href=#1023 id=1023 data-nosnippet>1023</a>        <span class="kw">let </span><span class="kw-2">mut </span>fields =
-<a href=#1024 id=1024 data-nosnippet>1024</a>            <span class="macro">vec!</span>[
-<a href=#1025 id=1025 data-nosnippet>1025</a>                Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#1026 id=1026 data-nosnippet>1026</a>                    PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#1027 id=1027 data-nosnippet>1027</a>                    <span class="string">"1"</span>.to_string(),
-<a href=#1028 id=1028 data-nosnippet>1028</a>                )])),
-<a href=#1029 id=1029 data-nosnippet>1029</a>            ];
-<a href=#1030 id=1030 data-nosnippet>1030</a>        fields.extend(extra_fields);
-<a href=#1031 id=1031 data-nosnippet>1031</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(fields));
-<a href=#1032 id=1032 data-nosnippet>1032</a>
-<a href=#1033 id=1033 data-nosnippet>1033</a>        <span class="kw">let </span><span class="kw-2">mut </span>columns: Vec&lt;ArrayRef&gt; = <span class="macro">vec!</span>[Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>]))];
-<a href=#1034 id=1034 data-nosnippet>1034</a>        columns.extend(extra_columns);
-<a href=#1035 id=1035 data-nosnippet>1035</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), columns).unwrap();
+<a href=#930 id=930 data-nosnippet>930</a>            .with_aad_prefix(aad_prefix)
+<a href=#931 id=931 data-nosnippet>931</a>            .encode()
+<a href=#932 id=932 data-nosnippet>932</a>            .unwrap();
+<a href=#933 id=933 data-nosnippet>933</a>
+<a href=#934 id=934 data-nosnippet>934</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#935 id=935 data-nosnippet>935</a>
+<a href=#936 id=936 data-nosnippet>936</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#937 id=937 data-nosnippet>937</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#938 id=938 data-nosnippet>938</a>            .with_start(<span class="number">0</span>)
+<a href=#939 id=939 data-nosnippet>939</a>            .with_length(<span class="number">0</span>)
+<a href=#940 id=940 data-nosnippet>940</a>            .with_data_file_path(file_path)
+<a href=#941 id=941 data-nosnippet>941</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#942 id=942 data-nosnippet>942</a>            .with_schema(schema)
+<a href=#943 id=943 data-nosnippet>943</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#944 id=944 data-nosnippet>944</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#945 id=945 data-nosnippet>945</a>            .with_key_metadata(<span class="prelude-val">Some</span>(key_metadata))
+<a href=#946 id=946 data-nosnippet>946</a>            .build();
+<a href=#947 id=947 data-nosnippet>947</a>
+<a href=#948 id=948 data-nosnippet>948</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#949 id=949 data-nosnippet>949</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#950 id=950 data-nosnippet>950</a>            .read(tasks)
+<a href=#951 id=951 data-nosnippet>951</a>            .unwrap()
+<a href=#952 id=952 data-nosnippet>952</a>            .stream()
+<a href=#953 id=953 data-nosnippet>953</a>            .try_collect()
+<a href=#954 id=954 data-nosnippet>954</a>            .<span class="kw">await
+<a href=#955 id=955 data-nosnippet>955</a>            </span>.unwrap();
+<a href=#956 id=956 data-nosnippet>956</a>
+<a href=#957 id=957 data-nosnippet>957</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
+<a href=#958 id=958 data-nosnippet>958</a>        <span class="kw">let </span>ids = batches[<span class="number">0</span>]
+<a href=#959 id=959 data-nosnippet>959</a>            .column(<span class="number">0</span>)
+<a href=#960 id=960 data-nosnippet>960</a>            .as_any()
+<a href=#961 id=961 data-nosnippet>961</a>            .downcast_ref::&lt;Int32Array&gt;()
+<a href=#962 id=962 data-nosnippet>962</a>            .unwrap();
+<a href=#963 id=963 data-nosnippet>963</a>        <span class="macro">assert_eq!</span>(ids.values(), <span class="kw-2">&amp;</span>[<span class="number">10</span>, <span class="number">20</span>, <span class="number">30</span>]);
+<a href=#964 id=964 data-nosnippet>964</a>    }
+<a href=#965 id=965 data-nosnippet>965</a>
+<a href=#966 id=966 data-nosnippet>966</a>    <span class="attr">#[tokio::test]
+<a href=#967 id=967 data-nosnippet>967</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_aes_128() {
+<a href=#968 id=968 data-nosnippet>968</a>        assert_encrypted_parquet_roundtrip(<span class="string">b"0123456789abcdef"</span>).<span class="kw">await</span>;
+<a href=#969 id=969 data-nosnippet>969</a>    }
+<a href=#970 id=970 data-nosnippet>970</a>
+<a href=#971 id=971 data-nosnippet>971</a>    <span class="attr">#[tokio::test]
+<a href=#972 id=972 data-nosnippet>972</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_aes_256() {
+<a href=#973 id=973 data-nosnippet>973</a>        assert_encrypted_parquet_roundtrip(<span class="string">b"0123456789abcdef0123456789abcdef"</span>).<span class="kw">await</span>;
+<a href=#974 id=974 data-nosnippet>974</a>    }
+<a href=#975 id=975 data-nosnippet>975</a>
+<a href=#976 id=976 data-nosnippet>976</a>    <span class="attr">#[tokio::test]
+<a href=#977 id=977 data-nosnippet>977</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_without_key_metadata_fails() {
+<a href=#978 id=978 data-nosnippet>978</a>        <span class="kw">let </span>encryption_key = <span class="string">b"0123456789abcdef"</span>;
+<a href=#979 id=979 data-nosnippet>979</a>
+<a href=#980 id=980 data-nosnippet>980</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#981 id=981 data-nosnippet>981</a>            Schema::builder()
+<a href=#982 id=982 data-nosnippet>982</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#983 id=983 data-nosnippet>983</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#984 id=984 data-nosnippet>984</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#985 id=985 data-nosnippet>985</a>                ])
+<a href=#986 id=986 data-nosnippet>986</a>                .build()
+<a href=#987 id=987 data-nosnippet>987</a>                .unwrap(),
+<a href=#988 id=988 data-nosnippet>988</a>        );
+<a href=#989 id=989 data-nosnippet>989</a>
+<a href=#990 id=990 data-nosnippet>990</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#991 id=991 data-nosnippet>991</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#992 id=992 data-nosnippet>992</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#993 id=993 data-nosnippet>993</a>                <span class="string">"1"</span>.to_string(),
+<a href=#994 id=994 data-nosnippet>994</a>            )])),
+<a href=#995 id=995 data-nosnippet>995</a>        ]));
+<a href=#996 id=996 data-nosnippet>996</a>
+<a href=#997 id=997 data-nosnippet>997</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#998 id=998 data-nosnippet>998</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#999 id=999 data-nosnippet>999</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#1000 id=1000 data-nosnippet>1000</a>
+<a href=#1001 id=1001 data-nosnippet>1001</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>])) <span class="kw">as </span>ArrayRef;
+<a href=#1002 id=1002 data-nosnippet>1002</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
+<a href=#1003 id=1003 data-nosnippet>1003</a>
+<a href=#1004 id=1004 data-nosnippet>1004</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted_no_key.parquet"</span>);
+<a href=#1005 id=1005 data-nosnippet>1005</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">None</span>);
+<a href=#1006 id=1006 data-nosnippet>1006</a>
+<a href=#1007 id=1007 data-nosnippet>1007</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#1008 id=1008 data-nosnippet>1008</a>
+<a href=#1009 id=1009 data-nosnippet>1009</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#1010 id=1010 data-nosnippet>1010</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#1011 id=1011 data-nosnippet>1011</a>            .with_start(<span class="number">0</span>)
+<a href=#1012 id=1012 data-nosnippet>1012</a>            .with_length(<span class="number">0</span>)
+<a href=#1013 id=1013 data-nosnippet>1013</a>            .with_data_file_path(file_path)
+<a href=#1014 id=1014 data-nosnippet>1014</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1015 id=1015 data-nosnippet>1015</a>            .with_schema(schema)
+<a href=#1016 id=1016 data-nosnippet>1016</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#1017 id=1017 data-nosnippet>1017</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1018 id=1018 data-nosnippet>1018</a>            .build();
+<a href=#1019 id=1019 data-nosnippet>1019</a>
+<a href=#1020 id=1020 data-nosnippet>1020</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1021 id=1021 data-nosnippet>1021</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
+<a href=#1022 id=1022 data-nosnippet>1022</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1023 id=1023 data-nosnippet>1023</a>
+<a href=#1024 id=1024 data-nosnippet>1024</a>        <span class="kw">let </span>err = result.unwrap_err();
+<a href=#1025 id=1025 data-nosnippet>1025</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::Unexpected);
+<a href=#1026 id=1026 data-nosnippet>1026</a>        <span class="kw">let </span>err_str = <span class="macro">format!</span>(<span class="string">"{err}"</span>);
+<a href=#1027 id=1027 data-nosnippet>1027</a>        <span class="macro">assert!</span>(
+<a href=#1028 id=1028 data-nosnippet>1028</a>            err_str.contains(<span class="string">"encrypted footer"</span>),
+<a href=#1029 id=1029 data-nosnippet>1029</a>            <span class="string">"Expected error about encrypted footer, got: {err_str}"
+<a href=#1030 id=1030 data-nosnippet>1030</a>        </span>);
+<a href=#1031 id=1031 data-nosnippet>1031</a>        <span class="macro">assert!</span>(
+<a href=#1032 id=1032 data-nosnippet>1032</a>            err_str.contains(<span class="string">"decryption properties were not provided"</span>),
+<a href=#1033 id=1033 data-nosnippet>1033</a>            <span class="string">"Expected error about missing decryption properties, got: {err_str}"
+<a href=#1034 id=1034 data-nosnippet>1034</a>        </span>);
+<a href=#1035 id=1035 data-nosnippet>1035</a>    }
 <a href=#1036 id=1036 data-nosnippet>1036</a>
-<a href=#1037 id=1037 data-nosnippet>1037</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{dir}/{name}"</span>);
-<a href=#1038 id=1038 data-nosnippet>1038</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
-<a href=#1039 id=1039 data-nosnippet>1039</a>        <span class="kw">let </span>props = WriterProperties::builder()
-<a href=#1040 id=1040 data-nosnippet>1040</a>            .set_compression(Compression::SNAPPY)
-<a href=#1041 id=1041 data-nosnippet>1041</a>            .build();
-<a href=#1042 id=1042 data-nosnippet>1042</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema, <span class="prelude-val">Some</span>(props)).unwrap();
-<a href=#1043 id=1043 data-nosnippet>1043</a>        writer.write(<span class="kw-2">&amp;</span>batch).unwrap();
-<a href=#1044 id=1044 data-nosnippet>1044</a>        writer.close().unwrap();
-<a href=#1045 id=1045 data-nosnippet>1045</a>        file_path
-<a href=#1046 id=1046 data-nosnippet>1046</a>    }
-<a href=#1047 id=1047 data-nosnippet>1047</a>
-<a href=#1048 id=1048 data-nosnippet>1048</a>    <span class="kw">fn </span>last_updated_seq_task(
-<a href=#1049 id=1049 data-nosnippet>1049</a>        file_path: String,
-<a href=#1050 id=1050 data-nosnippet>1050</a>        first_row_id: <span class="prelude-ty">Option</span>&lt;i64&gt;,
-<a href=#1051 id=1051 data-nosnippet>1051</a>        data_sequence_number: <span class="prelude-ty">Option</span>&lt;i64&gt;,
-<a href=#1052 id=1052 data-nosnippet>1052</a>    ) -&gt; FileScanTask {
-<a href=#1053 id=1053 data-nosnippet>1053</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER;
+<a href=#1037 id=1037 data-nosnippet>1037</a>    <span class="doccomment">/// Writes a plain (unencrypted) single-column Int32 "id" parquet file with the
+<a href=#1038 id=1038 data-nosnippet>1038</a>    /// given extra Arrow fields/columns appended, returning the file path.
+<a href=#1039 id=1039 data-nosnippet>1039</a>    </span><span class="kw">fn </span>write_plain_parquet(
+<a href=#1040 id=1040 data-nosnippet>1040</a>        dir: <span class="kw-2">&amp;</span>str,
+<a href=#1041 id=1041 data-nosnippet>1041</a>        name: <span class="kw-2">&amp;</span>str,
+<a href=#1042 id=1042 data-nosnippet>1042</a>        extra_fields: Vec&lt;Field&gt;,
+<a href=#1043 id=1043 data-nosnippet>1043</a>        extra_columns: Vec&lt;ArrayRef&gt;,
+<a href=#1044 id=1044 data-nosnippet>1044</a>    ) -&gt; String {
+<a href=#1045 id=1045 data-nosnippet>1045</a>        <span class="kw">let </span><span class="kw-2">mut </span>fields =
+<a href=#1046 id=1046 data-nosnippet>1046</a>            <span class="macro">vec!</span>[
+<a href=#1047 id=1047 data-nosnippet>1047</a>                Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#1048 id=1048 data-nosnippet>1048</a>                    PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#1049 id=1049 data-nosnippet>1049</a>                    <span class="string">"1"</span>.to_string(),
+<a href=#1050 id=1050 data-nosnippet>1050</a>                )])),
+<a href=#1051 id=1051 data-nosnippet>1051</a>            ];
+<a href=#1052 id=1052 data-nosnippet>1052</a>        fields.extend(extra_fields);
+<a href=#1053 id=1053 data-nosnippet>1053</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(fields));
 <a href=#1054 id=1054 data-nosnippet>1054</a>
-<a href=#1055 id=1055 data-nosnippet>1055</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#1056 id=1056 data-nosnippet>1056</a>            Schema::builder()
-<a href=#1057 id=1057 data-nosnippet>1057</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1058 id=1058 data-nosnippet>1058</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1059 id=1059 data-nosnippet>1059</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#1060 id=1060 data-nosnippet>1060</a>                ])
-<a href=#1061 id=1061 data-nosnippet>1061</a>                .build()
-<a href=#1062 id=1062 data-nosnippet>1062</a>                .unwrap(),
-<a href=#1063 id=1063 data-nosnippet>1063</a>        );
-<a href=#1064 id=1064 data-nosnippet>1064</a>
-<a href=#1065 id=1065 data-nosnippet>1065</a>        FileScanTask::builder()
-<a href=#1066 id=1066 data-nosnippet>1066</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
-<a href=#1067 id=1067 data-nosnippet>1067</a>            .with_start(<span class="number">0</span>)
-<a href=#1068 id=1068 data-nosnippet>1068</a>            .with_length(<span class="number">0</span>)
-<a href=#1069 id=1069 data-nosnippet>1069</a>            .with_data_file_path(file_path)
-<a href=#1070 id=1070 data-nosnippet>1070</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1071 id=1071 data-nosnippet>1071</a>            .with_schema(schema)
-<a href=#1072 id=1072 data-nosnippet>1072</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER])
-<a href=#1073 id=1073 data-nosnippet>1073</a>            .with_first_row_id(first_row_id)
-<a href=#1074 id=1074 data-nosnippet>1074</a>            .with_data_sequence_number(data_sequence_number)
-<a href=#1075 id=1075 data-nosnippet>1075</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1076 id=1076 data-nosnippet>1076</a>            .build()
-<a href=#1077 id=1077 data-nosnippet>1077</a>    }
-<a href=#1078 id=1078 data-nosnippet>1078</a>
-<a href=#1079 id=1079 data-nosnippet>1079</a>    <span class="doccomment">/// Asserts the logical per-row values of the `_last_updated_sequence_number`
-<a href=#1080 id=1080 data-nosnippet>1080</a>    /// column across all batches, independent of the physical (run-end) encoding.
-<a href=#1081 id=1081 data-nosnippet>1081</a>    </span><span class="kw">fn </span>assert_last_updated_seq_column(batches: <span class="kw-2">&amp;</span>[RecordBatch], expected: <span class="kw-2">&amp;</span>[<span class="prelude-ty">Option</span>&lt;i64&gt;]) {
-<a href=#1082 id=1082 data-nosnippet>1082</a>        <span class="kw">use </span>arrow_array::cast::AsArray;
-<a href=#1083 id=1083 data-nosnippet>1083</a>        <span class="kw">use </span>arrow_cast::cast;
-<a href=#1084 id=1084 data-nosnippet>1084</a>        <span class="kw">use </span>arrow_schema::DataType;
-<a href=#1085 id=1085 data-nosnippet>1085</a>
-<a href=#1086 id=1086 data-nosnippet>1086</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER;
-<a href=#1087 id=1087 data-nosnippet>1087</a>
-<a href=#1088 id=1088 data-nosnippet>1088</a>        <span class="kw">let </span><span class="kw-2">mut </span>actual = Vec::new();
-<a href=#1089 id=1089 data-nosnippet>1089</a>        <span class="kw">for </span>batch <span class="kw">in </span>batches {
-<a href=#1090 id=1090 data-nosnippet>1090</a>            <span class="kw">let </span>col = batch
-<a href=#1091 id=1091 data-nosnippet>1091</a>                .column_by_name(RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER)
-<a href=#1092 id=1092 data-nosnippet>1092</a>                .expect(<span class="string">"_last_updated_sequence_number column should be present"</span>);
-<a href=#1093 id=1093 data-nosnippet>1093</a>            <span class="kw">let </span>logical = cast(col, <span class="kw-2">&amp;</span>DataType::Int64).unwrap();
-<a href=#1094 id=1094 data-nosnippet>1094</a>            <span class="kw">let </span>values = logical.as_primitive::&lt;arrow_array::types::Int64Type&gt;();
-<a href=#1095 id=1095 data-nosnippet>1095</a>            <span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..values.len() {
-<a href=#1096 id=1096 data-nosnippet>1096</a>                actual.push((!values.is_null(i)).then(|| values.value(i)));
-<a href=#1097 id=1097 data-nosnippet>1097</a>            }
-<a href=#1098 id=1098 data-nosnippet>1098</a>        }
-<a href=#1099 id=1099 data-nosnippet>1099</a>        <span class="macro">assert_eq!</span>(actual, expected);
-<a href=#1100 id=1100 data-nosnippet>1100</a>    }
-<a href=#1101 id=1101 data-nosnippet>1101</a>
-<a href=#1102 id=1102 data-nosnippet>1102</a>    <span class="attr">#[tokio::test]
-<a href=#1103 id=1103 data-nosnippet>1103</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_null_when_no_first_row_id() {
-<a href=#1104 id=1104 data-nosnippet>1104</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1105 id=1105 data-nosnippet>1105</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1106 id=1106 data-nosnippet>1106</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"no_first_row_id.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#1055 id=1055 data-nosnippet>1055</a>        <span class="kw">let </span><span class="kw-2">mut </span>columns: Vec&lt;ArrayRef&gt; = <span class="macro">vec!</span>[Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>]))];
+<a href=#1056 id=1056 data-nosnippet>1056</a>        columns.extend(extra_columns);
+<a href=#1057 id=1057 data-nosnippet>1057</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), columns).unwrap();
+<a href=#1058 id=1058 data-nosnippet>1058</a>
+<a href=#1059 id=1059 data-nosnippet>1059</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{dir}/{name}"</span>);
+<a href=#1060 id=1060 data-nosnippet>1060</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
+<a href=#1061 id=1061 data-nosnippet>1061</a>        <span class="kw">let </span>props = WriterProperties::builder()
+<a href=#1062 id=1062 data-nosnippet>1062</a>            .set_compression(Compression::SNAPPY)
+<a href=#1063 id=1063 data-nosnippet>1063</a>            .build();
+<a href=#1064 id=1064 data-nosnippet>1064</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema, <span class="prelude-val">Some</span>(props)).unwrap();
+<a href=#1065 id=1065 data-nosnippet>1065</a>        writer.write(<span class="kw-2">&amp;</span>batch).unwrap();
+<a href=#1066 id=1066 data-nosnippet>1066</a>        writer.close().unwrap();
+<a href=#1067 id=1067 data-nosnippet>1067</a>        file_path
+<a href=#1068 id=1068 data-nosnippet>1068</a>    }
+<a href=#1069 id=1069 data-nosnippet>1069</a>
+<a href=#1070 id=1070 data-nosnippet>1070</a>    <span class="kw">fn </span>last_updated_seq_task(
+<a href=#1071 id=1071 data-nosnippet>1071</a>        file_path: String,
+<a href=#1072 id=1072 data-nosnippet>1072</a>        first_row_id: <span class="prelude-ty">Option</span>&lt;i64&gt;,
+<a href=#1073 id=1073 data-nosnippet>1073</a>        data_sequence_number: <span class="prelude-ty">Option</span>&lt;i64&gt;,
+<a href=#1074 id=1074 data-nosnippet>1074</a>    ) -&gt; FileScanTask {
+<a href=#1075 id=1075 data-nosnippet>1075</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER;
+<a href=#1076 id=1076 data-nosnippet>1076</a>
+<a href=#1077 id=1077 data-nosnippet>1077</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#1078 id=1078 data-nosnippet>1078</a>            Schema::builder()
+<a href=#1079 id=1079 data-nosnippet>1079</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1080 id=1080 data-nosnippet>1080</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1081 id=1081 data-nosnippet>1081</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#1082 id=1082 data-nosnippet>1082</a>                ])
+<a href=#1083 id=1083 data-nosnippet>1083</a>                .build()
+<a href=#1084 id=1084 data-nosnippet>1084</a>                .unwrap(),
+<a href=#1085 id=1085 data-nosnippet>1085</a>        );
+<a href=#1086 id=1086 data-nosnippet>1086</a>
+<a href=#1087 id=1087 data-nosnippet>1087</a>        FileScanTask::builder()
+<a href=#1088 id=1088 data-nosnippet>1088</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#1089 id=1089 data-nosnippet>1089</a>            .with_start(<span class="number">0</span>)
+<a href=#1090 id=1090 data-nosnippet>1090</a>            .with_length(<span class="number">0</span>)
+<a href=#1091 id=1091 data-nosnippet>1091</a>            .with_data_file_path(file_path)
+<a href=#1092 id=1092 data-nosnippet>1092</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1093 id=1093 data-nosnippet>1093</a>            .with_schema(schema)
+<a href=#1094 id=1094 data-nosnippet>1094</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER])
+<a href=#1095 id=1095 data-nosnippet>1095</a>            .with_first_row_id(first_row_id)
+<a href=#1096 id=1096 data-nosnippet>1096</a>            .with_data_sequence_number(data_sequence_number)
+<a href=#1097 id=1097 data-nosnippet>1097</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1098 id=1098 data-nosnippet>1098</a>            .build()
+<a href=#1099 id=1099 data-nosnippet>1099</a>    }
+<a href=#1100 id=1100 data-nosnippet>1100</a>
+<a href=#1101 id=1101 data-nosnippet>1101</a>    <span class="doccomment">/// Asserts the logical per-row values of the `_last_updated_sequence_number`
+<a href=#1102 id=1102 data-nosnippet>1102</a>    /// column across all batches, independent of the physical (run-end) encoding.
+<a href=#1103 id=1103 data-nosnippet>1103</a>    </span><span class="kw">fn </span>assert_last_updated_seq_column(batches: <span class="kw-2">&amp;</span>[RecordBatch], expected: <span class="kw-2">&amp;</span>[<span class="prelude-ty">Option</span>&lt;i64&gt;]) {
+<a href=#1104 id=1104 data-nosnippet>1104</a>        <span class="kw">use </span>arrow_array::cast::AsArray;
+<a href=#1105 id=1105 data-nosnippet>1105</a>        <span class="kw">use </span>arrow_cast::cast;
+<a href=#1106 id=1106 data-nosnippet>1106</a>        <span class="kw">use </span>arrow_schema::DataType;
 <a href=#1107 id=1107 data-nosnippet>1107</a>
-<a href=#1108 id=1108 data-nosnippet>1108</a>        <span class="comment">// A file with a null first_row_id (v1/v2, or a pre-upgrade v3 snapshot) produces
-<a href=#1109 id=1109 data-nosnippet>1109</a>        // a null _last_updated_sequence_number column, even though it has a data
-<a href=#1110 id=1110 data-nosnippet>1110</a>        // sequence number; the spec gates both lineage columns on first_row_id.
-<a href=#1111 id=1111 data-nosnippet>1111</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">9</span>));
-<a href=#1112 id=1112 data-nosnippet>1112</a>
-<a href=#1113 id=1113 data-nosnippet>1113</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1114 id=1114 data-nosnippet>1114</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1115 id=1115 data-nosnippet>1115</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1116 id=1116 data-nosnippet>1116</a>            .read(tasks)
-<a href=#1117 id=1117 data-nosnippet>1117</a>            .unwrap()
-<a href=#1118 id=1118 data-nosnippet>1118</a>            .stream()
-<a href=#1119 id=1119 data-nosnippet>1119</a>            .try_collect()
-<a href=#1120 id=1120 data-nosnippet>1120</a>            .<span class="kw">await
-<a href=#1121 id=1121 data-nosnippet>1121</a>            </span>.unwrap();
-<a href=#1122 id=1122 data-nosnippet>1122</a>
-<a href=#1123 id=1123 data-nosnippet>1123</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">None</span>, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>]);
-<a href=#1124 id=1124 data-nosnippet>1124</a>    }
-<a href=#1125 id=1125 data-nosnippet>1125</a>
-<a href=#1126 id=1126 data-nosnippet>1126</a>    <span class="attr">#[tokio::test]
-<a href=#1127 id=1127 data-nosnippet>1127</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_error_when_no_data_seq() {
-<a href=#1128 id=1128 data-nosnippet>1128</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1129 id=1129 data-nosnippet>1129</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1130 id=1130 data-nosnippet>1130</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"no_data_seq.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
-<a href=#1131 id=1131 data-nosnippet>1131</a>
-<a href=#1132 id=1132 data-nosnippet>1132</a>        <span class="comment">// first_row_id present but data_sequence_number absent: after manifest
-<a href=#1133 id=1133 data-nosnippet>1133</a>        // inheritance a committed entry always has one, so this is a malformed
-<a href=#1134 id=1134 data-nosnippet>1134</a>        // manifest and must error rather than fabricate or null the column.
-<a href=#1135 id=1135 data-nosnippet>1135</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">42</span>), <span class="prelude-val">None</span>);
-<a href=#1136 id=1136 data-nosnippet>1136</a>
-<a href=#1137 id=1137 data-nosnippet>1137</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1138 id=1138 data-nosnippet>1138</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1139 id=1139 data-nosnippet>1139</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
-<a href=#1140 id=1140 data-nosnippet>1140</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
-<a href=#1141 id=1141 data-nosnippet>1141</a>
-<a href=#1142 id=1142 data-nosnippet>1142</a>        <span class="kw">let </span>err = result.unwrap_err();
-<a href=#1143 id=1143 data-nosnippet>1143</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::DataInvalid);
-<a href=#1144 id=1144 data-nosnippet>1144</a>        <span class="macro">assert!</span>(
-<a href=#1145 id=1145 data-nosnippet>1145</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"no data sequence number"</span>),
-<a href=#1146 id=1146 data-nosnippet>1146</a>            <span class="string">"unexpected error: {err}"
-<a href=#1147 id=1147 data-nosnippet>1147</a>        </span>);
-<a href=#1148 id=1148 data-nosnippet>1148</a>    }
-<a href=#1149 id=1149 data-nosnippet>1149</a>
-<a href=#1150 id=1150 data-nosnippet>1150</a>    <span class="attr">#[tokio::test]
-<a href=#1151 id=1151 data-nosnippet>1151</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_derived_from_data_seq() {
-<a href=#1152 id=1152 data-nosnippet>1152</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1153 id=1153 data-nosnippet>1153</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1154 id=1154 data-nosnippet>1154</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"with_first_row_id.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
-<a href=#1155 id=1155 data-nosnippet>1155</a>
-<a href=#1156 id=1156 data-nosnippet>1156</a>        <span class="comment">// Non-null first_row_id + data sequence number -&gt; the derived value (the data
-<a href=#1157 id=1157 data-nosnippet>1157</a>        // sequence number) for every row. This is the only value-producing arm.
-<a href=#1158 id=1158 data-nosnippet>1158</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">42</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>));
-<a href=#1159 id=1159 data-nosnippet>1159</a>
-<a href=#1160 id=1160 data-nosnippet>1160</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1161 id=1161 data-nosnippet>1161</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1162 id=1162 data-nosnippet>1162</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1163 id=1163 data-nosnippet>1163</a>            .read(tasks)
-<a href=#1164 id=1164 data-nosnippet>1164</a>            .unwrap()
-<a href=#1165 id=1165 data-nosnippet>1165</a>            .stream()
-<a href=#1166 id=1166 data-nosnippet>1166</a>            .try_collect()
-<a href=#1167 id=1167 data-nosnippet>1167</a>            .<span class="kw">await
-<a href=#1168 id=1168 data-nosnippet>1168</a>            </span>.unwrap();
-<a href=#1169 id=1169 data-nosnippet>1169</a>
-<a href=#1170 id=1170 data-nosnippet>1170</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">7</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>)]);
-<a href=#1171 id=1171 data-nosnippet>1171</a>    }
-<a href=#1172 id=1172 data-nosnippet>1172</a>
-<a href=#1173 id=1173 data-nosnippet>1173</a>    <span class="attr">#[tokio::test]
-<a href=#1174 id=1174 data-nosnippet>1174</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_mixed_files_share_schema() {
-<a href=#1175 id=1175 data-nosnippet>1175</a>        <span class="kw">use </span>arrow_select::concat::concat_batches;
-<a href=#1176 id=1176 data-nosnippet>1176</a>
-<a href=#1177 id=1177 data-nosnippet>1177</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1178 id=1178 data-nosnippet>1178</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1179 id=1179 data-nosnippet>1179</a>
-<a href=#1180 id=1180 data-nosnippet>1180</a>        <span class="comment">// Three files in one scan exercising all three column paths, which must all
-<a href=#1181 id=1181 data-nosnippet>1181</a>        // produce the SAME Arrow type (run-end-encoded) or concatenation fails:
-<a href=#1182 id=1182 data-nosnippet>1182</a>        //   - constant: first_row_id set, no physical column -&gt; derived constant
-<a href=#1183 id=1183 data-nosnippet>1183</a>        //   - null gate: no first_row_id -&gt; null column
-<a href=#1184 id=1184 data-nosnippet>1184</a>        //   - coalesce: first_row_id set, physical column present -&gt; per-row + fallback
-<a href=#1185 id=1185 data-nosnippet>1185</a>        </span><span class="kw">let </span>constant = last_updated_seq_task(
-<a href=#1186 id=1186 data-nosnippet>1186</a>            write_plain_parquet(dir, <span class="string">"constant.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
-<a href=#1187 id=1187 data-nosnippet>1187</a>            <span class="prelude-val">Some</span>(<span class="number">42</span>),
-<a href=#1188 id=1188 data-nosnippet>1188</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
-<a href=#1189 id=1189 data-nosnippet>1189</a>        );
-<a href=#1190 id=1190 data-nosnippet>1190</a>        <span class="kw">let </span>nulled = last_updated_seq_task(
-<a href=#1191 id=1191 data-nosnippet>1191</a>            write_plain_parquet(dir, <span class="string">"nulled.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
-<a href=#1192 id=1192 data-nosnippet>1192</a>            <span class="prelude-val">None</span>,
-<a href=#1193 id=1193 data-nosnippet>1193</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
-<a href=#1194 id=1194 data-nosnippet>1194</a>        );
-<a href=#1195 id=1195 data-nosnippet>1195</a>        <span class="kw">let </span>coalesced = last_updated_seq_task(
-<a href=#1196 id=1196 data-nosnippet>1196</a>            write_plain_parquet(
-<a href=#1197 id=1197 data-nosnippet>1197</a>                dir,
-<a href=#1198 id=1198 data-nosnippet>1198</a>                <span class="string">"coalesced.parquet"</span>,
-<a href=#1199 id=1199 data-nosnippet>1199</a>                <span class="macro">vec!</span>[physical_last_updated_seq_field()],
-<a href=#1200 id=1200 data-nosnippet>1200</a>                <span class="macro">vec!</span>[Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef],
-<a href=#1201 id=1201 data-nosnippet>1201</a>            ),
-<a href=#1202 id=1202 data-nosnippet>1202</a>            <span class="prelude-val">Some</span>(<span class="number">50</span>),
-<a href=#1203 id=1203 data-nosnippet>1203</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
-<a href=#1204 id=1204 data-nosnippet>1204</a>        );
-<a href=#1205 id=1205 data-nosnippet>1205</a>
-<a href=#1206 id=1206 data-nosnippet>1206</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1207 id=1207 data-nosnippet>1207</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[
-<a href=#1208 id=1208 data-nosnippet>1208</a>            <span class="prelude-val">Ok</span>(constant),
-<a href=#1209 id=1209 data-nosnippet>1209</a>            <span class="prelude-val">Ok</span>(nulled),
-<a href=#1210 id=1210 data-nosnippet>1210</a>            <span class="prelude-val">Ok</span>(coalesced),
-<a href=#1211 id=1211 data-nosnippet>1211</a>        ])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1212 id=1212 data-nosnippet>1212</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1213 id=1213 data-nosnippet>1213</a>            .read(tasks)
-<a href=#1214 id=1214 data-nosnippet>1214</a>            .unwrap()
-<a href=#1215 id=1215 data-nosnippet>1215</a>            .stream()
-<a href=#1216 id=1216 data-nosnippet>1216</a>            .try_collect()
-<a href=#1217 id=1217 data-nosnippet>1217</a>            .<span class="kw">await
-<a href=#1218 id=1218 data-nosnippet>1218</a>            </span>.unwrap();
-<a href=#1219 id=1219 data-nosnippet>1219</a>
-<a href=#1220 id=1220 data-nosnippet>1220</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">3</span>);
-<a href=#1221 id=1221 data-nosnippet>1221</a>        <span class="comment">// Identical schema across all three paths -&gt; concat succeeds.
-<a href=#1222 id=1222 data-nosnippet>1222</a>        </span><span class="kw">let </span>schema = batches[<span class="number">0</span>].schema();
-<a href=#1223 id=1223 data-nosnippet>1223</a>        concat_batches(<span class="kw-2">&amp;</span>schema, <span class="kw-2">&amp;</span>batches)
-<a href=#1224 id=1224 data-nosnippet>1224</a>            .expect(<span class="string">"constant, null and coalesce files must share one column type"</span>);
-<a href=#1225 id=1225 data-nosnippet>1225</a>    }
-<a href=#1226 id=1226 data-nosnippet>1226</a>
-<a href=#1227 id=1227 data-nosnippet>1227</a>    <span class="doccomment">/// A parquet field carrying the embedded `_last_updated_sequence_number` field id.
-<a href=#1228 id=1228 data-nosnippet>1228</a>    </span><span class="kw">fn </span>physical_last_updated_seq_field() -&gt; Field {
-<a href=#1229 id=1229 data-nosnippet>1229</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
-<a href=#1230 id=1230 data-nosnippet>1230</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1231 id=1231 data-nosnippet>1231</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1232 id=1232 data-nosnippet>1232</a>        };
-<a href=#1233 id=1233 data-nosnippet>1233</a>        Field::new(
-<a href=#1234 id=1234 data-nosnippet>1234</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1235 id=1235 data-nosnippet>1235</a>            DataType::Int64,
-<a href=#1236 id=1236 data-nosnippet>1236</a>            <span class="bool-val">true</span>,
-<a href=#1237 id=1237 data-nosnippet>1237</a>        )
-<a href=#1238 id=1238 data-nosnippet>1238</a>        .with_metadata(HashMap::from([(
-<a href=#1239 id=1239 data-nosnippet>1239</a>            PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#1240 id=1240 data-nosnippet>1240</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER.to_string(),
-<a href=#1241 id=1241 data-nosnippet>1241</a>        )]))
-<a href=#1242 id=1242 data-nosnippet>1242</a>    }
-<a href=#1243 id=1243 data-nosnippet>1243</a>
-<a href=#1244 id=1244 data-nosnippet>1244</a>    <span class="attr">#[tokio::test]
-<a href=#1245 id=1245 data-nosnippet>1245</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_coalesced() {
-<a href=#1246 id=1246 data-nosnippet>1246</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1247 id=1247 data-nosnippet>1247</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1248 id=1248 data-nosnippet>1248</a>        <span class="comment">// A file that physically carries the column, as Iceberg Java writes when
-<a href=#1249 id=1249 data-nosnippet>1249</a>        // carrying rows forward across a rewrite: some rows have a stored value, some
-<a href=#1250 id=1250 data-nosnippet>1250</a>        // are null (added/modified rows, inherited on read).
-<a href=#1251 id=1251 data-nosnippet>1251</a>        </span><span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
-<a href=#1252 id=1252 data-nosnippet>1252</a>        <span class="kw">let </span>file_path = write_plain_parquet(
-<a href=#1253 id=1253 data-nosnippet>1253</a>            dir,
-<a href=#1254 id=1254 data-nosnippet>1254</a>            <span class="string">"with_seq.parquet"</span>,
-<a href=#1255 id=1255 data-nosnippet>1255</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
-<a href=#1256 id=1256 data-nosnippet>1256</a>            <span class="macro">vec!</span>[seq_col],
-<a href=#1257 id=1257 data-nosnippet>1257</a>        );
-<a href=#1258 id=1258 data-nosnippet>1258</a>
-<a href=#1259 id=1259 data-nosnippet>1259</a>        <span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>));
-<a href=#1260 id=1260 data-nosnippet>1260</a>
-<a href=#1261 id=1261 data-nosnippet>1261</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1262 id=1262 data-nosnippet>1262</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1263 id=1263 data-nosnippet>1263</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1264 id=1264 data-nosnippet>1264</a>            .read(tasks)
-<a href=#1265 id=1265 data-nosnippet>1265</a>            .unwrap()
-<a href=#1266 id=1266 data-nosnippet>1266</a>            .stream()
-<a href=#1267 id=1267 data-nosnippet>1267</a>            .try_collect()
-<a href=#1268 id=1268 data-nosnippet>1268</a>            .<span class="kw">await
-<a href=#1269 id=1269 data-nosnippet>1269</a>            </span>.unwrap();
-<a href=#1270 id=1270 data-nosnippet>1270</a>
-<a href=#1271 id=1271 data-nosnippet>1271</a>        <span class="comment">// Per-row value where non-null; the data sequence number (9) where null.
-<a href=#1272 id=1272 data-nosnippet>1272</a>        </span>assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>), <span class="prelude-val">Some</span>(<span class="number">8</span>)]);
-<a href=#1273 id=1273 data-nosnippet>1273</a>    }
-<a href=#1274 id=1274 data-nosnippet>1274</a>
-<a href=#1275 id=1275 data-nosnippet>1275</a>    <span class="attr">#[tokio::test]
-<a href=#1276 id=1276 data-nosnippet>1276</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_coalesced_with_pos_column() {
-<a href=#1277 id=1277 data-nosnippet>1277</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
-<a href=#1278 id=1278 data-nosnippet>1278</a>            RESERVED_COL_NAME_POS, RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1279 id=1279 data-nosnippet>1279</a>            RESERVED_FIELD_ID_POS,
-<a href=#1280 id=1280 data-nosnippet>1280</a>        };
-<a href=#1281 id=1281 data-nosnippet>1281</a>
-<a href=#1282 id=1282 data-nosnippet>1282</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1283 id=1283 data-nosnippet>1283</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1284 id=1284 data-nosnippet>1284</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
-<a href=#1285 id=1285 data-nosnippet>1285</a>        <span class="kw">let </span>file_path = write_plain_parquet(
-<a href=#1286 id=1286 data-nosnippet>1286</a>            dir,
-<a href=#1287 id=1287 data-nosnippet>1287</a>            <span class="string">"with_seq_and_pos.parquet"</span>,
-<a href=#1288 id=1288 data-nosnippet>1288</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
-<a href=#1289 id=1289 data-nosnippet>1289</a>            <span class="macro">vec!</span>[seq_col],
-<a href=#1290 id=1290 data-nosnippet>1290</a>        );
-<a href=#1291 id=1291 data-nosnippet>1291</a>
-<a href=#1292 id=1292 data-nosnippet>1292</a>        <span class="comment">// Co-project `_pos` (a virtual column appended to the Arrow output schema) with the
-<a href=#1293 id=1293 data-nosnippet>1293</a>        // physical coalesce column. This guards that the physical column's index is
-<a href=#1294 id=1294 data-nosnippet>1294</a>        // resolved in the Parquet schema, not the Arrow schema (whose indices shift once
-<a href=#1295 id=1295 data-nosnippet>1295</a>        // virtual columns are appended).
-<a href=#1296 id=1296 data-nosnippet>1296</a>        </span><span class="kw">let </span>schema = Arc::new(
-<a href=#1297 id=1297 data-nosnippet>1297</a>            Schema::builder()
-<a href=#1298 id=1298 data-nosnippet>1298</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1299 id=1299 data-nosnippet>1299</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1300 id=1300 data-nosnippet>1300</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#1301 id=1301 data-nosnippet>1301</a>                ])
-<a href=#1302 id=1302 data-nosnippet>1302</a>                .build()
-<a href=#1303 id=1303 data-nosnippet>1303</a>                .unwrap(),
-<a href=#1304 id=1304 data-nosnippet>1304</a>        );
-<a href=#1305 id=1305 data-nosnippet>1305</a>        <span class="kw">let </span>task = FileScanTask::builder()
-<a href=#1306 id=1306 data-nosnippet>1306</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
-<a href=#1307 id=1307 data-nosnippet>1307</a>            .with_start(<span class="number">0</span>)
-<a href=#1308 id=1308 data-nosnippet>1308</a>            .with_length(<span class="number">0</span>)
-<a href=#1309 id=1309 data-nosnippet>1309</a>            .with_data_file_path(file_path)
-<a href=#1310 id=1310 data-nosnippet>1310</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1311 id=1311 data-nosnippet>1311</a>            .with_schema(schema)
-<a href=#1312 id=1312 data-nosnippet>1312</a>            .with_project_field_ids(<span class="macro">vec!</span>[
-<a href=#1313 id=1313 data-nosnippet>1313</a>                <span class="number">1</span>,
-<a href=#1314 id=1314 data-nosnippet>1314</a>                RESERVED_FIELD_ID_POS,
-<a href=#1315 id=1315 data-nosnippet>1315</a>                RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1316 id=1316 data-nosnippet>1316</a>            ])
-<a href=#1317 id=1317 data-nosnippet>1317</a>            .with_first_row_id(<span class="prelude-val">Some</span>(<span class="number">100</span>))
-<a href=#1318 id=1318 data-nosnippet>1318</a>            .with_data_sequence_number(<span class="prelude-val">Some</span>(<span class="number">9</span>))
-<a href=#1319 id=1319 data-nosnippet>1319</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1320 id=1320 data-nosnippet>1320</a>            .build();
-<a href=#1321 id=1321 data-nosnippet>1321</a>
-<a href=#1322 id=1322 data-nosnippet>1322</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1323 id=1323 data-nosnippet>1323</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1324 id=1324 data-nosnippet>1324</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1325 id=1325 data-nosnippet>1325</a>            .read(tasks)
-<a href=#1326 id=1326 data-nosnippet>1326</a>            .unwrap()
-<a href=#1327 id=1327 data-nosnippet>1327</a>            .stream()
-<a href=#1328 id=1328 data-nosnippet>1328</a>            .try_collect()
-<a href=#1329 id=1329 data-nosnippet>1329</a>            .<span class="kw">await
-<a href=#1330 id=1330 data-nosnippet>1330</a>            </span>.unwrap();
-<a href=#1331 id=1331 data-nosnippet>1331</a>
-<a href=#1332 id=1332 data-nosnippet>1332</a>        <span class="comment">// The seq column still coalesces correctly...
-<a href=#1333 id=1333 data-nosnippet>1333</a>        </span>assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>), <span class="prelude-val">Some</span>(<span class="number">8</span>)]);
-<a href=#1334 id=1334 data-nosnippet>1334</a>        <span class="comment">// ...and `_pos` is the row position, unaffected by the physical-column union.
-<a href=#1335 id=1335 data-nosnippet>1335</a>        </span><span class="kw">let </span>pos_col = batches[<span class="number">0</span>]
-<a href=#1336 id=1336 data-nosnippet>1336</a>            .column_by_name(RESERVED_COL_NAME_POS)
-<a href=#1337 id=1337 data-nosnippet>1337</a>            .expect(<span class="string">"_pos column should be present"</span>)
-<a href=#1338 id=1338 data-nosnippet>1338</a>            .as_primitive::&lt;arrow_array::types::Int64Type&gt;();
-<a href=#1339 id=1339 data-nosnippet>1339</a>        <span class="macro">assert_eq!</span>(pos_col.values(), <span class="kw-2">&amp;</span>[<span class="number">0</span>, <span class="number">1</span>, <span class="number">2</span>]);
-<a href=#1340 id=1340 data-nosnippet>1340</a>    }
-<a href=#1341 id=1341 data-nosnippet>1341</a>
-<a href=#1342 id=1342 data-nosnippet>1342</a>    <span class="attr">#[tokio::test]
-<a href=#1343 id=1343 data-nosnippet>1343</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_nulled_without_first_row_id() {
-<a href=#1344 id=1344 data-nosnippet>1344</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1345 id=1345 data-nosnippet>1345</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1346 id=1346 data-nosnippet>1346</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
-<a href=#1347 id=1347 data-nosnippet>1347</a>        <span class="kw">let </span>file_path = write_plain_parquet(
-<a href=#1348 id=1348 data-nosnippet>1348</a>            dir,
-<a href=#1349 id=1349 data-nosnippet>1349</a>            <span class="string">"with_seq_no_first_row_id.parquet"</span>,
-<a href=#1350 id=1350 data-nosnippet>1350</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
-<a href=#1351 id=1351 data-nosnippet>1351</a>            <span class="macro">vec!</span>[seq_col],
-<a href=#1352 id=1352 data-nosnippet>1352</a>        );
+<a href=#1108 id=1108 data-nosnippet>1108</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER;
+<a href=#1109 id=1109 data-nosnippet>1109</a>
+<a href=#1110 id=1110 data-nosnippet>1110</a>        <span class="kw">let </span><span class="kw-2">mut </span>actual = Vec::new();
+<a href=#1111 id=1111 data-nosnippet>1111</a>        <span class="kw">for </span>batch <span class="kw">in </span>batches {
+<a href=#1112 id=1112 data-nosnippet>1112</a>            <span class="kw">let </span>col = batch
+<a href=#1113 id=1113 data-nosnippet>1113</a>                .column_by_name(RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER)
+<a href=#1114 id=1114 data-nosnippet>1114</a>                .expect(<span class="string">"_last_updated_sequence_number column should be present"</span>);
+<a href=#1115 id=1115 data-nosnippet>1115</a>            <span class="kw">let </span>logical = cast(col, <span class="kw-2">&amp;</span>DataType::Int64).unwrap();
+<a href=#1116 id=1116 data-nosnippet>1116</a>            <span class="kw">let </span>values = logical.as_primitive::&lt;arrow_array::types::Int64Type&gt;();
+<a href=#1117 id=1117 data-nosnippet>1117</a>            <span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..values.len() {
+<a href=#1118 id=1118 data-nosnippet>1118</a>                actual.push((!values.is_null(i)).then(|| values.value(i)));
+<a href=#1119 id=1119 data-nosnippet>1119</a>            }
+<a href=#1120 id=1120 data-nosnippet>1120</a>        }
+<a href=#1121 id=1121 data-nosnippet>1121</a>        <span class="macro">assert_eq!</span>(actual, expected);
+<a href=#1122 id=1122 data-nosnippet>1122</a>    }
+<a href=#1123 id=1123 data-nosnippet>1123</a>
+<a href=#1124 id=1124 data-nosnippet>1124</a>    <span class="attr">#[tokio::test]
+<a href=#1125 id=1125 data-nosnippet>1125</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_null_when_no_first_row_id() {
+<a href=#1126 id=1126 data-nosnippet>1126</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1127 id=1127 data-nosnippet>1127</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1128 id=1128 data-nosnippet>1128</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"no_first_row_id.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#1129 id=1129 data-nosnippet>1129</a>
+<a href=#1130 id=1130 data-nosnippet>1130</a>        <span class="comment">// A file with a null first_row_id (v1/v2, or a pre-upgrade v3 snapshot) produces
+<a href=#1131 id=1131 data-nosnippet>1131</a>        // a null _last_updated_sequence_number column, even though it has a data
+<a href=#1132 id=1132 data-nosnippet>1132</a>        // sequence number; the spec gates both lineage columns on first_row_id.
+<a href=#1133 id=1133 data-nosnippet>1133</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">9</span>));
+<a href=#1134 id=1134 data-nosnippet>1134</a>
+<a href=#1135 id=1135 data-nosnippet>1135</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1136 id=1136 data-nosnippet>1136</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1137 id=1137 data-nosnippet>1137</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1138 id=1138 data-nosnippet>1138</a>            .read(tasks)
+<a href=#1139 id=1139 data-nosnippet>1139</a>            .unwrap()
+<a href=#1140 id=1140 data-nosnippet>1140</a>            .stream()
+<a href=#1141 id=1141 data-nosnippet>1141</a>            .try_collect()
+<a href=#1142 id=1142 data-nosnippet>1142</a>            .<span class="kw">await
+<a href=#1143 id=1143 data-nosnippet>1143</a>            </span>.unwrap();
+<a href=#1144 id=1144 data-nosnippet>1144</a>
+<a href=#1145 id=1145 data-nosnippet>1145</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">None</span>, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>]);
+<a href=#1146 id=1146 data-nosnippet>1146</a>    }
+<a href=#1147 id=1147 data-nosnippet>1147</a>
+<a href=#1148 id=1148 data-nosnippet>1148</a>    <span class="attr">#[tokio::test]
+<a href=#1149 id=1149 data-nosnippet>1149</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_error_when_no_data_seq() {
+<a href=#1150 id=1150 data-nosnippet>1150</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1151 id=1151 data-nosnippet>1151</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1152 id=1152 data-nosnippet>1152</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"no_data_seq.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#1153 id=1153 data-nosnippet>1153</a>
+<a href=#1154 id=1154 data-nosnippet>1154</a>        <span class="comment">// first_row_id present but data_sequence_number absent: after manifest
+<a href=#1155 id=1155 data-nosnippet>1155</a>        // inheritance a committed entry always has one, so this is a malformed
+<a href=#1156 id=1156 data-nosnippet>1156</a>        // manifest and must error rather than fabricate or null the column.
+<a href=#1157 id=1157 data-nosnippet>1157</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">42</span>), <span class="prelude-val">None</span>);
+<a href=#1158 id=1158 data-nosnippet>1158</a>
+<a href=#1159 id=1159 data-nosnippet>1159</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1160 id=1160 data-nosnippet>1160</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1161 id=1161 data-nosnippet>1161</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
+<a href=#1162 id=1162 data-nosnippet>1162</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1163 id=1163 data-nosnippet>1163</a>
+<a href=#1164 id=1164 data-nosnippet>1164</a>        <span class="kw">let </span>err = result.unwrap_err();
+<a href=#1165 id=1165 data-nosnippet>1165</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::DataInvalid);
+<a href=#1166 id=1166 data-nosnippet>1166</a>        <span class="macro">assert!</span>(
+<a href=#1167 id=1167 data-nosnippet>1167</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"no data sequence number"</span>),
+<a href=#1168 id=1168 data-nosnippet>1168</a>            <span class="string">"unexpected error: {err}"
+<a href=#1169 id=1169 data-nosnippet>1169</a>        </span>);
+<a href=#1170 id=1170 data-nosnippet>1170</a>    }
+<a href=#1171 id=1171 data-nosnippet>1171</a>
+<a href=#1172 id=1172 data-nosnippet>1172</a>    <span class="attr">#[tokio::test]
+<a href=#1173 id=1173 data-nosnippet>1173</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_derived_from_data_seq() {
+<a href=#1174 id=1174 data-nosnippet>1174</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1175 id=1175 data-nosnippet>1175</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1176 id=1176 data-nosnippet>1176</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"with_first_row_id.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#1177 id=1177 data-nosnippet>1177</a>
+<a href=#1178 id=1178 data-nosnippet>1178</a>        <span class="comment">// Non-null first_row_id + data sequence number -&gt; the derived value (the data
+<a href=#1179 id=1179 data-nosnippet>1179</a>        // sequence number) for every row. This is the only value-producing arm.
+<a href=#1180 id=1180 data-nosnippet>1180</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">42</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>));
+<a href=#1181 id=1181 data-nosnippet>1181</a>
+<a href=#1182 id=1182 data-nosnippet>1182</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1183 id=1183 data-nosnippet>1183</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1184 id=1184 data-nosnippet>1184</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1185 id=1185 data-nosnippet>1185</a>            .read(tasks)
+<a href=#1186 id=1186 data-nosnippet>1186</a>            .unwrap()
+<a href=#1187 id=1187 data-nosnippet>1187</a>            .stream()
+<a href=#1188 id=1188 data-nosnippet>1188</a>            .try_collect()
+<a href=#1189 id=1189 data-nosnippet>1189</a>            .<span class="kw">await
+<a href=#1190 id=1190 data-nosnippet>1190</a>            </span>.unwrap();
+<a href=#1191 id=1191 data-nosnippet>1191</a>
+<a href=#1192 id=1192 data-nosnippet>1192</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">7</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>)]);
+<a href=#1193 id=1193 data-nosnippet>1193</a>    }
+<a href=#1194 id=1194 data-nosnippet>1194</a>
+<a href=#1195 id=1195 data-nosnippet>1195</a>    <span class="attr">#[tokio::test]
+<a href=#1196 id=1196 data-nosnippet>1196</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_mixed_files_share_schema() {
+<a href=#1197 id=1197 data-nosnippet>1197</a>        <span class="kw">use </span>arrow_select::concat::concat_batches;
+<a href=#1198 id=1198 data-nosnippet>1198</a>
+<a href=#1199 id=1199 data-nosnippet>1199</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1200 id=1200 data-nosnippet>1200</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1201 id=1201 data-nosnippet>1201</a>
+<a href=#1202 id=1202 data-nosnippet>1202</a>        <span class="comment">// Three files in one scan exercising all three column paths, which must all
+<a href=#1203 id=1203 data-nosnippet>1203</a>        // produce the SAME Arrow type (run-end-encoded) or concatenation fails:
+<a href=#1204 id=1204 data-nosnippet>1204</a>        //   - constant: first_row_id set, no physical column -&gt; derived constant
+<a href=#1205 id=1205 data-nosnippet>1205</a>        //   - null gate: no first_row_id -&gt; null column
+<a href=#1206 id=1206 data-nosnippet>1206</a>        //   - coalesce: first_row_id set, physical column present -&gt; per-row + fallback
+<a href=#1207 id=1207 data-nosnippet>1207</a>        </span><span class="kw">let </span>constant = last_updated_seq_task(
+<a href=#1208 id=1208 data-nosnippet>1208</a>            write_plain_parquet(dir, <span class="string">"constant.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
+<a href=#1209 id=1209 data-nosnippet>1209</a>            <span class="prelude-val">Some</span>(<span class="number">42</span>),
+<a href=#1210 id=1210 data-nosnippet>1210</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
+<a href=#1211 id=1211 data-nosnippet>1211</a>        );
+<a href=#1212 id=1212 data-nosnippet>1212</a>        <span class="kw">let </span>nulled = last_updated_seq_task(
+<a href=#1213 id=1213 data-nosnippet>1213</a>            write_plain_parquet(dir, <span class="string">"nulled.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
+<a href=#1214 id=1214 data-nosnippet>1214</a>            <span class="prelude-val">None</span>,
+<a href=#1215 id=1215 data-nosnippet>1215</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
+<a href=#1216 id=1216 data-nosnippet>1216</a>        );
+<a href=#1217 id=1217 data-nosnippet>1217</a>        <span class="kw">let </span>coalesced = last_updated_seq_task(
+<a href=#1218 id=1218 data-nosnippet>1218</a>            write_plain_parquet(
+<a href=#1219 id=1219 data-nosnippet>1219</a>                dir,
+<a href=#1220 id=1220 data-nosnippet>1220</a>                <span class="string">"coalesced.parquet"</span>,
+<a href=#1221 id=1221 data-nosnippet>1221</a>                <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#1222 id=1222 data-nosnippet>1222</a>                <span class="macro">vec!</span>[Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef],
+<a href=#1223 id=1223 data-nosnippet>1223</a>            ),
+<a href=#1224 id=1224 data-nosnippet>1224</a>            <span class="prelude-val">Some</span>(<span class="number">50</span>),
+<a href=#1225 id=1225 data-nosnippet>1225</a>            <span class="prelude-val">Some</span>(<span class="number">7</span>),
+<a href=#1226 id=1226 data-nosnippet>1226</a>        );
+<a href=#1227 id=1227 data-nosnippet>1227</a>
+<a href=#1228 id=1228 data-nosnippet>1228</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1229 id=1229 data-nosnippet>1229</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[
+<a href=#1230 id=1230 data-nosnippet>1230</a>            <span class="prelude-val">Ok</span>(constant),
+<a href=#1231 id=1231 data-nosnippet>1231</a>            <span class="prelude-val">Ok</span>(nulled),
+<a href=#1232 id=1232 data-nosnippet>1232</a>            <span class="prelude-val">Ok</span>(coalesced),
+<a href=#1233 id=1233 data-nosnippet>1233</a>        ])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1234 id=1234 data-nosnippet>1234</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1235 id=1235 data-nosnippet>1235</a>            .read(tasks)
+<a href=#1236 id=1236 data-nosnippet>1236</a>            .unwrap()
+<a href=#1237 id=1237 data-nosnippet>1237</a>            .stream()
+<a href=#1238 id=1238 data-nosnippet>1238</a>            .try_collect()
+<a href=#1239 id=1239 data-nosnippet>1239</a>            .<span class="kw">await
+<a href=#1240 id=1240 data-nosnippet>1240</a>            </span>.unwrap();
+<a href=#1241 id=1241 data-nosnippet>1241</a>
+<a href=#1242 id=1242 data-nosnippet>1242</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">3</span>);
+<a href=#1243 id=1243 data-nosnippet>1243</a>        <span class="comment">// Identical schema across all three paths -&gt; concat succeeds.
+<a href=#1244 id=1244 data-nosnippet>1244</a>        </span><span class="kw">let </span>schema = batches[<span class="number">0</span>].schema();
+<a href=#1245 id=1245 data-nosnippet>1245</a>        concat_batches(<span class="kw-2">&amp;</span>schema, <span class="kw-2">&amp;</span>batches)
+<a href=#1246 id=1246 data-nosnippet>1246</a>            .expect(<span class="string">"constant, null and coalesce files must share one column type"</span>);
+<a href=#1247 id=1247 data-nosnippet>1247</a>    }
+<a href=#1248 id=1248 data-nosnippet>1248</a>
+<a href=#1249 id=1249 data-nosnippet>1249</a>    <span class="doccomment">/// A parquet field carrying the embedded `_last_updated_sequence_number` field id.
+<a href=#1250 id=1250 data-nosnippet>1250</a>    </span><span class="kw">fn </span>physical_last_updated_seq_field() -&gt; Field {
+<a href=#1251 id=1251 data-nosnippet>1251</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#1252 id=1252 data-nosnippet>1252</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1253 id=1253 data-nosnippet>1253</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1254 id=1254 data-nosnippet>1254</a>        };
+<a href=#1255 id=1255 data-nosnippet>1255</a>        Field::new(
+<a href=#1256 id=1256 data-nosnippet>1256</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1257 id=1257 data-nosnippet>1257</a>            DataType::Int64,
+<a href=#1258 id=1258 data-nosnippet>1258</a>            <span class="bool-val">true</span>,
+<a href=#1259 id=1259 data-nosnippet>1259</a>        )
+<a href=#1260 id=1260 data-nosnippet>1260</a>        .with_metadata(HashMap::from([(
+<a href=#1261 id=1261 data-nosnippet>1261</a>            PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#1262 id=1262 data-nosnippet>1262</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER.to_string(),
+<a href=#1263 id=1263 data-nosnippet>1263</a>        )]))
+<a href=#1264 id=1264 data-nosnippet>1264</a>    }
+<a href=#1265 id=1265 data-nosnippet>1265</a>
+<a href=#1266 id=1266 data-nosnippet>1266</a>    <span class="attr">#[tokio::test]
+<a href=#1267 id=1267 data-nosnippet>1267</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_coalesced() {
+<a href=#1268 id=1268 data-nosnippet>1268</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1269 id=1269 data-nosnippet>1269</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1270 id=1270 data-nosnippet>1270</a>        <span class="comment">// A file that physically carries the column, as Iceberg Java writes when
+<a href=#1271 id=1271 data-nosnippet>1271</a>        // carrying rows forward across a rewrite: some rows have a stored value, some
+<a href=#1272 id=1272 data-nosnippet>1272</a>        // are null (added/modified rows, inherited on read).
+<a href=#1273 id=1273 data-nosnippet>1273</a>        </span><span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
+<a href=#1274 id=1274 data-nosnippet>1274</a>        <span class="kw">let </span>file_path = write_plain_parquet(
+<a href=#1275 id=1275 data-nosnippet>1275</a>            dir,
+<a href=#1276 id=1276 data-nosnippet>1276</a>            <span class="string">"with_seq.parquet"</span>,
+<a href=#1277 id=1277 data-nosnippet>1277</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#1278 id=1278 data-nosnippet>1278</a>            <span class="macro">vec!</span>[seq_col],
+<a href=#1279 id=1279 data-nosnippet>1279</a>        );
+<a href=#1280 id=1280 data-nosnippet>1280</a>
+<a href=#1281 id=1281 data-nosnippet>1281</a>        <span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>));
+<a href=#1282 id=1282 data-nosnippet>1282</a>
+<a href=#1283 id=1283 data-nosnippet>1283</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1284 id=1284 data-nosnippet>1284</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1285 id=1285 data-nosnippet>1285</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1286 id=1286 data-nosnippet>1286</a>            .read(tasks)
+<a href=#1287 id=1287 data-nosnippet>1287</a>            .unwrap()
+<a href=#1288 id=1288 data-nosnippet>1288</a>            .stream()
+<a href=#1289 id=1289 data-nosnippet>1289</a>            .try_collect()
+<a href=#1290 id=1290 data-nosnippet>1290</a>            .<span class="kw">await
+<a href=#1291 id=1291 data-nosnippet>1291</a>            </span>.unwrap();
+<a href=#1292 id=1292 data-nosnippet>1292</a>
+<a href=#1293 id=1293 data-nosnippet>1293</a>        <span class="comment">// Per-row value where non-null; the data sequence number (9) where null.
+<a href=#1294 id=1294 data-nosnippet>1294</a>        </span>assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>), <span class="prelude-val">Some</span>(<span class="number">8</span>)]);
+<a href=#1295 id=1295 data-nosnippet>1295</a>    }
+<a href=#1296 id=1296 data-nosnippet>1296</a>
+<a href=#1297 id=1297 data-nosnippet>1297</a>    <span class="attr">#[tokio::test]
+<a href=#1298 id=1298 data-nosnippet>1298</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_coalesced_with_pos_column() {
+<a href=#1299 id=1299 data-nosnippet>1299</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#1300 id=1300 data-nosnippet>1300</a>            RESERVED_COL_NAME_POS, RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1301 id=1301 data-nosnippet>1301</a>            RESERVED_FIELD_ID_POS,
+<a href=#1302 id=1302 data-nosnippet>1302</a>        };
+<a href=#1303 id=1303 data-nosnippet>1303</a>
+<a href=#1304 id=1304 data-nosnippet>1304</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1305 id=1305 data-nosnippet>1305</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1306 id=1306 data-nosnippet>1306</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
+<a href=#1307 id=1307 data-nosnippet>1307</a>        <span class="kw">let </span>file_path = write_plain_parquet(
+<a href=#1308 id=1308 data-nosnippet>1308</a>            dir,
+<a href=#1309 id=1309 data-nosnippet>1309</a>            <span class="string">"with_seq_and_pos.parquet"</span>,
+<a href=#1310 id=1310 data-nosnippet>1310</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#1311 id=1311 data-nosnippet>1311</a>            <span class="macro">vec!</span>[seq_col],
+<a href=#1312 id=1312 data-nosnippet>1312</a>        );
+<a href=#1313 id=1313 data-nosnippet>1313</a>
+<a href=#1314 id=1314 data-nosnippet>1314</a>        <span class="comment">// Co-project `_pos` (a virtual column appended to the Arrow output schema) with the
+<a href=#1315 id=1315 data-nosnippet>1315</a>        // physical coalesce column. This guards that the physical column's index is
+<a href=#1316 id=1316 data-nosnippet>1316</a>        // resolved in the Parquet schema, not the Arrow schema (whose indices shift once
+<a href=#1317 id=1317 data-nosnippet>1317</a>        // virtual columns are appended).
+<a href=#1318 id=1318 data-nosnippet>1318</a>        </span><span class="kw">let </span>schema = Arc::new(
+<a href=#1319 id=1319 data-nosnippet>1319</a>            Schema::builder()
+<a href=#1320 id=1320 data-nosnippet>1320</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1321 id=1321 data-nosnippet>1321</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1322 id=1322 data-nosnippet>1322</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#1323 id=1323 data-nosnippet>1323</a>                ])
+<a href=#1324 id=1324 data-nosnippet>1324</a>                .build()
+<a href=#1325 id=1325 data-nosnippet>1325</a>                .unwrap(),
+<a href=#1326 id=1326 data-nosnippet>1326</a>        );
+<a href=#1327 id=1327 data-nosnippet>1327</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#1328 id=1328 data-nosnippet>1328</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#1329 id=1329 data-nosnippet>1329</a>            .with_start(<span class="number">0</span>)
+<a href=#1330 id=1330 data-nosnippet>1330</a>            .with_length(<span class="number">0</span>)
+<a href=#1331 id=1331 data-nosnippet>1331</a>            .with_data_file_path(file_path)
+<a href=#1332 id=1332 data-nosnippet>1332</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1333 id=1333 data-nosnippet>1333</a>            .with_schema(schema)
+<a href=#1334 id=1334 data-nosnippet>1334</a>            .with_project_field_ids(<span class="macro">vec!</span>[
+<a href=#1335 id=1335 data-nosnippet>1335</a>                <span class="number">1</span>,
+<a href=#1336 id=1336 data-nosnippet>1336</a>                RESERVED_FIELD_ID_POS,
+<a href=#1337 id=1337 data-nosnippet>1337</a>                RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1338 id=1338 data-nosnippet>1338</a>            ])
+<a href=#1339 id=1339 data-nosnippet>1339</a>            .with_first_row_id(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#1340 id=1340 data-nosnippet>1340</a>            .with_data_sequence_number(<span class="prelude-val">Some</span>(<span class="number">9</span>))
+<a href=#1341 id=1341 data-nosnippet>1341</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1342 id=1342 data-nosnippet>1342</a>            .build();
+<a href=#1343 id=1343 data-nosnippet>1343</a>
+<a href=#1344 id=1344 data-nosnippet>1344</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1345 id=1345 data-nosnippet>1345</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1346 id=1346 data-nosnippet>1346</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1347 id=1347 data-nosnippet>1347</a>            .read(tasks)
+<a href=#1348 id=1348 data-nosnippet>1348</a>            .unwrap()
+<a href=#1349 id=1349 data-nosnippet>1349</a>            .stream()
+<a href=#1350 id=1350 data-nosnippet>1350</a>            .try_collect()
+<a href=#1351 id=1351 data-nosnippet>1351</a>            .<span class="kw">await
+<a href=#1352 id=1352 data-nosnippet>1352</a>            </span>.unwrap();
 <a href=#1353 id=1353 data-nosnippet>1353</a>
-<a href=#1354 id=1354 data-nosnippet>1354</a>        <span class="comment">// Null first_row_id: the whole column is null even though the file physically
-<a href=#1355 id=1355 data-nosnippet>1355</a>        // carries per-row values -- the gate wins, and the physical column is not read.
-<a href=#1356 id=1356 data-nosnippet>1356</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">9</span>));
-<a href=#1357 id=1357 data-nosnippet>1357</a>
-<a href=#1358 id=1358 data-nosnippet>1358</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1359 id=1359 data-nosnippet>1359</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1360 id=1360 data-nosnippet>1360</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
-<a href=#1361 id=1361 data-nosnippet>1361</a>            .read(tasks)
-<a href=#1362 id=1362 data-nosnippet>1362</a>            .unwrap()
-<a href=#1363 id=1363 data-nosnippet>1363</a>            .stream()
-<a href=#1364 id=1364 data-nosnippet>1364</a>            .try_collect()
-<a href=#1365 id=1365 data-nosnippet>1365</a>            .<span class="kw">await
-<a href=#1366 id=1366 data-nosnippet>1366</a>            </span>.unwrap();
-<a href=#1367 id=1367 data-nosnippet>1367</a>
-<a href=#1368 id=1368 data-nosnippet>1368</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">None</span>, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>]);
-<a href=#1369 id=1369 data-nosnippet>1369</a>    }
-<a href=#1370 id=1370 data-nosnippet>1370</a>
-<a href=#1371 id=1371 data-nosnippet>1371</a>    <span class="attr">#[tokio::test]
-<a href=#1372 id=1372 data-nosnippet>1372</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_present_by_name_without_id_unsupported() {
-<a href=#1373 id=1373 data-nosnippet>1373</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1374 id=1374 data-nosnippet>1374</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1375 id=1375 data-nosnippet>1375</a>        <span class="comment">// Column present by name but WITHOUT the embedded field id (e.g. name mapping /
-<a href=#1376 id=1376 data-nosnippet>1376</a>        // positional fallback). The transformer keys the source column by field id, so
-<a href=#1377 id=1377 data-nosnippet>1377</a>        // this shape can't be threaded and is rejected loudly.
-<a href=#1378 id=1378 data-nosnippet>1378</a>        </span><span class="kw">let </span>seq_field = Field::new(
-<a href=#1379 id=1379 data-nosnippet>1379</a>            <span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
-<a href=#1380 id=1380 data-nosnippet>1380</a>            DataType::Int64,
-<a href=#1381 id=1381 data-nosnippet>1381</a>            <span class="bool-val">true</span>,
-<a href=#1382 id=1382 data-nosnippet>1382</a>        );
-<a href=#1383 id=1383 data-nosnippet>1383</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
-<a href=#1384 id=1384 data-nosnippet>1384</a>        <span class="kw">let </span>file_path =
-<a href=#1385 id=1385 data-nosnippet>1385</a>            write_plain_parquet(dir, <span class="string">"with_seq_by_name.parquet"</span>, <span class="macro">vec!</span>[seq_field], <span class="macro">vec!</span>[
-<a href=#1386 id=1386 data-nosnippet>1386</a>                seq_col,
-<a href=#1387 id=1387 data-nosnippet>1387</a>            ]);
-<a href=#1388 id=1388 data-nosnippet>1388</a>
-<a href=#1389 id=1389 data-nosnippet>1389</a>        <span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>));
-<a href=#1390 id=1390 data-nosnippet>1390</a>
-<a href=#1391 id=1391 data-nosnippet>1391</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1392 id=1392 data-nosnippet>1392</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1393 id=1393 data-nosnippet>1393</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
-<a href=#1394 id=1394 data-nosnippet>1394</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
-<a href=#1395 id=1395 data-nosnippet>1395</a>
-<a href=#1396 id=1396 data-nosnippet>1396</a>        <span class="kw">let </span>err = result.unwrap_err();
-<a href=#1397 id=1397 data-nosnippet>1397</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::FeatureUnsupported);
-<a href=#1398 id=1398 data-nosnippet>1398</a>        <span class="macro">assert!</span>(
-<a href=#1399 id=1399 data-nosnippet>1399</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"without an embedded field id"</span>),
-<a href=#1400 id=1400 data-nosnippet>1400</a>            <span class="string">"unexpected error: {err}"
-<a href=#1401 id=1401 data-nosnippet>1401</a>        </span>);
-<a href=#1402 id=1402 data-nosnippet>1402</a>    }
-<a href=#1403 id=1403 data-nosnippet>1403</a>
-<a href=#1404 id=1404 data-nosnippet>1404</a>    <span class="attr">#[tokio::test]
-<a href=#1405 id=1405 data-nosnippet>1405</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_first_row_id_without_data_seq() {
-<a href=#1406 id=1406 data-nosnippet>1406</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1407 id=1407 data-nosnippet>1407</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
-<a href=#1408 id=1408 data-nosnippet>1408</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
-<a href=#1409 id=1409 data-nosnippet>1409</a>        <span class="kw">let </span>file_path = write_plain_parquet(
-<a href=#1410 id=1410 data-nosnippet>1410</a>            dir,
-<a href=#1411 id=1411 data-nosnippet>1411</a>            <span class="string">"with_seq_no_data_seq.parquet"</span>,
-<a href=#1412 id=1412 data-nosnippet>1412</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
-<a href=#1413 id=1413 data-nosnippet>1413</a>            <span class="macro">vec!</span>[seq_col],
-<a href=#1414 id=1414 data-nosnippet>1414</a>        );
-<a href=#1415 id=1415 data-nosnippet>1415</a>
-<a href=#1416 id=1416 data-nosnippet>1416</a>        <span class="comment">// first_row_id set but no data sequence number: after manifest inheritance a
-<a href=#1417 id=1417 data-nosnippet>1417</a>        // committed entry always has one, so this is a malformed manifest, rejected loudly
-<a href=#1418 id=1418 data-nosnippet>1418</a>        // rather than nulled.
-<a href=#1419 id=1419 data-nosnippet>1419</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">None</span>);
-<a href=#1420 id=1420 data-nosnippet>1420</a>
-<a href=#1421 id=1421 data-nosnippet>1421</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
-<a href=#1422 id=1422 data-nosnippet>1422</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1423 id=1423 data-nosnippet>1423</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
-<a href=#1424 id=1424 data-nosnippet>1424</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1354 id=1354 data-nosnippet>1354</a>        <span class="comment">// The seq column still coalesces correctly...
+<a href=#1355 id=1355 data-nosnippet>1355</a>        </span>assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>), <span class="prelude-val">Some</span>(<span class="number">8</span>)]);
+<a href=#1356 id=1356 data-nosnippet>1356</a>        <span class="comment">// ...and `_pos` is the row position, unaffected by the physical-column union.
+<a href=#1357 id=1357 data-nosnippet>1357</a>        </span><span class="kw">let </span>pos_col = batches[<span class="number">0</span>]
+<a href=#1358 id=1358 data-nosnippet>1358</a>            .column_by_name(RESERVED_COL_NAME_POS)
+<a href=#1359 id=1359 data-nosnippet>1359</a>            .expect(<span class="string">"_pos column should be present"</span>)
+<a href=#1360 id=1360 data-nosnippet>1360</a>            .as_primitive::&lt;arrow_array::types::Int64Type&gt;();
+<a href=#1361 id=1361 data-nosnippet>1361</a>        <span class="macro">assert_eq!</span>(pos_col.values(), <span class="kw-2">&amp;</span>[<span class="number">0</span>, <span class="number">1</span>, <span class="number">2</span>]);
+<a href=#1362 id=1362 data-nosnippet>1362</a>    }
+<a href=#1363 id=1363 data-nosnippet>1363</a>
+<a href=#1364 id=1364 data-nosnippet>1364</a>    <span class="attr">#[tokio::test]
+<a href=#1365 id=1365 data-nosnippet>1365</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_nulled_without_first_row_id() {
+<a href=#1366 id=1366 data-nosnippet>1366</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1367 id=1367 data-nosnippet>1367</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1368 id=1368 data-nosnippet>1368</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
+<a href=#1369 id=1369 data-nosnippet>1369</a>        <span class="kw">let </span>file_path = write_plain_parquet(
+<a href=#1370 id=1370 data-nosnippet>1370</a>            dir,
+<a href=#1371 id=1371 data-nosnippet>1371</a>            <span class="string">"with_seq_no_first_row_id.parquet"</span>,
+<a href=#1372 id=1372 data-nosnippet>1372</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#1373 id=1373 data-nosnippet>1373</a>            <span class="macro">vec!</span>[seq_col],
+<a href=#1374 id=1374 data-nosnippet>1374</a>        );
+<a href=#1375 id=1375 data-nosnippet>1375</a>
+<a href=#1376 id=1376 data-nosnippet>1376</a>        <span class="comment">// Null first_row_id: the whole column is null even though the file physically
+<a href=#1377 id=1377 data-nosnippet>1377</a>        // carries per-row values -- the gate wins, and the physical column is not read.
+<a href=#1378 id=1378 data-nosnippet>1378</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">9</span>));
+<a href=#1379 id=1379 data-nosnippet>1379</a>
+<a href=#1380 id=1380 data-nosnippet>1380</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1381 id=1381 data-nosnippet>1381</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1382 id=1382 data-nosnippet>1382</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#1383 id=1383 data-nosnippet>1383</a>            .read(tasks)
+<a href=#1384 id=1384 data-nosnippet>1384</a>            .unwrap()
+<a href=#1385 id=1385 data-nosnippet>1385</a>            .stream()
+<a href=#1386 id=1386 data-nosnippet>1386</a>            .try_collect()
+<a href=#1387 id=1387 data-nosnippet>1387</a>            .<span class="kw">await
+<a href=#1388 id=1388 data-nosnippet>1388</a>            </span>.unwrap();
+<a href=#1389 id=1389 data-nosnippet>1389</a>
+<a href=#1390 id=1390 data-nosnippet>1390</a>        assert_last_updated_seq_column(<span class="kw-2">&amp;</span>batches, <span class="kw-2">&amp;</span>[<span class="prelude-val">None</span>, <span class="prelude-val">None</span>, <span class="prelude-val">None</span>]);
+<a href=#1391 id=1391 data-nosnippet>1391</a>    }
+<a href=#1392 id=1392 data-nosnippet>1392</a>
+<a href=#1393 id=1393 data-nosnippet>1393</a>    <span class="attr">#[tokio::test]
+<a href=#1394 id=1394 data-nosnippet>1394</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_present_by_name_without_id_unsupported() {
+<a href=#1395 id=1395 data-nosnippet>1395</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1396 id=1396 data-nosnippet>1396</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1397 id=1397 data-nosnippet>1397</a>        <span class="comment">// Column present by name but WITHOUT the embedded field id (e.g. name mapping /
+<a href=#1398 id=1398 data-nosnippet>1398</a>        // positional fallback). The transformer keys the source column by field id, so
+<a href=#1399 id=1399 data-nosnippet>1399</a>        // this shape can't be threaded and is rejected loudly.
+<a href=#1400 id=1400 data-nosnippet>1400</a>        </span><span class="kw">let </span>seq_field = Field::new(
+<a href=#1401 id=1401 data-nosnippet>1401</a>            <span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#1402 id=1402 data-nosnippet>1402</a>            DataType::Int64,
+<a href=#1403 id=1403 data-nosnippet>1403</a>            <span class="bool-val">true</span>,
+<a href=#1404 id=1404 data-nosnippet>1404</a>        );
+<a href=#1405 id=1405 data-nosnippet>1405</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
+<a href=#1406 id=1406 data-nosnippet>1406</a>        <span class="kw">let </span>file_path =
+<a href=#1407 id=1407 data-nosnippet>1407</a>            write_plain_parquet(dir, <span class="string">"with_seq_by_name.parquet"</span>, <span class="macro">vec!</span>[seq_field], <span class="macro">vec!</span>[
+<a href=#1408 id=1408 data-nosnippet>1408</a>                seq_col,
+<a href=#1409 id=1409 data-nosnippet>1409</a>            ]);
+<a href=#1410 id=1410 data-nosnippet>1410</a>
+<a href=#1411 id=1411 data-nosnippet>1411</a>        <span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">Some</span>(<span class="number">9</span>));
+<a href=#1412 id=1412 data-nosnippet>1412</a>
+<a href=#1413 id=1413 data-nosnippet>1413</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1414 id=1414 data-nosnippet>1414</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1415 id=1415 data-nosnippet>1415</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
+<a href=#1416 id=1416 data-nosnippet>1416</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1417 id=1417 data-nosnippet>1417</a>
+<a href=#1418 id=1418 data-nosnippet>1418</a>        <span class="kw">let </span>err = result.unwrap_err();
+<a href=#1419 id=1419 data-nosnippet>1419</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::FeatureUnsupported);
+<a href=#1420 id=1420 data-nosnippet>1420</a>        <span class="macro">assert!</span>(
+<a href=#1421 id=1421 data-nosnippet>1421</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"without an embedded field id"</span>),
+<a href=#1422 id=1422 data-nosnippet>1422</a>            <span class="string">"unexpected error: {err}"
+<a href=#1423 id=1423 data-nosnippet>1423</a>        </span>);
+<a href=#1424 id=1424 data-nosnippet>1424</a>    }
 <a href=#1425 id=1425 data-nosnippet>1425</a>
-<a href=#1426 id=1426 data-nosnippet>1426</a>        <span class="kw">let </span>err = result.unwrap_err();
-<a href=#1427 id=1427 data-nosnippet>1427</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::DataInvalid);
-<a href=#1428 id=1428 data-nosnippet>1428</a>        <span class="macro">assert!</span>(
-<a href=#1429 id=1429 data-nosnippet>1429</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"no data sequence number"</span>),
-<a href=#1430 id=1430 data-nosnippet>1430</a>            <span class="string">"unexpected error: {err}"
-<a href=#1431 id=1431 data-nosnippet>1431</a>        </span>);
-<a href=#1432 id=1432 data-nosnippet>1432</a>    }
-<a href=#1433 id=1433 data-nosnippet>1433</a>
-<a href=#1434 id=1434 data-nosnippet>1434</a>    <span class="attr">#[tokio::test]
-<a href=#1435 id=1435 data-nosnippet>1435</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_with_wrong_key_fails() {
-<a href=#1436 id=1436 data-nosnippet>1436</a>        <span class="kw">let </span>encryption_key = <span class="string">b"0123456789abcdef"</span>;
-<a href=#1437 id=1437 data-nosnippet>1437</a>        <span class="kw">let </span>wrong_key = <span class="string">b"fedcba9876543210"</span>;
-<a href=#1438 id=1438 data-nosnippet>1438</a>
-<a href=#1439 id=1439 data-nosnippet>1439</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#1440 id=1440 data-nosnippet>1440</a>            Schema::builder()
-<a href=#1441 id=1441 data-nosnippet>1441</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1442 id=1442 data-nosnippet>1442</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1443 id=1443 data-nosnippet>1443</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#1444 id=1444 data-nosnippet>1444</a>                ])
-<a href=#1445 id=1445 data-nosnippet>1445</a>                .build()
-<a href=#1446 id=1446 data-nosnippet>1446</a>                .unwrap(),
-<a href=#1447 id=1447 data-nosnippet>1447</a>        );
-<a href=#1448 id=1448 data-nosnippet>1448</a>
-<a href=#1449 id=1449 data-nosnippet>1449</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#1450 id=1450 data-nosnippet>1450</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#1451 id=1451 data-nosnippet>1451</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#1452 id=1452 data-nosnippet>1452</a>                <span class="string">"1"</span>.to_string(),
-<a href=#1453 id=1453 data-nosnippet>1453</a>            )])),
-<a href=#1454 id=1454 data-nosnippet>1454</a>        ]));
+<a href=#1426 id=1426 data-nosnippet>1426</a>    <span class="attr">#[tokio::test]
+<a href=#1427 id=1427 data-nosnippet>1427</a>    </span><span class="kw">async fn </span>test_last_updated_sequence_number_physical_column_first_row_id_without_data_seq() {
+<a href=#1428 id=1428 data-nosnippet>1428</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1429 id=1429 data-nosnippet>1429</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#1430 id=1430 data-nosnippet>1430</a>        <span class="kw">let </span>seq_col = Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef;
+<a href=#1431 id=1431 data-nosnippet>1431</a>        <span class="kw">let </span>file_path = write_plain_parquet(
+<a href=#1432 id=1432 data-nosnippet>1432</a>            dir,
+<a href=#1433 id=1433 data-nosnippet>1433</a>            <span class="string">"with_seq_no_data_seq.parquet"</span>,
+<a href=#1434 id=1434 data-nosnippet>1434</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#1435 id=1435 data-nosnippet>1435</a>            <span class="macro">vec!</span>[seq_col],
+<a href=#1436 id=1436 data-nosnippet>1436</a>        );
+<a href=#1437 id=1437 data-nosnippet>1437</a>
+<a href=#1438 id=1438 data-nosnippet>1438</a>        <span class="comment">// first_row_id set but no data sequence number: after manifest inheritance a
+<a href=#1439 id=1439 data-nosnippet>1439</a>        // committed entry always has one, so this is a malformed manifest, rejected loudly
+<a href=#1440 id=1440 data-nosnippet>1440</a>        // rather than nulled.
+<a href=#1441 id=1441 data-nosnippet>1441</a>        </span><span class="kw">let </span>task = last_updated_seq_task(file_path, <span class="prelude-val">Some</span>(<span class="number">100</span>), <span class="prelude-val">None</span>);
+<a href=#1442 id=1442 data-nosnippet>1442</a>
+<a href=#1443 id=1443 data-nosnippet>1443</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#1444 id=1444 data-nosnippet>1444</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1445 id=1445 data-nosnippet>1445</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
+<a href=#1446 id=1446 data-nosnippet>1446</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1447 id=1447 data-nosnippet>1447</a>
+<a href=#1448 id=1448 data-nosnippet>1448</a>        <span class="kw">let </span>err = result.unwrap_err();
+<a href=#1449 id=1449 data-nosnippet>1449</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::DataInvalid);
+<a href=#1450 id=1450 data-nosnippet>1450</a>        <span class="macro">assert!</span>(
+<a href=#1451 id=1451 data-nosnippet>1451</a>            <span class="macro">format!</span>(<span class="string">"{err}"</span>).contains(<span class="string">"no data sequence number"</span>),
+<a href=#1452 id=1452 data-nosnippet>1452</a>            <span class="string">"unexpected error: {err}"
+<a href=#1453 id=1453 data-nosnippet>1453</a>        </span>);
+<a href=#1454 id=1454 data-nosnippet>1454</a>    }
 <a href=#1455 id=1455 data-nosnippet>1455</a>
-<a href=#1456 id=1456 data-nosnippet>1456</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1457 id=1457 data-nosnippet>1457</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1458 id=1458 data-nosnippet>1458</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#1459 id=1459 data-nosnippet>1459</a>
-<a href=#1460 id=1460 data-nosnippet>1460</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>])) <span class="kw">as </span>ArrayRef;
-<a href=#1461 id=1461 data-nosnippet>1461</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
-<a href=#1462 id=1462 data-nosnippet>1462</a>
-<a href=#1463 id=1463 data-nosnippet>1463</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted_wrong_key.parquet"</span>);
-<a href=#1464 id=1464 data-nosnippet>1464</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">None</span>);
-<a href=#1465 id=1465 data-nosnippet>1465</a>
-<a href=#1466 id=1466 data-nosnippet>1466</a>        <span class="kw">let </span>wrong_key_metadata = <span class="kw">crate</span>::encryption::StandardKeyMetadata::try_new(wrong_key)
-<a href=#1467 id=1467 data-nosnippet>1467</a>            .unwrap()
-<a href=#1468 id=1468 data-nosnippet>1468</a>            .encode()
-<a href=#1469 id=1469 data-nosnippet>1469</a>            .unwrap();
+<a href=#1456 id=1456 data-nosnippet>1456</a>    <span class="attr">#[tokio::test]
+<a href=#1457 id=1457 data-nosnippet>1457</a>    </span><span class="kw">async fn </span>test_read_encrypted_parquet_with_wrong_key_fails() {
+<a href=#1458 id=1458 data-nosnippet>1458</a>        <span class="kw">let </span>encryption_key = <span class="string">b"0123456789abcdef"</span>;
+<a href=#1459 id=1459 data-nosnippet>1459</a>        <span class="kw">let </span>wrong_key = <span class="string">b"fedcba9876543210"</span>;
+<a href=#1460 id=1460 data-nosnippet>1460</a>
+<a href=#1461 id=1461 data-nosnippet>1461</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#1462 id=1462 data-nosnippet>1462</a>            Schema::builder()
+<a href=#1463 id=1463 data-nosnippet>1463</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1464 id=1464 data-nosnippet>1464</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1465 id=1465 data-nosnippet>1465</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#1466 id=1466 data-nosnippet>1466</a>                ])
+<a href=#1467 id=1467 data-nosnippet>1467</a>                .build()
+<a href=#1468 id=1468 data-nosnippet>1468</a>                .unwrap(),
+<a href=#1469 id=1469 data-nosnippet>1469</a>        );
 <a href=#1470 id=1470 data-nosnippet>1470</a>
-<a href=#1471 id=1471 data-nosnippet>1471</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#1472 id=1472 data-nosnippet>1472</a>
-<a href=#1473 id=1473 data-nosnippet>1473</a>        <span class="kw">let </span>task = FileScanTask::builder()
-<a href=#1474 id=1474 data-nosnippet>1474</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
-<a href=#1475 id=1475 data-nosnippet>1475</a>            .with_start(<span class="number">0</span>)
-<a href=#1476 id=1476 data-nosnippet>1476</a>            .with_length(<span class="number">0</span>)
-<a href=#1477 id=1477 data-nosnippet>1477</a>            .with_data_file_path(file_path)
-<a href=#1478 id=1478 data-nosnippet>1478</a>            .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1479 id=1479 data-nosnippet>1479</a>            .with_schema(schema)
-<a href=#1480 id=1480 data-nosnippet>1480</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#1481 id=1481 data-nosnippet>1481</a>            .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1482 id=1482 data-nosnippet>1482</a>            .with_key_metadata(<span class="prelude-val">Some</span>(wrong_key_metadata))
-<a href=#1483 id=1483 data-nosnippet>1483</a>            .build();
+<a href=#1471 id=1471 data-nosnippet>1471</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#1472 id=1472 data-nosnippet>1472</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#1473 id=1473 data-nosnippet>1473</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#1474 id=1474 data-nosnippet>1474</a>                <span class="string">"1"</span>.to_string(),
+<a href=#1475 id=1475 data-nosnippet>1475</a>            )])),
+<a href=#1476 id=1476 data-nosnippet>1476</a>        ]));
+<a href=#1477 id=1477 data-nosnippet>1477</a>
+<a href=#1478 id=1478 data-nosnippet>1478</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1479 id=1479 data-nosnippet>1479</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1480 id=1480 data-nosnippet>1480</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#1481 id=1481 data-nosnippet>1481</a>
+<a href=#1482 id=1482 data-nosnippet>1482</a>        <span class="kw">let </span>id_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>, <span class="number">3</span>])) <span class="kw">as </span>ArrayRef;
+<a href=#1483 id=1483 data-nosnippet>1483</a>        <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data]).unwrap();
 <a href=#1484 id=1484 data-nosnippet>1484</a>
-<a href=#1485 id=1485 data-nosnippet>1485</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1486 id=1486 data-nosnippet>1486</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
-<a href=#1487 id=1487 data-nosnippet>1487</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
-<a href=#1488 id=1488 data-nosnippet>1488</a>
-<a href=#1489 id=1489 data-nosnippet>1489</a>        <span class="kw">let </span>err = result.unwrap_err();
-<a href=#1490 id=1490 data-nosnippet>1490</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::Unexpected);
-<a href=#1491 id=1491 data-nosnippet>1491</a>        <span class="kw">let </span>err_str = <span class="macro">format!</span>(<span class="string">"{err}"</span>);
-<a href=#1492 id=1492 data-nosnippet>1492</a>        <span class="macro">assert!</span>(
-<a href=#1493 id=1493 data-nosnippet>1493</a>            err_str.contains(<span class="string">"unable to decrypt parquet footer"</span>),
-<a href=#1494 id=1494 data-nosnippet>1494</a>            <span class="string">"Expected error about decryption failure, got: {err_str}"
-<a href=#1495 id=1495 data-nosnippet>1495</a>        </span>);
-<a href=#1496 id=1496 data-nosnippet>1496</a>    }
-<a href=#1497 id=1497 data-nosnippet>1497</a>
-<a href=#1498 id=1498 data-nosnippet>1498</a>    <span class="doccomment">/// Test that concurrency=1 reads all files correctly and in deterministic order.
-<a href=#1499 id=1499 data-nosnippet>1499</a>    /// This verifies the fast-path optimization for single concurrency.
-<a href=#1500 id=1500 data-nosnippet>1500</a>    </span><span class="attr">#[tokio::test]
-<a href=#1501 id=1501 data-nosnippet>1501</a>    </span><span class="kw">async fn </span>test_read_with_concurrency_one() {
-<a href=#1502 id=1502 data-nosnippet>1502</a>        <span class="kw">use </span>arrow_array::Int32Array;
-<a href=#1503 id=1503 data-nosnippet>1503</a>
-<a href=#1504 id=1504 data-nosnippet>1504</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#1505 id=1505 data-nosnippet>1505</a>            Schema::builder()
-<a href=#1506 id=1506 data-nosnippet>1506</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1507 id=1507 data-nosnippet>1507</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1508 id=1508 data-nosnippet>1508</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#1509 id=1509 data-nosnippet>1509</a>                    NestedField::required(<span class="number">2</span>, <span class="string">"file_num"</span>, Type::Primitive(PrimitiveType::Int))
-<a href=#1510 id=1510 data-nosnippet>1510</a>                        .into(),
-<a href=#1511 id=1511 data-nosnippet>1511</a>                ])
-<a href=#1512 id=1512 data-nosnippet>1512</a>                .build()
-<a href=#1513 id=1513 data-nosnippet>1513</a>                .unwrap(),
-<a href=#1514 id=1514 data-nosnippet>1514</a>        );
-<a href=#1515 id=1515 data-nosnippet>1515</a>
-<a href=#1516 id=1516 data-nosnippet>1516</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#1517 id=1517 data-nosnippet>1517</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#1518 id=1518 data-nosnippet>1518</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#1519 id=1519 data-nosnippet>1519</a>                <span class="string">"1"</span>.to_string(),
-<a href=#1520 id=1520 data-nosnippet>1520</a>            )])),
-<a href=#1521 id=1521 data-nosnippet>1521</a>            Field::new(<span class="string">"file_num"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#1522 id=1522 data-nosnippet>1522</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#1523 id=1523 data-nosnippet>1523</a>                <span class="string">"2"</span>.to_string(),
-<a href=#1524 id=1524 data-nosnippet>1524</a>            )])),
-<a href=#1525 id=1525 data-nosnippet>1525</a>        ]));
-<a href=#1526 id=1526 data-nosnippet>1526</a>
-<a href=#1527 id=1527 data-nosnippet>1527</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1528 id=1528 data-nosnippet>1528</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1529 id=1529 data-nosnippet>1529</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#1530 id=1530 data-nosnippet>1530</a>
-<a href=#1531 id=1531 data-nosnippet>1531</a>        <span class="comment">// Create 3 parquet files with different data
-<a href=#1532 id=1532 data-nosnippet>1532</a>        </span><span class="kw">let </span>props = WriterProperties::builder()
-<a href=#1533 id=1533 data-nosnippet>1533</a>            .set_compression(Compression::SNAPPY)
-<a href=#1534 id=1534 data-nosnippet>1534</a>            .build();
-<a href=#1535 id=1535 data-nosnippet>1535</a>
-<a href=#1536 id=1536 data-nosnippet>1536</a>        <span class="kw">for </span>file_num <span class="kw">in </span><span class="number">0</span>..<span class="number">3 </span>{
-<a href=#1537 id=1537 data-nosnippet>1537</a>            <span class="kw">let </span>id_data = Arc::new(Int32Array::from_iter_values(
-<a href=#1538 id=1538 data-nosnippet>1538</a>                file_num * <span class="number">10</span>..(file_num + <span class="number">1</span>) * <span class="number">10</span>,
-<a href=#1539 id=1539 data-nosnippet>1539</a>            )) <span class="kw">as </span>ArrayRef;
-<a href=#1540 id=1540 data-nosnippet>1540</a>            <span class="kw">let </span>file_num_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[file_num; <span class="number">10</span>])) <span class="kw">as </span>ArrayRef;
-<a href=#1541 id=1541 data-nosnippet>1541</a>
-<a href=#1542 id=1542 data-nosnippet>1542</a>            <span class="kw">let </span>to_write =
-<a href=#1543 id=1543 data-nosnippet>1543</a>                RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data, file_num_data]).unwrap();
-<a href=#1544 id=1544 data-nosnippet>1544</a>
-<a href=#1545 id=1545 data-nosnippet>1545</a>            <span class="kw">let </span>file = File::create(<span class="macro">format!</span>(<span class="string">"{table_location}/file_{file_num}.parquet"</span>)).unwrap();
-<a href=#1546 id=1546 data-nosnippet>1546</a>            <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#1547 id=1547 data-nosnippet>1547</a>                ArrowWriter::try_new(file, to_write.schema(), <span class="prelude-val">Some</span>(props.clone())).unwrap();
-<a href=#1548 id=1548 data-nosnippet>1548</a>            writer.write(<span class="kw-2">&amp;</span>to_write).expect(<span class="string">"Writing batch"</span>);
-<a href=#1549 id=1549 data-nosnippet>1549</a>            writer.close().unwrap();
-<a href=#1550 id=1550 data-nosnippet>1550</a>        }
-<a href=#1551 id=1551 data-nosnippet>1551</a>
-<a href=#1552 id=1552 data-nosnippet>1552</a>        <span class="comment">// Read with concurrency=1 (fast-path)
-<a href=#1553 id=1553 data-nosnippet>1553</a>        </span><span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current())
-<a href=#1554 id=1554 data-nosnippet>1554</a>            .with_data_file_concurrency_limit(<span class="number">1</span>)
-<a href=#1555 id=1555 data-nosnippet>1555</a>            .build();
-<a href=#1556 id=1556 data-nosnippet>1556</a>
-<a href=#1557 id=1557 data-nosnippet>1557</a>        <span class="comment">// Create tasks in a specific order: file_0, file_1, file_2
-<a href=#1558 id=1558 data-nosnippet>1558</a>        </span><span class="kw">let </span>tasks = <span class="macro">vec!</span>[
-<a href=#1559 id=1559 data-nosnippet>1559</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
-<a href=#1560 id=1560 data-nosnippet>1560</a>                .with_file_size_in_bytes(
-<a href=#1561 id=1561 data-nosnippet>1561</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_0.parquet"</span>))
-<a href=#1562 id=1562 data-nosnippet>1562</a>                        .unwrap()
-<a href=#1563 id=1563 data-nosnippet>1563</a>                        .len(),
-<a href=#1564 id=1564 data-nosnippet>1564</a>                )
-<a href=#1565 id=1565 data-nosnippet>1565</a>                .with_start(<span class="number">0</span>)
-<a href=#1566 id=1566 data-nosnippet>1566</a>                .with_length(<span class="number">0</span>)
-<a href=#1567 id=1567 data-nosnippet>1567</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_0.parquet"</span>))
-<a href=#1568 id=1568 data-nosnippet>1568</a>                .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1569 id=1569 data-nosnippet>1569</a>                .with_schema(schema.clone())
-<a href=#1570 id=1570 data-nosnippet>1570</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
-<a href=#1571 id=1571 data-nosnippet>1571</a>                .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1572 id=1572 data-nosnippet>1572</a>                .build()),
-<a href=#1573 id=1573 data-nosnippet>1573</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
-<a href=#1574 id=1574 data-nosnippet>1574</a>                .with_file_size_in_bytes(
-<a href=#1575 id=1575 data-nosnippet>1575</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_1.parquet"</span>))
-<a href=#1576 id=1576 data-nosnippet>1576</a>                        .unwrap()
-<a href=#1577 id=1577 data-nosnippet>1577</a>                        .len(),
-<a href=#1578 id=1578 data-nosnippet>1578</a>                )
-<a href=#1579 id=1579 data-nosnippet>1579</a>                .with_start(<span class="number">0</span>)
-<a href=#1580 id=1580 data-nosnippet>1580</a>                .with_length(<span class="number">0</span>)
-<a href=#1581 id=1581 data-nosnippet>1581</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_1.parquet"</span>))
-<a href=#1582 id=1582 data-nosnippet>1582</a>                .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1583 id=1583 data-nosnippet>1583</a>                .with_schema(schema.clone())
-<a href=#1584 id=1584 data-nosnippet>1584</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
-<a href=#1585 id=1585 data-nosnippet>1585</a>                .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1586 id=1586 data-nosnippet>1586</a>                .build()),
-<a href=#1587 id=1587 data-nosnippet>1587</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
-<a href=#1588 id=1588 data-nosnippet>1588</a>                .with_file_size_in_bytes(
-<a href=#1589 id=1589 data-nosnippet>1589</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_2.parquet"</span>))
-<a href=#1590 id=1590 data-nosnippet>1590</a>                        .unwrap()
-<a href=#1591 id=1591 data-nosnippet>1591</a>                        .len(),
-<a href=#1592 id=1592 data-nosnippet>1592</a>                )
-<a href=#1593 id=1593 data-nosnippet>1593</a>                .with_start(<span class="number">0</span>)
-<a href=#1594 id=1594 data-nosnippet>1594</a>                .with_length(<span class="number">0</span>)
-<a href=#1595 id=1595 data-nosnippet>1595</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_2.parquet"</span>))
-<a href=#1596 id=1596 data-nosnippet>1596</a>                .with_data_file_format(DataFileFormat::Parquet)
-<a href=#1597 id=1597 data-nosnippet>1597</a>                .with_schema(schema.clone())
-<a href=#1598 id=1598 data-nosnippet>1598</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
-<a href=#1599 id=1599 data-nosnippet>1599</a>                .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#1600 id=1600 data-nosnippet>1600</a>                .build()),
-<a href=#1601 id=1601 data-nosnippet>1601</a>        ];
-<a href=#1602 id=1602 data-nosnippet>1602</a>
-<a href=#1603 id=1603 data-nosnippet>1603</a>        <span class="kw">let </span>tasks_stream = Box::pin(futures::stream::iter(tasks)) <span class="kw">as </span>FileScanTaskStream;
-<a href=#1604 id=1604 data-nosnippet>1604</a>
-<a href=#1605 id=1605 data-nosnippet>1605</a>        <span class="kw">let </span>result = reader
-<a href=#1606 id=1606 data-nosnippet>1606</a>            .read(tasks_stream)
-<a href=#1607 id=1607 data-nosnippet>1607</a>            .unwrap()
-<a href=#1608 id=1608 data-nosnippet>1608</a>            .stream()
-<a href=#1609 id=1609 data-nosnippet>1609</a>            .try_collect::&lt;Vec&lt;RecordBatch&gt;&gt;()
-<a href=#1610 id=1610 data-nosnippet>1610</a>            .<span class="kw">await
-<a href=#1611 id=1611 data-nosnippet>1611</a>            </span>.unwrap();
-<a href=#1612 id=1612 data-nosnippet>1612</a>
-<a href=#1613 id=1613 data-nosnippet>1613</a>        <span class="comment">// Verify we got all 30 rows (10 from each file)
-<a href=#1614 id=1614 data-nosnippet>1614</a>        </span><span class="kw">let </span>total_rows: usize = result.iter().map(|b| b.num_rows()).sum();
-<a href=#1615 id=1615 data-nosnippet>1615</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">30</span>, <span class="string">"Should have 30 total rows"</span>);
-<a href=#1616 id=1616 data-nosnippet>1616</a>
-<a href=#1617 id=1617 data-nosnippet>1617</a>        <span class="comment">// Collect all ids and file_nums to verify data
-<a href=#1618 id=1618 data-nosnippet>1618</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>all_ids = Vec::new();
-<a href=#1619 id=1619 data-nosnippet>1619</a>        <span class="kw">let </span><span class="kw-2">mut </span>all_file_nums = Vec::new();
-<a href=#1620 id=1620 data-nosnippet>1620</a>
-<a href=#1621 id=1621 data-nosnippet>1621</a>        <span class="kw">for </span>batch <span class="kw">in </span><span class="kw-2">&amp;</span>result {
-<a href=#1622 id=1622 data-nosnippet>1622</a>            <span class="kw">let </span>id_col = batch
-<a href=#1623 id=1623 data-nosnippet>1623</a>                .column(<span class="number">0</span>)
-<a href=#1624 id=1624 data-nosnippet>1624</a>                .as_primitive::&lt;arrow_array::types::Int32Type&gt;();
-<a href=#1625 id=1625 data-nosnippet>1625</a>            <span class="kw">let </span>file_num_col = batch
-<a href=#1626 id=1626 data-nosnippet>1626</a>                .column(<span class="number">1</span>)
-<a href=#1627 id=1627 data-nosnippet>1627</a>                .as_primitive::&lt;arrow_array::types::Int32Type&gt;();
-<a href=#1628 id=1628 data-nosnippet>1628</a>
-<a href=#1629 id=1629 data-nosnippet>1629</a>            <span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..batch.num_rows() {
-<a href=#1630 id=1630 data-nosnippet>1630</a>                all_ids.push(id_col.value(i));
-<a href=#1631 id=1631 data-nosnippet>1631</a>                all_file_nums.push(file_num_col.value(i));
-<a href=#1632 id=1632 data-nosnippet>1632</a>            }
-<a href=#1633 id=1633 data-nosnippet>1633</a>        }
+<a href=#1485 id=1485 data-nosnippet>1485</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/encrypted_wrong_key.parquet"</span>);
+<a href=#1486 id=1486 data-nosnippet>1486</a>        write_encrypted_parquet(<span class="kw-2">&amp;</span>file_path, <span class="kw-2">&amp;</span>batch, encryption_key, <span class="prelude-val">None</span>);
+<a href=#1487 id=1487 data-nosnippet>1487</a>
+<a href=#1488 id=1488 data-nosnippet>1488</a>        <span class="kw">let </span>wrong_key_metadata = <span class="kw">crate</span>::encryption::StandardKeyMetadata::try_new(wrong_key)
+<a href=#1489 id=1489 data-nosnippet>1489</a>            .unwrap()
+<a href=#1490 id=1490 data-nosnippet>1490</a>            .encode()
+<a href=#1491 id=1491 data-nosnippet>1491</a>            .unwrap();
+<a href=#1492 id=1492 data-nosnippet>1492</a>
+<a href=#1493 id=1493 data-nosnippet>1493</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#1494 id=1494 data-nosnippet>1494</a>
+<a href=#1495 id=1495 data-nosnippet>1495</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#1496 id=1496 data-nosnippet>1496</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#1497 id=1497 data-nosnippet>1497</a>            .with_start(<span class="number">0</span>)
+<a href=#1498 id=1498 data-nosnippet>1498</a>            .with_length(<span class="number">0</span>)
+<a href=#1499 id=1499 data-nosnippet>1499</a>            .with_data_file_path(file_path)
+<a href=#1500 id=1500 data-nosnippet>1500</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1501 id=1501 data-nosnippet>1501</a>            .with_schema(schema)
+<a href=#1502 id=1502 data-nosnippet>1502</a>            .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#1503 id=1503 data-nosnippet>1503</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1504 id=1504 data-nosnippet>1504</a>            .with_key_metadata(<span class="prelude-val">Some</span>(wrong_key_metadata))
+<a href=#1505 id=1505 data-nosnippet>1505</a>            .build();
+<a href=#1506 id=1506 data-nosnippet>1506</a>
+<a href=#1507 id=1507 data-nosnippet>1507</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1508 id=1508 data-nosnippet>1508</a>        <span class="kw">let </span>result: <span class="prelude-ty">Result</span>&lt;Vec&lt;RecordBatch&gt;, <span class="kw">_</span>&gt; =
+<a href=#1509 id=1509 data-nosnippet>1509</a>            reader.read(tasks).unwrap().stream().try_collect().<span class="kw">await</span>;
+<a href=#1510 id=1510 data-nosnippet>1510</a>
+<a href=#1511 id=1511 data-nosnippet>1511</a>        <span class="kw">let </span>err = result.unwrap_err();
+<a href=#1512 id=1512 data-nosnippet>1512</a>        <span class="macro">assert_eq!</span>(err.kind(), <span class="kw">crate</span>::ErrorKind::Unexpected);
+<a href=#1513 id=1513 data-nosnippet>1513</a>        <span class="kw">let </span>err_str = <span class="macro">format!</span>(<span class="string">"{err}"</span>);
+<a href=#1514 id=1514 data-nosnippet>1514</a>        <span class="macro">assert!</span>(
+<a href=#1515 id=1515 data-nosnippet>1515</a>            err_str.contains(<span class="string">"unable to decrypt parquet footer"</span>),
+<a href=#1516 id=1516 data-nosnippet>1516</a>            <span class="string">"Expected error about decryption failure, got: {err_str}"
+<a href=#1517 id=1517 data-nosnippet>1517</a>        </span>);
+<a href=#1518 id=1518 data-nosnippet>1518</a>    }
+<a href=#1519 id=1519 data-nosnippet>1519</a>
+<a href=#1520 id=1520 data-nosnippet>1520</a>    <span class="doccomment">/// Test that concurrency=1 reads all files correctly and in deterministic order.
+<a href=#1521 id=1521 data-nosnippet>1521</a>    /// This verifies the fast-path optimization for single concurrency.
+<a href=#1522 id=1522 data-nosnippet>1522</a>    </span><span class="attr">#[tokio::test]
+<a href=#1523 id=1523 data-nosnippet>1523</a>    </span><span class="kw">async fn </span>test_read_with_concurrency_one() {
+<a href=#1524 id=1524 data-nosnippet>1524</a>        <span class="kw">use </span>arrow_array::Int32Array;
+<a href=#1525 id=1525 data-nosnippet>1525</a>
+<a href=#1526 id=1526 data-nosnippet>1526</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#1527 id=1527 data-nosnippet>1527</a>            Schema::builder()
+<a href=#1528 id=1528 data-nosnippet>1528</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1529 id=1529 data-nosnippet>1529</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1530 id=1530 data-nosnippet>1530</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#1531 id=1531 data-nosnippet>1531</a>                    NestedField::required(<span class="number">2</span>, <span class="string">"file_num"</span>, Type::Primitive(PrimitiveType::Int))
+<a href=#1532 id=1532 data-nosnippet>1532</a>                        .into(),
+<a href=#1533 id=1533 data-nosnippet>1533</a>                ])
+<a href=#1534 id=1534 data-nosnippet>1534</a>                .build()
+<a href=#1535 id=1535 data-nosnippet>1535</a>                .unwrap(),
+<a href=#1536 id=1536 data-nosnippet>1536</a>        );
+<a href=#1537 id=1537 data-nosnippet>1537</a>
+<a href=#1538 id=1538 data-nosnippet>1538</a>        <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#1539 id=1539 data-nosnippet>1539</a>            Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#1540 id=1540 data-nosnippet>1540</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#1541 id=1541 data-nosnippet>1541</a>                <span class="string">"1"</span>.to_string(),
+<a href=#1542 id=1542 data-nosnippet>1542</a>            )])),
+<a href=#1543 id=1543 data-nosnippet>1543</a>            Field::new(<span class="string">"file_num"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#1544 id=1544 data-nosnippet>1544</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#1545 id=1545 data-nosnippet>1545</a>                <span class="string">"2"</span>.to_string(),
+<a href=#1546 id=1546 data-nosnippet>1546</a>            )])),
+<a href=#1547 id=1547 data-nosnippet>1547</a>        ]));
+<a href=#1548 id=1548 data-nosnippet>1548</a>
+<a href=#1549 id=1549 data-nosnippet>1549</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1550 id=1550 data-nosnippet>1550</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1551 id=1551 data-nosnippet>1551</a>        <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#1552 id=1552 data-nosnippet>1552</a>
+<a href=#1553 id=1553 data-nosnippet>1553</a>        <span class="comment">// Create 3 parquet files with different data
+<a href=#1554 id=1554 data-nosnippet>1554</a>        </span><span class="kw">let </span>props = WriterProperties::builder()
+<a href=#1555 id=1555 data-nosnippet>1555</a>            .set_compression(Compression::SNAPPY)
+<a href=#1556 id=1556 data-nosnippet>1556</a>            .build();
+<a href=#1557 id=1557 data-nosnippet>1557</a>
+<a href=#1558 id=1558 data-nosnippet>1558</a>        <span class="kw">for </span>file_num <span class="kw">in </span><span class="number">0</span>..<span class="number">3 </span>{
+<a href=#1559 id=1559 data-nosnippet>1559</a>            <span class="kw">let </span>id_data = Arc::new(Int32Array::from_iter_values(
+<a href=#1560 id=1560 data-nosnippet>1560</a>                file_num * <span class="number">10</span>..(file_num + <span class="number">1</span>) * <span class="number">10</span>,
+<a href=#1561 id=1561 data-nosnippet>1561</a>            )) <span class="kw">as </span>ArrayRef;
+<a href=#1562 id=1562 data-nosnippet>1562</a>            <span class="kw">let </span>file_num_data = Arc::new(Int32Array::from(<span class="macro">vec!</span>[file_num; <span class="number">10</span>])) <span class="kw">as </span>ArrayRef;
+<a href=#1563 id=1563 data-nosnippet>1563</a>
+<a href=#1564 id=1564 data-nosnippet>1564</a>            <span class="kw">let </span>to_write =
+<a href=#1565 id=1565 data-nosnippet>1565</a>                RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[id_data, file_num_data]).unwrap();
+<a href=#1566 id=1566 data-nosnippet>1566</a>
+<a href=#1567 id=1567 data-nosnippet>1567</a>            <span class="kw">let </span>file = File::create(<span class="macro">format!</span>(<span class="string">"{table_location}/file_{file_num}.parquet"</span>)).unwrap();
+<a href=#1568 id=1568 data-nosnippet>1568</a>            <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#1569 id=1569 data-nosnippet>1569</a>                ArrowWriter::try_new(file, to_write.schema(), <span class="prelude-val">Some</span>(props.clone())).unwrap();
+<a href=#1570 id=1570 data-nosnippet>1570</a>            writer.write(<span class="kw-2">&amp;</span>to_write).expect(<span class="string">"Writing batch"</span>);
+<a href=#1571 id=1571 data-nosnippet>1571</a>            writer.close().unwrap();
+<a href=#1572 id=1572 data-nosnippet>1572</a>        }
+<a href=#1573 id=1573 data-nosnippet>1573</a>
+<a href=#1574 id=1574 data-nosnippet>1574</a>        <span class="comment">// Read with concurrency=1 (fast-path)
+<a href=#1575 id=1575 data-nosnippet>1575</a>        </span><span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current())
+<a href=#1576 id=1576 data-nosnippet>1576</a>            .with_data_file_concurrency_limit(<span class="number">1</span>)
+<a href=#1577 id=1577 data-nosnippet>1577</a>            .build();
+<a href=#1578 id=1578 data-nosnippet>1578</a>
+<a href=#1579 id=1579 data-nosnippet>1579</a>        <span class="comment">// Create tasks in a specific order: file_0, file_1, file_2
+<a href=#1580 id=1580 data-nosnippet>1580</a>        </span><span class="kw">let </span>tasks = <span class="macro">vec!</span>[
+<a href=#1581 id=1581 data-nosnippet>1581</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
+<a href=#1582 id=1582 data-nosnippet>1582</a>                .with_file_size_in_bytes(
+<a href=#1583 id=1583 data-nosnippet>1583</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_0.parquet"</span>))
+<a href=#1584 id=1584 data-nosnippet>1584</a>                        .unwrap()
+<a href=#1585 id=1585 data-nosnippet>1585</a>                        .len(),
+<a href=#1586 id=1586 data-nosnippet>1586</a>                )
+<a href=#1587 id=1587 data-nosnippet>1587</a>                .with_start(<span class="number">0</span>)
+<a href=#1588 id=1588 data-nosnippet>1588</a>                .with_length(<span class="number">0</span>)
+<a href=#1589 id=1589 data-nosnippet>1589</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_0.parquet"</span>))
+<a href=#1590 id=1590 data-nosnippet>1590</a>                .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1591 id=1591 data-nosnippet>1591</a>                .with_schema(schema.clone())
+<a href=#1592 id=1592 data-nosnippet>1592</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
+<a href=#1593 id=1593 data-nosnippet>1593</a>                .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1594 id=1594 data-nosnippet>1594</a>                .build()),
+<a href=#1595 id=1595 data-nosnippet>1595</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
+<a href=#1596 id=1596 data-nosnippet>1596</a>                .with_file_size_in_bytes(
+<a href=#1597 id=1597 data-nosnippet>1597</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_1.parquet"</span>))
+<a href=#1598 id=1598 data-nosnippet>1598</a>                        .unwrap()
+<a href=#1599 id=1599 data-nosnippet>1599</a>                        .len(),
+<a href=#1600 id=1600 data-nosnippet>1600</a>                )
+<a href=#1601 id=1601 data-nosnippet>1601</a>                .with_start(<span class="number">0</span>)
+<a href=#1602 id=1602 data-nosnippet>1602</a>                .with_length(<span class="number">0</span>)
+<a href=#1603 id=1603 data-nosnippet>1603</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_1.parquet"</span>))
+<a href=#1604 id=1604 data-nosnippet>1604</a>                .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1605 id=1605 data-nosnippet>1605</a>                .with_schema(schema.clone())
+<a href=#1606 id=1606 data-nosnippet>1606</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
+<a href=#1607 id=1607 data-nosnippet>1607</a>                .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1608 id=1608 data-nosnippet>1608</a>                .build()),
+<a href=#1609 id=1609 data-nosnippet>1609</a>            <span class="prelude-val">Ok</span>(FileScanTask::builder()
+<a href=#1610 id=1610 data-nosnippet>1610</a>                .with_file_size_in_bytes(
+<a href=#1611 id=1611 data-nosnippet>1611</a>                    std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/file_2.parquet"</span>))
+<a href=#1612 id=1612 data-nosnippet>1612</a>                        .unwrap()
+<a href=#1613 id=1613 data-nosnippet>1613</a>                        .len(),
+<a href=#1614 id=1614 data-nosnippet>1614</a>                )
+<a href=#1615 id=1615 data-nosnippet>1615</a>                .with_start(<span class="number">0</span>)
+<a href=#1616 id=1616 data-nosnippet>1616</a>                .with_length(<span class="number">0</span>)
+<a href=#1617 id=1617 data-nosnippet>1617</a>                .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/file_2.parquet"</span>))
+<a href=#1618 id=1618 data-nosnippet>1618</a>                .with_data_file_format(DataFileFormat::Parquet)
+<a href=#1619 id=1619 data-nosnippet>1619</a>                .with_schema(schema.clone())
+<a href=#1620 id=1620 data-nosnippet>1620</a>                .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>])
+<a href=#1621 id=1621 data-nosnippet>1621</a>                .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#1622 id=1622 data-nosnippet>1622</a>                .build()),
+<a href=#1623 id=1623 data-nosnippet>1623</a>        ];
+<a href=#1624 id=1624 data-nosnippet>1624</a>
+<a href=#1625 id=1625 data-nosnippet>1625</a>        <span class="kw">let </span>tasks_stream = Box::pin(futures::stream::iter(tasks)) <span class="kw">as </span>FileScanTaskStream;
+<a href=#1626 id=1626 data-nosnippet>1626</a>
+<a href=#1627 id=1627 data-nosnippet>1627</a>        <span class="kw">let </span>result = reader
+<a href=#1628 id=1628 data-nosnippet>1628</a>            .read(tasks_stream)
+<a href=#1629 id=1629 data-nosnippet>1629</a>            .unwrap()
+<a href=#1630 id=1630 data-nosnippet>1630</a>            .stream()
+<a href=#1631 id=1631 data-nosnippet>1631</a>            .try_collect::&lt;Vec&lt;RecordBatch&gt;&gt;()
+<a href=#1632 id=1632 data-nosnippet>1632</a>            .<span class="kw">await
+<a href=#1633 id=1633 data-nosnippet>1633</a>            </span>.unwrap();
 <a href=#1634 id=1634 data-nosnippet>1634</a>
-<a href=#1635 id=1635 data-nosnippet>1635</a>        <span class="macro">assert_eq!</span>(all_ids.len(), <span class="number">30</span>);
-<a href=#1636 id=1636 data-nosnippet>1636</a>        <span class="macro">assert_eq!</span>(all_file_nums.len(), <span class="number">30</span>);
-<a href=#1637 id=1637 data-nosnippet>1637</a>
-<a href=#1638 id=1638 data-nosnippet>1638</a>        <span class="comment">// With concurrency=1 and sequential processing, files should be processed in order
-<a href=#1639 id=1639 data-nosnippet>1639</a>        // file_0: ids 0-9, file_num=0
-<a href=#1640 id=1640 data-nosnippet>1640</a>        // file_1: ids 10-19, file_num=1
-<a href=#1641 id=1641 data-nosnippet>1641</a>        // file_2: ids 20-29, file_num=2
-<a href=#1642 id=1642 data-nosnippet>1642</a>        </span><span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..<span class="number">10 </span>{
-<a href=#1643 id=1643 data-nosnippet>1643</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">0</span>, <span class="string">"First 10 rows should be from file_0"</span>);
-<a href=#1644 id=1644 data-nosnippet>1644</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 0-9"</span>);
-<a href=#1645 id=1645 data-nosnippet>1645</a>        }
-<a href=#1646 id=1646 data-nosnippet>1646</a>        <span class="kw">for </span>i <span class="kw">in </span><span class="number">10</span>..<span class="number">20 </span>{
-<a href=#1647 id=1647 data-nosnippet>1647</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">1</span>, <span class="string">"Next 10 rows should be from file_1"</span>);
-<a href=#1648 id=1648 data-nosnippet>1648</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 10-19"</span>);
-<a href=#1649 id=1649 data-nosnippet>1649</a>        }
-<a href=#1650 id=1650 data-nosnippet>1650</a>        <span class="kw">for </span>i <span class="kw">in </span><span class="number">20</span>..<span class="number">30 </span>{
-<a href=#1651 id=1651 data-nosnippet>1651</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">2</span>, <span class="string">"Last 10 rows should be from file_2"</span>);
-<a href=#1652 id=1652 data-nosnippet>1652</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 20-29"</span>);
-<a href=#1653 id=1653 data-nosnippet>1653</a>        }
-<a href=#1654 id=1654 data-nosnippet>1654</a>    }
-<a href=#1655 id=1655 data-nosnippet>1655</a>
-<a href=#1656 id=1656 data-nosnippet>1656</a>    <span class="attr">#[tokio::test]
-<a href=#1657 id=1657 data-nosnippet>1657</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_with_field_ids() {
-<a href=#1658 id=1658 data-nosnippet>1658</a>        <span class="kw">let </span>schema = Arc::new(
-<a href=#1659 id=1659 data-nosnippet>1659</a>            Schema::builder()
-<a href=#1660 id=1660 data-nosnippet>1660</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1661 id=1661 data-nosnippet>1661</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1662 id=1662 data-nosnippet>1662</a>                    NestedField::optional(<span class="number">1</span>, <span class="string">"ts"</span>, Type::Primitive(PrimitiveType::Timestamp))
-<a href=#1663 id=1663 data-nosnippet>1663</a>                        .into(),
-<a href=#1664 id=1664 data-nosnippet>1664</a>                    NestedField::required(<span class="number">2</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#1665 id=1665 data-nosnippet>1665</a>                ])
-<a href=#1666 id=1666 data-nosnippet>1666</a>                .build()
-<a href=#1667 id=1667 data-nosnippet>1667</a>                .unwrap(),
-<a href=#1668 id=1668 data-nosnippet>1668</a>        );
-<a href=#1669 id=1669 data-nosnippet>1669</a>
-<a href=#1670 id=1670 data-nosnippet>1670</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1671 id=1671 data-nosnippet>1671</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1672 id=1672 data-nosnippet>1672</a>        <span class="kw">let </span>(file_path, expected_micros) =
-<a href=#1673 id=1673 data-nosnippet>1673</a>            write_int96_parquet_file(<span class="kw-2">&amp;</span>table_location, <span class="string">"with_ids.parquet"</span>, <span class="bool-val">true</span>);
-<a href=#1674 id=1674 data-nosnippet>1674</a>
-<a href=#1675 id=1675 data-nosnippet>1675</a>        assert_int96_read_matches(<span class="kw-2">&amp;</span>file_path, schema, <span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>], <span class="kw-2">&amp;</span>expected_micros).<span class="kw">await</span>;
+<a href=#1635 id=1635 data-nosnippet>1635</a>        <span class="comment">// Verify we got all 30 rows (10 from each file)
+<a href=#1636 id=1636 data-nosnippet>1636</a>        </span><span class="kw">let </span>total_rows: usize = result.iter().map(|b| b.num_rows()).sum();
+<a href=#1637 id=1637 data-nosnippet>1637</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">30</span>, <span class="string">"Should have 30 total rows"</span>);
+<a href=#1638 id=1638 data-nosnippet>1638</a>
+<a href=#1639 id=1639 data-nosnippet>1639</a>        <span class="comment">// Collect all ids and file_nums to verify data
+<a href=#1640 id=1640 data-nosnippet>1640</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>all_ids = Vec::new();
+<a href=#1641 id=1641 data-nosnippet>1641</a>        <span class="kw">let </span><span class="kw-2">mut </span>all_file_nums = Vec::new();
+<a href=#1642 id=1642 data-nosnippet>1642</a>
+<a href=#1643 id=1643 data-nosnippet>1643</a>        <span class="kw">for </span>batch <span class="kw">in </span><span class="kw-2">&amp;</span>result {
+<a href=#1644 id=1644 data-nosnippet>1644</a>            <span class="kw">let </span>id_col = batch
+<a href=#1645 id=1645 data-nosnippet>1645</a>                .column(<span class="number">0</span>)
+<a href=#1646 id=1646 data-nosnippet>1646</a>                .as_primitive::&lt;arrow_array::types::Int32Type&gt;();
+<a href=#1647 id=1647 data-nosnippet>1647</a>            <span class="kw">let </span>file_num_col = batch
+<a href=#1648 id=1648 data-nosnippet>1648</a>                .column(<span class="number">1</span>)
+<a href=#1649 id=1649 data-nosnippet>1649</a>                .as_primitive::&lt;arrow_array::types::Int32Type&gt;();
+<a href=#1650 id=1650 data-nosnippet>1650</a>
+<a href=#1651 id=1651 data-nosnippet>1651</a>            <span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..batch.num_rows() {
+<a href=#1652 id=1652 data-nosnippet>1652</a>                all_ids.push(id_col.value(i));
+<a href=#1653 id=1653 data-nosnippet>1653</a>                all_file_nums.push(file_num_col.value(i));
+<a href=#1654 id=1654 data-nosnippet>1654</a>            }
+<a href=#1655 id=1655 data-nosnippet>1655</a>        }
+<a href=#1656 id=1656 data-nosnippet>1656</a>
+<a href=#1657 id=1657 data-nosnippet>1657</a>        <span class="macro">assert_eq!</span>(all_ids.len(), <span class="number">30</span>);
+<a href=#1658 id=1658 data-nosnippet>1658</a>        <span class="macro">assert_eq!</span>(all_file_nums.len(), <span class="number">30</span>);
+<a href=#1659 id=1659 data-nosnippet>1659</a>
+<a href=#1660 id=1660 data-nosnippet>1660</a>        <span class="comment">// With concurrency=1 and sequential processing, files should be processed in order
+<a href=#1661 id=1661 data-nosnippet>1661</a>        // file_0: ids 0-9, file_num=0
+<a href=#1662 id=1662 data-nosnippet>1662</a>        // file_1: ids 10-19, file_num=1
+<a href=#1663 id=1663 data-nosnippet>1663</a>        // file_2: ids 20-29, file_num=2
+<a href=#1664 id=1664 data-nosnippet>1664</a>        </span><span class="kw">for </span>i <span class="kw">in </span><span class="number">0</span>..<span class="number">10 </span>{
+<a href=#1665 id=1665 data-nosnippet>1665</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">0</span>, <span class="string">"First 10 rows should be from file_0"</span>);
+<a href=#1666 id=1666 data-nosnippet>1666</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 0-9"</span>);
+<a href=#1667 id=1667 data-nosnippet>1667</a>        }
+<a href=#1668 id=1668 data-nosnippet>1668</a>        <span class="kw">for </span>i <span class="kw">in </span><span class="number">10</span>..<span class="number">20 </span>{
+<a href=#1669 id=1669 data-nosnippet>1669</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">1</span>, <span class="string">"Next 10 rows should be from file_1"</span>);
+<a href=#1670 id=1670 data-nosnippet>1670</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 10-19"</span>);
+<a href=#1671 id=1671 data-nosnippet>1671</a>        }
+<a href=#1672 id=1672 data-nosnippet>1672</a>        <span class="kw">for </span>i <span class="kw">in </span><span class="number">20</span>..<span class="number">30 </span>{
+<a href=#1673 id=1673 data-nosnippet>1673</a>            <span class="macro">assert_eq!</span>(all_file_nums[i], <span class="number">2</span>, <span class="string">"Last 10 rows should be from file_2"</span>);
+<a href=#1674 id=1674 data-nosnippet>1674</a>            <span class="macro">assert_eq!</span>(all_ids[i], i <span class="kw">as </span>i32, <span class="string">"IDs should be 20-29"</span>);
+<a href=#1675 id=1675 data-nosnippet>1675</a>        }
 <a href=#1676 id=1676 data-nosnippet>1676</a>    }
 <a href=#1677 id=1677 data-nosnippet>1677</a>
 <a href=#1678 id=1678 data-nosnippet>1678</a>    <span class="attr">#[tokio::test]
-<a href=#1679 id=1679 data-nosnippet>1679</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_without_field_ids() {
+<a href=#1679 id=1679 data-nosnippet>1679</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_with_field_ids() {
 <a href=#1680 id=1680 data-nosnippet>1680</a>        <span class="kw">let </span>schema = Arc::new(
 <a href=#1681 id=1681 data-nosnippet>1681</a>            Schema::builder()
 <a href=#1682 id=1682 data-nosnippet>1682</a>                .with_schema_id(<span class="number">1</span>)
@@ -1692,342 +1692,808 @@
 <a href=#1692 id=1692 data-nosnippet>1692</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
 <a href=#1693 id=1693 data-nosnippet>1693</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
 <a href=#1694 id=1694 data-nosnippet>1694</a>        <span class="kw">let </span>(file_path, expected_micros) =
-<a href=#1695 id=1695 data-nosnippet>1695</a>            write_int96_parquet_file(<span class="kw-2">&amp;</span>table_location, <span class="string">"no_ids.parquet"</span>, <span class="bool-val">false</span>);
+<a href=#1695 id=1695 data-nosnippet>1695</a>            write_int96_parquet_file(<span class="kw-2">&amp;</span>table_location, <span class="string">"with_ids.parquet"</span>, <span class="bool-val">true</span>);
 <a href=#1696 id=1696 data-nosnippet>1696</a>
 <a href=#1697 id=1697 data-nosnippet>1697</a>        assert_int96_read_matches(<span class="kw-2">&amp;</span>file_path, schema, <span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>], <span class="kw-2">&amp;</span>expected_micros).<span class="kw">await</span>;
 <a href=#1698 id=1698 data-nosnippet>1698</a>    }
 <a href=#1699 id=1699 data-nosnippet>1699</a>
 <a href=#1700 id=1700 data-nosnippet>1700</a>    <span class="attr">#[tokio::test]
-<a href=#1701 id=1701 data-nosnippet>1701</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_struct() {
-<a href=#1702 id=1702 data-nosnippet>1702</a>        <span class="kw">use </span>arrow_array::{StructArray, TimestampMicrosecondArray};
-<a href=#1703 id=1703 data-nosnippet>1703</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
-<a href=#1704 id=1704 data-nosnippet>1704</a>        <span class="kw">use </span>parquet::data_type::Int96Type;
-<a href=#1705 id=1705 data-nosnippet>1705</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
-<a href=#1706 id=1706 data-nosnippet>1706</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
-<a href=#1707 id=1707 data-nosnippet>1707</a>
-<a href=#1708 id=1708 data-nosnippet>1708</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1709 id=1709 data-nosnippet>1709</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1710 id=1710 data-nosnippet>1710</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/struct_int96.parquet"</span>);
-<a href=#1711 id=1711 data-nosnippet>1711</a>
-<a href=#1712 id=1712 data-nosnippet>1712</a>        <span class="kw">let </span>ts_type = SchemaType::primitive_type_builder(<span class="string">"ts"</span>, PhysicalType::INT96)
-<a href=#1713 id=1713 data-nosnippet>1713</a>            .with_repetition(Repetition::OPTIONAL)
-<a href=#1714 id=1714 data-nosnippet>1714</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
-<a href=#1715 id=1715 data-nosnippet>1715</a>            .build()
-<a href=#1716 id=1716 data-nosnippet>1716</a>            .unwrap();
-<a href=#1717 id=1717 data-nosnippet>1717</a>
-<a href=#1718 id=1718 data-nosnippet>1718</a>        <span class="kw">let </span>struct_type = SchemaType::group_type_builder(<span class="string">"data"</span>)
-<a href=#1719 id=1719 data-nosnippet>1719</a>            .with_repetition(Repetition::REQUIRED)
-<a href=#1720 id=1720 data-nosnippet>1720</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
-<a href=#1721 id=1721 data-nosnippet>1721</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(ts_type)])
-<a href=#1722 id=1722 data-nosnippet>1722</a>            .build()
-<a href=#1723 id=1723 data-nosnippet>1723</a>            .unwrap();
-<a href=#1724 id=1724 data-nosnippet>1724</a>
-<a href=#1725 id=1725 data-nosnippet>1725</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
-<a href=#1726 id=1726 data-nosnippet>1726</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(struct_type)])
-<a href=#1727 id=1727 data-nosnippet>1727</a>            .build()
-<a href=#1728 id=1728 data-nosnippet>1728</a>            .unwrap();
+<a href=#1701 id=1701 data-nosnippet>1701</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_without_field_ids() {
+<a href=#1702 id=1702 data-nosnippet>1702</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#1703 id=1703 data-nosnippet>1703</a>            Schema::builder()
+<a href=#1704 id=1704 data-nosnippet>1704</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1705 id=1705 data-nosnippet>1705</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1706 id=1706 data-nosnippet>1706</a>                    NestedField::optional(<span class="number">1</span>, <span class="string">"ts"</span>, Type::Primitive(PrimitiveType::Timestamp))
+<a href=#1707 id=1707 data-nosnippet>1707</a>                        .into(),
+<a href=#1708 id=1708 data-nosnippet>1708</a>                    NestedField::required(<span class="number">2</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#1709 id=1709 data-nosnippet>1709</a>                ])
+<a href=#1710 id=1710 data-nosnippet>1710</a>                .build()
+<a href=#1711 id=1711 data-nosnippet>1711</a>                .unwrap(),
+<a href=#1712 id=1712 data-nosnippet>1712</a>        );
+<a href=#1713 id=1713 data-nosnippet>1713</a>
+<a href=#1714 id=1714 data-nosnippet>1714</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1715 id=1715 data-nosnippet>1715</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1716 id=1716 data-nosnippet>1716</a>        <span class="kw">let </span>(file_path, expected_micros) =
+<a href=#1717 id=1717 data-nosnippet>1717</a>            write_int96_parquet_file(<span class="kw-2">&amp;</span>table_location, <span class="string">"no_ids.parquet"</span>, <span class="bool-val">false</span>);
+<a href=#1718 id=1718 data-nosnippet>1718</a>
+<a href=#1719 id=1719 data-nosnippet>1719</a>        assert_int96_read_matches(<span class="kw-2">&amp;</span>file_path, schema, <span class="macro">vec!</span>[<span class="number">1</span>, <span class="number">2</span>], <span class="kw-2">&amp;</span>expected_micros).<span class="kw">await</span>;
+<a href=#1720 id=1720 data-nosnippet>1720</a>    }
+<a href=#1721 id=1721 data-nosnippet>1721</a>
+<a href=#1722 id=1722 data-nosnippet>1722</a>    <span class="attr">#[tokio::test]
+<a href=#1723 id=1723 data-nosnippet>1723</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_struct() {
+<a href=#1724 id=1724 data-nosnippet>1724</a>        <span class="kw">use </span>arrow_array::{StructArray, TimestampMicrosecondArray};
+<a href=#1725 id=1725 data-nosnippet>1725</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
+<a href=#1726 id=1726 data-nosnippet>1726</a>        <span class="kw">use </span>parquet::data_type::Int96Type;
+<a href=#1727 id=1727 data-nosnippet>1727</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
+<a href=#1728 id=1728 data-nosnippet>1728</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
 <a href=#1729 id=1729 data-nosnippet>1729</a>
-<a href=#1730 id=1730 data-nosnippet>1730</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
-<a href=#1731 id=1731 data-nosnippet>1731</a>
-<a href=#1732 id=1732 data-nosnippet>1732</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
-<a href=#1733 id=1733 data-nosnippet>1733</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#1734 id=1734 data-nosnippet>1734</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
-<a href=#1735 id=1735 data-nosnippet>1735</a>
-<a href=#1736 id=1736 data-nosnippet>1736</a>        <span class="comment">// def=1: struct is REQUIRED so no level, ts is OPTIONAL and present (1).
-<a href=#1737 id=1737 data-nosnippet>1737</a>        // No repetition levels needed (no repeated groups).
-<a href=#1738 id=1738 data-nosnippet>1738</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
-<a href=#1739 id=1739 data-nosnippet>1739</a>        {
-<a href=#1740 id=1740 data-nosnippet>1740</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#1741 id=1741 data-nosnippet>1741</a>            col.typed::&lt;Int96Type&gt;()
-<a href=#1742 id=1742 data-nosnippet>1742</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">1</span>]), <span class="prelude-val">None</span>)
-<a href=#1743 id=1743 data-nosnippet>1743</a>                .unwrap();
-<a href=#1744 id=1744 data-nosnippet>1744</a>            col.close().unwrap();
-<a href=#1745 id=1745 data-nosnippet>1745</a>        }
-<a href=#1746 id=1746 data-nosnippet>1746</a>        row_group.close().unwrap();
-<a href=#1747 id=1747 data-nosnippet>1747</a>        writer.close().unwrap();
-<a href=#1748 id=1748 data-nosnippet>1748</a>
-<a href=#1749 id=1749 data-nosnippet>1749</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
-<a href=#1750 id=1750 data-nosnippet>1750</a>            Schema::builder()
-<a href=#1751 id=1751 data-nosnippet>1751</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1752 id=1752 data-nosnippet>1752</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1753 id=1753 data-nosnippet>1753</a>                    NestedField::required(
-<a href=#1754 id=1754 data-nosnippet>1754</a>                        <span class="number">1</span>,
-<a href=#1755 id=1755 data-nosnippet>1755</a>                        <span class="string">"data"</span>,
-<a href=#1756 id=1756 data-nosnippet>1756</a>                        Type::Struct(<span class="kw">crate</span>::spec::StructType::new(<span class="macro">vec!</span>[
-<a href=#1757 id=1757 data-nosnippet>1757</a>                            NestedField::optional(
-<a href=#1758 id=1758 data-nosnippet>1758</a>                                <span class="number">2</span>,
-<a href=#1759 id=1759 data-nosnippet>1759</a>                                <span class="string">"ts"</span>,
-<a href=#1760 id=1760 data-nosnippet>1760</a>                                Type::Primitive(PrimitiveType::Timestamp),
-<a href=#1761 id=1761 data-nosnippet>1761</a>                            )
-<a href=#1762 id=1762 data-nosnippet>1762</a>                            .into(),
-<a href=#1763 id=1763 data-nosnippet>1763</a>                        ])),
-<a href=#1764 id=1764 data-nosnippet>1764</a>                    )
-<a href=#1765 id=1765 data-nosnippet>1765</a>                    .into(),
-<a href=#1766 id=1766 data-nosnippet>1766</a>                ])
-<a href=#1767 id=1767 data-nosnippet>1767</a>                .build()
-<a href=#1768 id=1768 data-nosnippet>1768</a>                .unwrap(),
-<a href=#1769 id=1769 data-nosnippet>1769</a>        );
+<a href=#1730 id=1730 data-nosnippet>1730</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1731 id=1731 data-nosnippet>1731</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1732 id=1732 data-nosnippet>1732</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/struct_int96.parquet"</span>);
+<a href=#1733 id=1733 data-nosnippet>1733</a>
+<a href=#1734 id=1734 data-nosnippet>1734</a>        <span class="kw">let </span>ts_type = SchemaType::primitive_type_builder(<span class="string">"ts"</span>, PhysicalType::INT96)
+<a href=#1735 id=1735 data-nosnippet>1735</a>            .with_repetition(Repetition::OPTIONAL)
+<a href=#1736 id=1736 data-nosnippet>1736</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
+<a href=#1737 id=1737 data-nosnippet>1737</a>            .build()
+<a href=#1738 id=1738 data-nosnippet>1738</a>            .unwrap();
+<a href=#1739 id=1739 data-nosnippet>1739</a>
+<a href=#1740 id=1740 data-nosnippet>1740</a>        <span class="kw">let </span>struct_type = SchemaType::group_type_builder(<span class="string">"data"</span>)
+<a href=#1741 id=1741 data-nosnippet>1741</a>            .with_repetition(Repetition::REQUIRED)
+<a href=#1742 id=1742 data-nosnippet>1742</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
+<a href=#1743 id=1743 data-nosnippet>1743</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(ts_type)])
+<a href=#1744 id=1744 data-nosnippet>1744</a>            .build()
+<a href=#1745 id=1745 data-nosnippet>1745</a>            .unwrap();
+<a href=#1746 id=1746 data-nosnippet>1746</a>
+<a href=#1747 id=1747 data-nosnippet>1747</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
+<a href=#1748 id=1748 data-nosnippet>1748</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(struct_type)])
+<a href=#1749 id=1749 data-nosnippet>1749</a>            .build()
+<a href=#1750 id=1750 data-nosnippet>1750</a>            .unwrap();
+<a href=#1751 id=1751 data-nosnippet>1751</a>
+<a href=#1752 id=1752 data-nosnippet>1752</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
+<a href=#1753 id=1753 data-nosnippet>1753</a>
+<a href=#1754 id=1754 data-nosnippet>1754</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
+<a href=#1755 id=1755 data-nosnippet>1755</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#1756 id=1756 data-nosnippet>1756</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
+<a href=#1757 id=1757 data-nosnippet>1757</a>
+<a href=#1758 id=1758 data-nosnippet>1758</a>        <span class="comment">// def=1: struct is REQUIRED so no level, ts is OPTIONAL and present (1).
+<a href=#1759 id=1759 data-nosnippet>1759</a>        // No repetition levels needed (no repeated groups).
+<a href=#1760 id=1760 data-nosnippet>1760</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
+<a href=#1761 id=1761 data-nosnippet>1761</a>        {
+<a href=#1762 id=1762 data-nosnippet>1762</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#1763 id=1763 data-nosnippet>1763</a>            col.typed::&lt;Int96Type&gt;()
+<a href=#1764 id=1764 data-nosnippet>1764</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">1</span>]), <span class="prelude-val">None</span>)
+<a href=#1765 id=1765 data-nosnippet>1765</a>                .unwrap();
+<a href=#1766 id=1766 data-nosnippet>1766</a>            col.close().unwrap();
+<a href=#1767 id=1767 data-nosnippet>1767</a>        }
+<a href=#1768 id=1768 data-nosnippet>1768</a>        row_group.close().unwrap();
+<a href=#1769 id=1769 data-nosnippet>1769</a>        writer.close().unwrap();
 <a href=#1770 id=1770 data-nosnippet>1770</a>
-<a href=#1771 id=1771 data-nosnippet>1771</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
-<a href=#1772 id=1772 data-nosnippet>1772</a>
-<a href=#1773 id=1773 data-nosnippet>1773</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
-<a href=#1774 id=1774 data-nosnippet>1774</a>        <span class="kw">let </span>struct_array = batches[<span class="number">0</span>]
-<a href=#1775 id=1775 data-nosnippet>1775</a>            .column(<span class="number">0</span>)
-<a href=#1776 id=1776 data-nosnippet>1776</a>            .as_any()
-<a href=#1777 id=1777 data-nosnippet>1777</a>            .downcast_ref::&lt;StructArray&gt;()
-<a href=#1778 id=1778 data-nosnippet>1778</a>            .expect(<span class="string">"Expected StructArray"</span>);
-<a href=#1779 id=1779 data-nosnippet>1779</a>        <span class="kw">let </span>ts_array = struct_array
-<a href=#1780 id=1780 data-nosnippet>1780</a>            .column(<span class="number">0</span>)
-<a href=#1781 id=1781 data-nosnippet>1781</a>            .as_any()
-<a href=#1782 id=1782 data-nosnippet>1782</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
-<a href=#1783 id=1783 data-nosnippet>1783</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray inside struct"</span>);
-<a href=#1784 id=1784 data-nosnippet>1784</a>
-<a href=#1785 id=1785 data-nosnippet>1785</a>        <span class="macro">assert_eq!</span>(
-<a href=#1786 id=1786 data-nosnippet>1786</a>            ts_array.value(<span class="number">0</span>),
-<a href=#1787 id=1787 data-nosnippet>1787</a>            expected_micros,
-<a href=#1788 id=1788 data-nosnippet>1788</a>            <span class="string">"INT96 in struct: got {}, expected {expected_micros}"</span>,
-<a href=#1789 id=1789 data-nosnippet>1789</a>            ts_array.value(<span class="number">0</span>)
-<a href=#1790 id=1790 data-nosnippet>1790</a>        );
-<a href=#1791 id=1791 data-nosnippet>1791</a>    }
+<a href=#1771 id=1771 data-nosnippet>1771</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
+<a href=#1772 id=1772 data-nosnippet>1772</a>            Schema::builder()
+<a href=#1773 id=1773 data-nosnippet>1773</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1774 id=1774 data-nosnippet>1774</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1775 id=1775 data-nosnippet>1775</a>                    NestedField::required(
+<a href=#1776 id=1776 data-nosnippet>1776</a>                        <span class="number">1</span>,
+<a href=#1777 id=1777 data-nosnippet>1777</a>                        <span class="string">"data"</span>,
+<a href=#1778 id=1778 data-nosnippet>1778</a>                        Type::Struct(<span class="kw">crate</span>::spec::StructType::new(<span class="macro">vec!</span>[
+<a href=#1779 id=1779 data-nosnippet>1779</a>                            NestedField::optional(
+<a href=#1780 id=1780 data-nosnippet>1780</a>                                <span class="number">2</span>,
+<a href=#1781 id=1781 data-nosnippet>1781</a>                                <span class="string">"ts"</span>,
+<a href=#1782 id=1782 data-nosnippet>1782</a>                                Type::Primitive(PrimitiveType::Timestamp),
+<a href=#1783 id=1783 data-nosnippet>1783</a>                            )
+<a href=#1784 id=1784 data-nosnippet>1784</a>                            .into(),
+<a href=#1785 id=1785 data-nosnippet>1785</a>                        ])),
+<a href=#1786 id=1786 data-nosnippet>1786</a>                    )
+<a href=#1787 id=1787 data-nosnippet>1787</a>                    .into(),
+<a href=#1788 id=1788 data-nosnippet>1788</a>                ])
+<a href=#1789 id=1789 data-nosnippet>1789</a>                .build()
+<a href=#1790 id=1790 data-nosnippet>1790</a>                .unwrap(),
+<a href=#1791 id=1791 data-nosnippet>1791</a>        );
 <a href=#1792 id=1792 data-nosnippet>1792</a>
-<a href=#1793 id=1793 data-nosnippet>1793</a>    <span class="attr">#[tokio::test]
-<a href=#1794 id=1794 data-nosnippet>1794</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_list() {
-<a href=#1795 id=1795 data-nosnippet>1795</a>        <span class="kw">use </span>arrow_array::{ListArray, TimestampMicrosecondArray};
-<a href=#1796 id=1796 data-nosnippet>1796</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
-<a href=#1797 id=1797 data-nosnippet>1797</a>        <span class="kw">use </span>parquet::data_type::Int96Type;
-<a href=#1798 id=1798 data-nosnippet>1798</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
-<a href=#1799 id=1799 data-nosnippet>1799</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
-<a href=#1800 id=1800 data-nosnippet>1800</a>
-<a href=#1801 id=1801 data-nosnippet>1801</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1802 id=1802 data-nosnippet>1802</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1803 id=1803 data-nosnippet>1803</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/list_int96.parquet"</span>);
-<a href=#1804 id=1804 data-nosnippet>1804</a>
-<a href=#1805 id=1805 data-nosnippet>1805</a>        <span class="comment">// 3-level LIST encoding:
-<a href=#1806 id=1806 data-nosnippet>1806</a>        //   optional group timestamps (LIST) {
-<a href=#1807 id=1807 data-nosnippet>1807</a>        //     repeated group list {
-<a href=#1808 id=1808 data-nosnippet>1808</a>        //       optional int96 element;
-<a href=#1809 id=1809 data-nosnippet>1809</a>        //     }
-<a href=#1810 id=1810 data-nosnippet>1810</a>        //   }
-<a href=#1811 id=1811 data-nosnippet>1811</a>        </span><span class="kw">let </span>element_type = SchemaType::primitive_type_builder(<span class="string">"element"</span>, PhysicalType::INT96)
-<a href=#1812 id=1812 data-nosnippet>1812</a>            .with_repetition(Repetition::OPTIONAL)
-<a href=#1813 id=1813 data-nosnippet>1813</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
-<a href=#1814 id=1814 data-nosnippet>1814</a>            .build()
-<a href=#1815 id=1815 data-nosnippet>1815</a>            .unwrap();
-<a href=#1816 id=1816 data-nosnippet>1816</a>
-<a href=#1817 id=1817 data-nosnippet>1817</a>        <span class="kw">let </span>list_group = SchemaType::group_type_builder(<span class="string">"list"</span>)
-<a href=#1818 id=1818 data-nosnippet>1818</a>            .with_repetition(Repetition::REPEATED)
-<a href=#1819 id=1819 data-nosnippet>1819</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(element_type)])
-<a href=#1820 id=1820 data-nosnippet>1820</a>            .build()
-<a href=#1821 id=1821 data-nosnippet>1821</a>            .unwrap();
+<a href=#1793 id=1793 data-nosnippet>1793</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
+<a href=#1794 id=1794 data-nosnippet>1794</a>
+<a href=#1795 id=1795 data-nosnippet>1795</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
+<a href=#1796 id=1796 data-nosnippet>1796</a>        <span class="kw">let </span>struct_array = batches[<span class="number">0</span>]
+<a href=#1797 id=1797 data-nosnippet>1797</a>            .column(<span class="number">0</span>)
+<a href=#1798 id=1798 data-nosnippet>1798</a>            .as_any()
+<a href=#1799 id=1799 data-nosnippet>1799</a>            .downcast_ref::&lt;StructArray&gt;()
+<a href=#1800 id=1800 data-nosnippet>1800</a>            .expect(<span class="string">"Expected StructArray"</span>);
+<a href=#1801 id=1801 data-nosnippet>1801</a>        <span class="kw">let </span>ts_array = struct_array
+<a href=#1802 id=1802 data-nosnippet>1802</a>            .column(<span class="number">0</span>)
+<a href=#1803 id=1803 data-nosnippet>1803</a>            .as_any()
+<a href=#1804 id=1804 data-nosnippet>1804</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
+<a href=#1805 id=1805 data-nosnippet>1805</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray inside struct"</span>);
+<a href=#1806 id=1806 data-nosnippet>1806</a>
+<a href=#1807 id=1807 data-nosnippet>1807</a>        <span class="macro">assert_eq!</span>(
+<a href=#1808 id=1808 data-nosnippet>1808</a>            ts_array.value(<span class="number">0</span>),
+<a href=#1809 id=1809 data-nosnippet>1809</a>            expected_micros,
+<a href=#1810 id=1810 data-nosnippet>1810</a>            <span class="string">"INT96 in struct: got {}, expected {expected_micros}"</span>,
+<a href=#1811 id=1811 data-nosnippet>1811</a>            ts_array.value(<span class="number">0</span>)
+<a href=#1812 id=1812 data-nosnippet>1812</a>        );
+<a href=#1813 id=1813 data-nosnippet>1813</a>    }
+<a href=#1814 id=1814 data-nosnippet>1814</a>
+<a href=#1815 id=1815 data-nosnippet>1815</a>    <span class="attr">#[tokio::test]
+<a href=#1816 id=1816 data-nosnippet>1816</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_list() {
+<a href=#1817 id=1817 data-nosnippet>1817</a>        <span class="kw">use </span>arrow_array::{ListArray, TimestampMicrosecondArray};
+<a href=#1818 id=1818 data-nosnippet>1818</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
+<a href=#1819 id=1819 data-nosnippet>1819</a>        <span class="kw">use </span>parquet::data_type::Int96Type;
+<a href=#1820 id=1820 data-nosnippet>1820</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
+<a href=#1821 id=1821 data-nosnippet>1821</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
 <a href=#1822 id=1822 data-nosnippet>1822</a>
-<a href=#1823 id=1823 data-nosnippet>1823</a>        <span class="kw">let </span>list_type = SchemaType::group_type_builder(<span class="string">"timestamps"</span>)
-<a href=#1824 id=1824 data-nosnippet>1824</a>            .with_repetition(Repetition::OPTIONAL)
-<a href=#1825 id=1825 data-nosnippet>1825</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
-<a href=#1826 id=1826 data-nosnippet>1826</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::List))
-<a href=#1827 id=1827 data-nosnippet>1827</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(list_group)])
-<a href=#1828 id=1828 data-nosnippet>1828</a>            .build()
-<a href=#1829 id=1829 data-nosnippet>1829</a>            .unwrap();
-<a href=#1830 id=1830 data-nosnippet>1830</a>
-<a href=#1831 id=1831 data-nosnippet>1831</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
-<a href=#1832 id=1832 data-nosnippet>1832</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(list_type)])
-<a href=#1833 id=1833 data-nosnippet>1833</a>            .build()
-<a href=#1834 id=1834 data-nosnippet>1834</a>            .unwrap();
-<a href=#1835 id=1835 data-nosnippet>1835</a>
-<a href=#1836 id=1836 data-nosnippet>1836</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
-<a href=#1837 id=1837 data-nosnippet>1837</a>
-<a href=#1838 id=1838 data-nosnippet>1838</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
-<a href=#1839 id=1839 data-nosnippet>1839</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#1840 id=1840 data-nosnippet>1840</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
-<a href=#1841 id=1841 data-nosnippet>1841</a>
-<a href=#1842 id=1842 data-nosnippet>1842</a>        <span class="comment">// Write a single row with a list containing one INT96 element.
-<a href=#1843 id=1843 data-nosnippet>1843</a>        // def=3: list present (1) + repeated group (2) + element present (3)
-<a href=#1844 id=1844 data-nosnippet>1844</a>        // rep=0: start of a new list
-<a href=#1845 id=1845 data-nosnippet>1845</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
-<a href=#1846 id=1846 data-nosnippet>1846</a>        {
-<a href=#1847 id=1847 data-nosnippet>1847</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#1848 id=1848 data-nosnippet>1848</a>            col.typed::&lt;Int96Type&gt;()
-<a href=#1849 id=1849 data-nosnippet>1849</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">3</span>]), <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]))
-<a href=#1850 id=1850 data-nosnippet>1850</a>                .unwrap();
-<a href=#1851 id=1851 data-nosnippet>1851</a>            col.close().unwrap();
-<a href=#1852 id=1852 data-nosnippet>1852</a>        }
-<a href=#1853 id=1853 data-nosnippet>1853</a>        row_group.close().unwrap();
-<a href=#1854 id=1854 data-nosnippet>1854</a>        writer.close().unwrap();
-<a href=#1855 id=1855 data-nosnippet>1855</a>
-<a href=#1856 id=1856 data-nosnippet>1856</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
-<a href=#1857 id=1857 data-nosnippet>1857</a>            Schema::builder()
-<a href=#1858 id=1858 data-nosnippet>1858</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1859 id=1859 data-nosnippet>1859</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1860 id=1860 data-nosnippet>1860</a>                    NestedField::optional(
-<a href=#1861 id=1861 data-nosnippet>1861</a>                        <span class="number">1</span>,
-<a href=#1862 id=1862 data-nosnippet>1862</a>                        <span class="string">"timestamps"</span>,
-<a href=#1863 id=1863 data-nosnippet>1863</a>                        Type::List(<span class="kw">crate</span>::spec::ListType {
-<a href=#1864 id=1864 data-nosnippet>1864</a>                            element_field: NestedField::optional(
-<a href=#1865 id=1865 data-nosnippet>1865</a>                                <span class="number">2</span>,
-<a href=#1866 id=1866 data-nosnippet>1866</a>                                <span class="string">"element"</span>,
-<a href=#1867 id=1867 data-nosnippet>1867</a>                                Type::Primitive(PrimitiveType::Timestamp),
-<a href=#1868 id=1868 data-nosnippet>1868</a>                            )
-<a href=#1869 id=1869 data-nosnippet>1869</a>                            .into(),
-<a href=#1870 id=1870 data-nosnippet>1870</a>                        }),
-<a href=#1871 id=1871 data-nosnippet>1871</a>                    )
-<a href=#1872 id=1872 data-nosnippet>1872</a>                    .into(),
-<a href=#1873 id=1873 data-nosnippet>1873</a>                ])
-<a href=#1874 id=1874 data-nosnippet>1874</a>                .build()
-<a href=#1875 id=1875 data-nosnippet>1875</a>                .unwrap(),
-<a href=#1876 id=1876 data-nosnippet>1876</a>        );
+<a href=#1823 id=1823 data-nosnippet>1823</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1824 id=1824 data-nosnippet>1824</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1825 id=1825 data-nosnippet>1825</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/list_int96.parquet"</span>);
+<a href=#1826 id=1826 data-nosnippet>1826</a>
+<a href=#1827 id=1827 data-nosnippet>1827</a>        <span class="comment">// 3-level LIST encoding:
+<a href=#1828 id=1828 data-nosnippet>1828</a>        //   optional group timestamps (LIST) {
+<a href=#1829 id=1829 data-nosnippet>1829</a>        //     repeated group list {
+<a href=#1830 id=1830 data-nosnippet>1830</a>        //       optional int96 element;
+<a href=#1831 id=1831 data-nosnippet>1831</a>        //     }
+<a href=#1832 id=1832 data-nosnippet>1832</a>        //   }
+<a href=#1833 id=1833 data-nosnippet>1833</a>        </span><span class="kw">let </span>element_type = SchemaType::primitive_type_builder(<span class="string">"element"</span>, PhysicalType::INT96)
+<a href=#1834 id=1834 data-nosnippet>1834</a>            .with_repetition(Repetition::OPTIONAL)
+<a href=#1835 id=1835 data-nosnippet>1835</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
+<a href=#1836 id=1836 data-nosnippet>1836</a>            .build()
+<a href=#1837 id=1837 data-nosnippet>1837</a>            .unwrap();
+<a href=#1838 id=1838 data-nosnippet>1838</a>
+<a href=#1839 id=1839 data-nosnippet>1839</a>        <span class="kw">let </span>list_group = SchemaType::group_type_builder(<span class="string">"list"</span>)
+<a href=#1840 id=1840 data-nosnippet>1840</a>            .with_repetition(Repetition::REPEATED)
+<a href=#1841 id=1841 data-nosnippet>1841</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(element_type)])
+<a href=#1842 id=1842 data-nosnippet>1842</a>            .build()
+<a href=#1843 id=1843 data-nosnippet>1843</a>            .unwrap();
+<a href=#1844 id=1844 data-nosnippet>1844</a>
+<a href=#1845 id=1845 data-nosnippet>1845</a>        <span class="kw">let </span>list_type = SchemaType::group_type_builder(<span class="string">"timestamps"</span>)
+<a href=#1846 id=1846 data-nosnippet>1846</a>            .with_repetition(Repetition::OPTIONAL)
+<a href=#1847 id=1847 data-nosnippet>1847</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
+<a href=#1848 id=1848 data-nosnippet>1848</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::List))
+<a href=#1849 id=1849 data-nosnippet>1849</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(list_group)])
+<a href=#1850 id=1850 data-nosnippet>1850</a>            .build()
+<a href=#1851 id=1851 data-nosnippet>1851</a>            .unwrap();
+<a href=#1852 id=1852 data-nosnippet>1852</a>
+<a href=#1853 id=1853 data-nosnippet>1853</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
+<a href=#1854 id=1854 data-nosnippet>1854</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(list_type)])
+<a href=#1855 id=1855 data-nosnippet>1855</a>            .build()
+<a href=#1856 id=1856 data-nosnippet>1856</a>            .unwrap();
+<a href=#1857 id=1857 data-nosnippet>1857</a>
+<a href=#1858 id=1858 data-nosnippet>1858</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
+<a href=#1859 id=1859 data-nosnippet>1859</a>
+<a href=#1860 id=1860 data-nosnippet>1860</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
+<a href=#1861 id=1861 data-nosnippet>1861</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#1862 id=1862 data-nosnippet>1862</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
+<a href=#1863 id=1863 data-nosnippet>1863</a>
+<a href=#1864 id=1864 data-nosnippet>1864</a>        <span class="comment">// Write a single row with a list containing one INT96 element.
+<a href=#1865 id=1865 data-nosnippet>1865</a>        // def=3: list present (1) + repeated group (2) + element present (3)
+<a href=#1866 id=1866 data-nosnippet>1866</a>        // rep=0: start of a new list
+<a href=#1867 id=1867 data-nosnippet>1867</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
+<a href=#1868 id=1868 data-nosnippet>1868</a>        {
+<a href=#1869 id=1869 data-nosnippet>1869</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#1870 id=1870 data-nosnippet>1870</a>            col.typed::&lt;Int96Type&gt;()
+<a href=#1871 id=1871 data-nosnippet>1871</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">3</span>]), <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]))
+<a href=#1872 id=1872 data-nosnippet>1872</a>                .unwrap();
+<a href=#1873 id=1873 data-nosnippet>1873</a>            col.close().unwrap();
+<a href=#1874 id=1874 data-nosnippet>1874</a>        }
+<a href=#1875 id=1875 data-nosnippet>1875</a>        row_group.close().unwrap();
+<a href=#1876 id=1876 data-nosnippet>1876</a>        writer.close().unwrap();
 <a href=#1877 id=1877 data-nosnippet>1877</a>
-<a href=#1878 id=1878 data-nosnippet>1878</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
-<a href=#1879 id=1879 data-nosnippet>1879</a>
-<a href=#1880 id=1880 data-nosnippet>1880</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
-<a href=#1881 id=1881 data-nosnippet>1881</a>        <span class="kw">let </span>list_array = batches[<span class="number">0</span>]
-<a href=#1882 id=1882 data-nosnippet>1882</a>            .column(<span class="number">0</span>)
-<a href=#1883 id=1883 data-nosnippet>1883</a>            .as_any()
-<a href=#1884 id=1884 data-nosnippet>1884</a>            .downcast_ref::&lt;ListArray&gt;()
-<a href=#1885 id=1885 data-nosnippet>1885</a>            .expect(<span class="string">"Expected ListArray"</span>);
-<a href=#1886 id=1886 data-nosnippet>1886</a>        <span class="kw">let </span>ts_array = list_array
-<a href=#1887 id=1887 data-nosnippet>1887</a>            .values()
-<a href=#1888 id=1888 data-nosnippet>1888</a>            .as_any()
-<a href=#1889 id=1889 data-nosnippet>1889</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
-<a href=#1890 id=1890 data-nosnippet>1890</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray inside list"</span>);
-<a href=#1891 id=1891 data-nosnippet>1891</a>
-<a href=#1892 id=1892 data-nosnippet>1892</a>        <span class="macro">assert_eq!</span>(
-<a href=#1893 id=1893 data-nosnippet>1893</a>            ts_array.value(<span class="number">0</span>),
-<a href=#1894 id=1894 data-nosnippet>1894</a>            expected_micros,
-<a href=#1895 id=1895 data-nosnippet>1895</a>            <span class="string">"INT96 in list: got {}, expected {expected_micros}"</span>,
-<a href=#1896 id=1896 data-nosnippet>1896</a>            ts_array.value(<span class="number">0</span>)
-<a href=#1897 id=1897 data-nosnippet>1897</a>        );
-<a href=#1898 id=1898 data-nosnippet>1898</a>    }
+<a href=#1878 id=1878 data-nosnippet>1878</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
+<a href=#1879 id=1879 data-nosnippet>1879</a>            Schema::builder()
+<a href=#1880 id=1880 data-nosnippet>1880</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#1881 id=1881 data-nosnippet>1881</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#1882 id=1882 data-nosnippet>1882</a>                    NestedField::optional(
+<a href=#1883 id=1883 data-nosnippet>1883</a>                        <span class="number">1</span>,
+<a href=#1884 id=1884 data-nosnippet>1884</a>                        <span class="string">"timestamps"</span>,
+<a href=#1885 id=1885 data-nosnippet>1885</a>                        Type::List(<span class="kw">crate</span>::spec::ListType {
+<a href=#1886 id=1886 data-nosnippet>1886</a>                            element_field: NestedField::optional(
+<a href=#1887 id=1887 data-nosnippet>1887</a>                                <span class="number">2</span>,
+<a href=#1888 id=1888 data-nosnippet>1888</a>                                <span class="string">"element"</span>,
+<a href=#1889 id=1889 data-nosnippet>1889</a>                                Type::Primitive(PrimitiveType::Timestamp),
+<a href=#1890 id=1890 data-nosnippet>1890</a>                            )
+<a href=#1891 id=1891 data-nosnippet>1891</a>                            .into(),
+<a href=#1892 id=1892 data-nosnippet>1892</a>                        }),
+<a href=#1893 id=1893 data-nosnippet>1893</a>                    )
+<a href=#1894 id=1894 data-nosnippet>1894</a>                    .into(),
+<a href=#1895 id=1895 data-nosnippet>1895</a>                ])
+<a href=#1896 id=1896 data-nosnippet>1896</a>                .build()
+<a href=#1897 id=1897 data-nosnippet>1897</a>                .unwrap(),
+<a href=#1898 id=1898 data-nosnippet>1898</a>        );
 <a href=#1899 id=1899 data-nosnippet>1899</a>
-<a href=#1900 id=1900 data-nosnippet>1900</a>    <span class="attr">#[tokio::test]
-<a href=#1901 id=1901 data-nosnippet>1901</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_map() {
-<a href=#1902 id=1902 data-nosnippet>1902</a>        <span class="kw">use </span>arrow_array::{MapArray, TimestampMicrosecondArray};
-<a href=#1903 id=1903 data-nosnippet>1903</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
-<a href=#1904 id=1904 data-nosnippet>1904</a>        <span class="kw">use </span>parquet::data_type::{ByteArrayType, Int96Type};
-<a href=#1905 id=1905 data-nosnippet>1905</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
-<a href=#1906 id=1906 data-nosnippet>1906</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
-<a href=#1907 id=1907 data-nosnippet>1907</a>
-<a href=#1908 id=1908 data-nosnippet>1908</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#1909 id=1909 data-nosnippet>1909</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#1910 id=1910 data-nosnippet>1910</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/map_int96.parquet"</span>);
-<a href=#1911 id=1911 data-nosnippet>1911</a>
-<a href=#1912 id=1912 data-nosnippet>1912</a>        <span class="comment">// MAP encoding:
-<a href=#1913 id=1913 data-nosnippet>1913</a>        //   optional group ts_map (MAP) {
-<a href=#1914 id=1914 data-nosnippet>1914</a>        //     repeated group key_value {
-<a href=#1915 id=1915 data-nosnippet>1915</a>        //       required binary key (UTF8);
-<a href=#1916 id=1916 data-nosnippet>1916</a>        //       optional int96 value;
-<a href=#1917 id=1917 data-nosnippet>1917</a>        //     }
-<a href=#1918 id=1918 data-nosnippet>1918</a>        //   }
-<a href=#1919 id=1919 data-nosnippet>1919</a>        </span><span class="kw">let </span>key_type = SchemaType::primitive_type_builder(<span class="string">"key"</span>, PhysicalType::BYTE_ARRAY)
-<a href=#1920 id=1920 data-nosnippet>1920</a>            .with_repetition(Repetition::REQUIRED)
-<a href=#1921 id=1921 data-nosnippet>1921</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::String))
-<a href=#1922 id=1922 data-nosnippet>1922</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
-<a href=#1923 id=1923 data-nosnippet>1923</a>            .build()
-<a href=#1924 id=1924 data-nosnippet>1924</a>            .unwrap();
-<a href=#1925 id=1925 data-nosnippet>1925</a>
-<a href=#1926 id=1926 data-nosnippet>1926</a>        <span class="kw">let </span>value_type = SchemaType::primitive_type_builder(<span class="string">"value"</span>, PhysicalType::INT96)
-<a href=#1927 id=1927 data-nosnippet>1927</a>            .with_repetition(Repetition::OPTIONAL)
-<a href=#1928 id=1928 data-nosnippet>1928</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">3</span>))
-<a href=#1929 id=1929 data-nosnippet>1929</a>            .build()
-<a href=#1930 id=1930 data-nosnippet>1930</a>            .unwrap();
-<a href=#1931 id=1931 data-nosnippet>1931</a>
-<a href=#1932 id=1932 data-nosnippet>1932</a>        <span class="kw">let </span>key_value_group = SchemaType::group_type_builder(<span class="string">"key_value"</span>)
-<a href=#1933 id=1933 data-nosnippet>1933</a>            .with_repetition(Repetition::REPEATED)
-<a href=#1934 id=1934 data-nosnippet>1934</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(key_type), Arc::new(value_type)])
-<a href=#1935 id=1935 data-nosnippet>1935</a>            .build()
-<a href=#1936 id=1936 data-nosnippet>1936</a>            .unwrap();
-<a href=#1937 id=1937 data-nosnippet>1937</a>
-<a href=#1938 id=1938 data-nosnippet>1938</a>        <span class="kw">let </span>map_type = SchemaType::group_type_builder(<span class="string">"ts_map"</span>)
-<a href=#1939 id=1939 data-nosnippet>1939</a>            .with_repetition(Repetition::OPTIONAL)
-<a href=#1940 id=1940 data-nosnippet>1940</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
-<a href=#1941 id=1941 data-nosnippet>1941</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::Map))
-<a href=#1942 id=1942 data-nosnippet>1942</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(key_value_group)])
-<a href=#1943 id=1943 data-nosnippet>1943</a>            .build()
-<a href=#1944 id=1944 data-nosnippet>1944</a>            .unwrap();
-<a href=#1945 id=1945 data-nosnippet>1945</a>
-<a href=#1946 id=1946 data-nosnippet>1946</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
-<a href=#1947 id=1947 data-nosnippet>1947</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(map_type)])
-<a href=#1948 id=1948 data-nosnippet>1948</a>            .build()
-<a href=#1949 id=1949 data-nosnippet>1949</a>            .unwrap();
-<a href=#1950 id=1950 data-nosnippet>1950</a>
-<a href=#1951 id=1951 data-nosnippet>1951</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
-<a href=#1952 id=1952 data-nosnippet>1952</a>
-<a href=#1953 id=1953 data-nosnippet>1953</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
-<a href=#1954 id=1954 data-nosnippet>1954</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#1955 id=1955 data-nosnippet>1955</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
-<a href=#1956 id=1956 data-nosnippet>1956</a>
-<a href=#1957 id=1957 data-nosnippet>1957</a>        <span class="comment">// Write a single row with a map containing one key-value pair.
-<a href=#1958 id=1958 data-nosnippet>1958</a>        // rep=0 for both columns: start of a new map.
-<a href=#1959 id=1959 data-nosnippet>1959</a>        // key def=2: map present (1) + key_value entry present (2), key is REQUIRED.
-<a href=#1960 id=1960 data-nosnippet>1960</a>        // value def=3: map present (1) + key_value entry present (2) + value present (3).
-<a href=#1961 id=1961 data-nosnippet>1961</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
-<a href=#1962 id=1962 data-nosnippet>1962</a>        {
-<a href=#1963 id=1963 data-nosnippet>1963</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#1964 id=1964 data-nosnippet>1964</a>            col.typed::&lt;ByteArrayType&gt;()
-<a href=#1965 id=1965 data-nosnippet>1965</a>                .write_batch(
-<a href=#1966 id=1966 data-nosnippet>1966</a>                    <span class="kw-2">&amp;</span>[parquet::data_type::ByteArray::from(<span class="string">"event_time"</span>)],
-<a href=#1967 id=1967 data-nosnippet>1967</a>                    <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">2</span>]),
-<a href=#1968 id=1968 data-nosnippet>1968</a>                    <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]),
-<a href=#1969 id=1969 data-nosnippet>1969</a>                )
-<a href=#1970 id=1970 data-nosnippet>1970</a>                .unwrap();
-<a href=#1971 id=1971 data-nosnippet>1971</a>            col.close().unwrap();
-<a href=#1972 id=1972 data-nosnippet>1972</a>        }
-<a href=#1973 id=1973 data-nosnippet>1973</a>        {
-<a href=#1974 id=1974 data-nosnippet>1974</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
-<a href=#1975 id=1975 data-nosnippet>1975</a>            col.typed::&lt;Int96Type&gt;()
-<a href=#1976 id=1976 data-nosnippet>1976</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">3</span>]), <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]))
-<a href=#1977 id=1977 data-nosnippet>1977</a>                .unwrap();
-<a href=#1978 id=1978 data-nosnippet>1978</a>            col.close().unwrap();
-<a href=#1979 id=1979 data-nosnippet>1979</a>        }
-<a href=#1980 id=1980 data-nosnippet>1980</a>        row_group.close().unwrap();
-<a href=#1981 id=1981 data-nosnippet>1981</a>        writer.close().unwrap();
-<a href=#1982 id=1982 data-nosnippet>1982</a>
-<a href=#1983 id=1983 data-nosnippet>1983</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
-<a href=#1984 id=1984 data-nosnippet>1984</a>            Schema::builder()
-<a href=#1985 id=1985 data-nosnippet>1985</a>                .with_schema_id(<span class="number">1</span>)
-<a href=#1986 id=1986 data-nosnippet>1986</a>                .with_fields(<span class="macro">vec!</span>[
-<a href=#1987 id=1987 data-nosnippet>1987</a>                    NestedField::optional(
-<a href=#1988 id=1988 data-nosnippet>1988</a>                        <span class="number">1</span>,
-<a href=#1989 id=1989 data-nosnippet>1989</a>                        <span class="string">"ts_map"</span>,
-<a href=#1990 id=1990 data-nosnippet>1990</a>                        Type::Map(<span class="kw">crate</span>::spec::MapType {
-<a href=#1991 id=1991 data-nosnippet>1991</a>                            key_field: NestedField::required(
-<a href=#1992 id=1992 data-nosnippet>1992</a>                                <span class="number">2</span>,
-<a href=#1993 id=1993 data-nosnippet>1993</a>                                <span class="string">"key"</span>,
-<a href=#1994 id=1994 data-nosnippet>1994</a>                                Type::Primitive(PrimitiveType::String),
-<a href=#1995 id=1995 data-nosnippet>1995</a>                            )
-<a href=#1996 id=1996 data-nosnippet>1996</a>                            .into(),
-<a href=#1997 id=1997 data-nosnippet>1997</a>                            value_field: NestedField::optional(
-<a href=#1998 id=1998 data-nosnippet>1998</a>                                <span class="number">3</span>,
-<a href=#1999 id=1999 data-nosnippet>1999</a>                                <span class="string">"value"</span>,
-<a href=#2000 id=2000 data-nosnippet>2000</a>                                Type::Primitive(PrimitiveType::Timestamp),
-<a href=#2001 id=2001 data-nosnippet>2001</a>                            )
-<a href=#2002 id=2002 data-nosnippet>2002</a>                            .into(),
-<a href=#2003 id=2003 data-nosnippet>2003</a>                        }),
-<a href=#2004 id=2004 data-nosnippet>2004</a>                    )
-<a href=#2005 id=2005 data-nosnippet>2005</a>                    .into(),
-<a href=#2006 id=2006 data-nosnippet>2006</a>                ])
-<a href=#2007 id=2007 data-nosnippet>2007</a>                .build()
-<a href=#2008 id=2008 data-nosnippet>2008</a>                .unwrap(),
-<a href=#2009 id=2009 data-nosnippet>2009</a>        );
-<a href=#2010 id=2010 data-nosnippet>2010</a>
-<a href=#2011 id=2011 data-nosnippet>2011</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
-<a href=#2012 id=2012 data-nosnippet>2012</a>
-<a href=#2013 id=2013 data-nosnippet>2013</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
-<a href=#2014 id=2014 data-nosnippet>2014</a>        <span class="kw">let </span>map_array = batches[<span class="number">0</span>]
-<a href=#2015 id=2015 data-nosnippet>2015</a>            .column(<span class="number">0</span>)
-<a href=#2016 id=2016 data-nosnippet>2016</a>            .as_any()
-<a href=#2017 id=2017 data-nosnippet>2017</a>            .downcast_ref::&lt;MapArray&gt;()
-<a href=#2018 id=2018 data-nosnippet>2018</a>            .expect(<span class="string">"Expected MapArray"</span>);
-<a href=#2019 id=2019 data-nosnippet>2019</a>        <span class="kw">let </span>ts_array = map_array
-<a href=#2020 id=2020 data-nosnippet>2020</a>            .values()
-<a href=#2021 id=2021 data-nosnippet>2021</a>            .as_any()
-<a href=#2022 id=2022 data-nosnippet>2022</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
-<a href=#2023 id=2023 data-nosnippet>2023</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray as map values"</span>);
-<a href=#2024 id=2024 data-nosnippet>2024</a>
-<a href=#2025 id=2025 data-nosnippet>2025</a>        <span class="macro">assert_eq!</span>(
-<a href=#2026 id=2026 data-nosnippet>2026</a>            ts_array.value(<span class="number">0</span>),
-<a href=#2027 id=2027 data-nosnippet>2027</a>            expected_micros,
-<a href=#2028 id=2028 data-nosnippet>2028</a>            <span class="string">"INT96 in map: got {}, expected {expected_micros}"</span>,
-<a href=#2029 id=2029 data-nosnippet>2029</a>            ts_array.value(<span class="number">0</span>)
-<a href=#2030 id=2030 data-nosnippet>2030</a>        );
-<a href=#2031 id=2031 data-nosnippet>2031</a>    }
-<a href=#2032 id=2032 data-nosnippet>2032</a>}
+<a href=#1900 id=1900 data-nosnippet>1900</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
+<a href=#1901 id=1901 data-nosnippet>1901</a>
+<a href=#1902 id=1902 data-nosnippet>1902</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
+<a href=#1903 id=1903 data-nosnippet>1903</a>        <span class="kw">let </span>list_array = batches[<span class="number">0</span>]
+<a href=#1904 id=1904 data-nosnippet>1904</a>            .column(<span class="number">0</span>)
+<a href=#1905 id=1905 data-nosnippet>1905</a>            .as_any()
+<a href=#1906 id=1906 data-nosnippet>1906</a>            .downcast_ref::&lt;ListArray&gt;()
+<a href=#1907 id=1907 data-nosnippet>1907</a>            .expect(<span class="string">"Expected ListArray"</span>);
+<a href=#1908 id=1908 data-nosnippet>1908</a>        <span class="kw">let </span>ts_array = list_array
+<a href=#1909 id=1909 data-nosnippet>1909</a>            .values()
+<a href=#1910 id=1910 data-nosnippet>1910</a>            .as_any()
+<a href=#1911 id=1911 data-nosnippet>1911</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
+<a href=#1912 id=1912 data-nosnippet>1912</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray inside list"</span>);
+<a href=#1913 id=1913 data-nosnippet>1913</a>
+<a href=#1914 id=1914 data-nosnippet>1914</a>        <span class="macro">assert_eq!</span>(
+<a href=#1915 id=1915 data-nosnippet>1915</a>            ts_array.value(<span class="number">0</span>),
+<a href=#1916 id=1916 data-nosnippet>1916</a>            expected_micros,
+<a href=#1917 id=1917 data-nosnippet>1917</a>            <span class="string">"INT96 in list: got {}, expected {expected_micros}"</span>,
+<a href=#1918 id=1918 data-nosnippet>1918</a>            ts_array.value(<span class="number">0</span>)
+<a href=#1919 id=1919 data-nosnippet>1919</a>        );
+<a href=#1920 id=1920 data-nosnippet>1920</a>    }
+<a href=#1921 id=1921 data-nosnippet>1921</a>
+<a href=#1922 id=1922 data-nosnippet>1922</a>    <span class="attr">#[tokio::test]
+<a href=#1923 id=1923 data-nosnippet>1923</a>    </span><span class="kw">async fn </span>test_read_int96_timestamps_in_map() {
+<a href=#1924 id=1924 data-nosnippet>1924</a>        <span class="kw">use </span>arrow_array::{MapArray, TimestampMicrosecondArray};
+<a href=#1925 id=1925 data-nosnippet>1925</a>        <span class="kw">use </span>parquet::basic::{Repetition, Type <span class="kw">as </span>PhysicalType};
+<a href=#1926 id=1926 data-nosnippet>1926</a>        <span class="kw">use </span>parquet::data_type::{ByteArrayType, Int96Type};
+<a href=#1927 id=1927 data-nosnippet>1927</a>        <span class="kw">use </span>parquet::file::writer::SerializedFileWriter;
+<a href=#1928 id=1928 data-nosnippet>1928</a>        <span class="kw">use </span>parquet::schema::types::Type <span class="kw">as </span>SchemaType;
+<a href=#1929 id=1929 data-nosnippet>1929</a>
+<a href=#1930 id=1930 data-nosnippet>1930</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#1931 id=1931 data-nosnippet>1931</a>        <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#1932 id=1932 data-nosnippet>1932</a>        <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/map_int96.parquet"</span>);
+<a href=#1933 id=1933 data-nosnippet>1933</a>
+<a href=#1934 id=1934 data-nosnippet>1934</a>        <span class="comment">// MAP encoding:
+<a href=#1935 id=1935 data-nosnippet>1935</a>        //   optional group ts_map (MAP) {
+<a href=#1936 id=1936 data-nosnippet>1936</a>        //     repeated group key_value {
+<a href=#1937 id=1937 data-nosnippet>1937</a>        //       required binary key (UTF8);
+<a href=#1938 id=1938 data-nosnippet>1938</a>        //       optional int96 value;
+<a href=#1939 id=1939 data-nosnippet>1939</a>        //     }
+<a href=#1940 id=1940 data-nosnippet>1940</a>        //   }
+<a href=#1941 id=1941 data-nosnippet>1941</a>        </span><span class="kw">let </span>key_type = SchemaType::primitive_type_builder(<span class="string">"key"</span>, PhysicalType::BYTE_ARRAY)
+<a href=#1942 id=1942 data-nosnippet>1942</a>            .with_repetition(Repetition::REQUIRED)
+<a href=#1943 id=1943 data-nosnippet>1943</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::String))
+<a href=#1944 id=1944 data-nosnippet>1944</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">2</span>))
+<a href=#1945 id=1945 data-nosnippet>1945</a>            .build()
+<a href=#1946 id=1946 data-nosnippet>1946</a>            .unwrap();
+<a href=#1947 id=1947 data-nosnippet>1947</a>
+<a href=#1948 id=1948 data-nosnippet>1948</a>        <span class="kw">let </span>value_type = SchemaType::primitive_type_builder(<span class="string">"value"</span>, PhysicalType::INT96)
+<a href=#1949 id=1949 data-nosnippet>1949</a>            .with_repetition(Repetition::OPTIONAL)
+<a href=#1950 id=1950 data-nosnippet>1950</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">3</span>))
+<a href=#1951 id=1951 data-nosnippet>1951</a>            .build()
+<a href=#1952 id=1952 data-nosnippet>1952</a>            .unwrap();
+<a href=#1953 id=1953 data-nosnippet>1953</a>
+<a href=#1954 id=1954 data-nosnippet>1954</a>        <span class="kw">let </span>key_value_group = SchemaType::group_type_builder(<span class="string">"key_value"</span>)
+<a href=#1955 id=1955 data-nosnippet>1955</a>            .with_repetition(Repetition::REPEATED)
+<a href=#1956 id=1956 data-nosnippet>1956</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(key_type), Arc::new(value_type)])
+<a href=#1957 id=1957 data-nosnippet>1957</a>            .build()
+<a href=#1958 id=1958 data-nosnippet>1958</a>            .unwrap();
+<a href=#1959 id=1959 data-nosnippet>1959</a>
+<a href=#1960 id=1960 data-nosnippet>1960</a>        <span class="kw">let </span>map_type = SchemaType::group_type_builder(<span class="string">"ts_map"</span>)
+<a href=#1961 id=1961 data-nosnippet>1961</a>            .with_repetition(Repetition::OPTIONAL)
+<a href=#1962 id=1962 data-nosnippet>1962</a>            .with_id(<span class="prelude-val">Some</span>(<span class="number">1</span>))
+<a href=#1963 id=1963 data-nosnippet>1963</a>            .with_logical_type(<span class="prelude-val">Some</span>(parquet::basic::LogicalType::Map))
+<a href=#1964 id=1964 data-nosnippet>1964</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(key_value_group)])
+<a href=#1965 id=1965 data-nosnippet>1965</a>            .build()
+<a href=#1966 id=1966 data-nosnippet>1966</a>            .unwrap();
+<a href=#1967 id=1967 data-nosnippet>1967</a>
+<a href=#1968 id=1968 data-nosnippet>1968</a>        <span class="kw">let </span>parquet_schema = SchemaType::group_type_builder(<span class="string">"schema"</span>)
+<a href=#1969 id=1969 data-nosnippet>1969</a>            .with_fields(<span class="macro">vec!</span>[Arc::new(map_type)])
+<a href=#1970 id=1970 data-nosnippet>1970</a>            .build()
+<a href=#1971 id=1971 data-nosnippet>1971</a>            .unwrap();
+<a href=#1972 id=1972 data-nosnippet>1972</a>
+<a href=#1973 id=1973 data-nosnippet>1973</a>        <span class="kw">let </span>(int96_val, expected_micros) = make_int96_test_value();
+<a href=#1974 id=1974 data-nosnippet>1974</a>
+<a href=#1975 id=1975 data-nosnippet>1975</a>        <span class="kw">let </span>file = File::create(<span class="kw-2">&amp;</span>file_path).unwrap();
+<a href=#1976 id=1976 data-nosnippet>1976</a>        <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#1977 id=1977 data-nosnippet>1977</a>            SerializedFileWriter::new(file, Arc::new(parquet_schema), Default::default()).unwrap();
+<a href=#1978 id=1978 data-nosnippet>1978</a>
+<a href=#1979 id=1979 data-nosnippet>1979</a>        <span class="comment">// Write a single row with a map containing one key-value pair.
+<a href=#1980 id=1980 data-nosnippet>1980</a>        // rep=0 for both columns: start of a new map.
+<a href=#1981 id=1981 data-nosnippet>1981</a>        // key def=2: map present (1) + key_value entry present (2), key is REQUIRED.
+<a href=#1982 id=1982 data-nosnippet>1982</a>        // value def=3: map present (1) + key_value entry present (2) + value present (3).
+<a href=#1983 id=1983 data-nosnippet>1983</a>        </span><span class="kw">let </span><span class="kw-2">mut </span>row_group = writer.next_row_group().unwrap();
+<a href=#1984 id=1984 data-nosnippet>1984</a>        {
+<a href=#1985 id=1985 data-nosnippet>1985</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#1986 id=1986 data-nosnippet>1986</a>            col.typed::&lt;ByteArrayType&gt;()
+<a href=#1987 id=1987 data-nosnippet>1987</a>                .write_batch(
+<a href=#1988 id=1988 data-nosnippet>1988</a>                    <span class="kw-2">&amp;</span>[parquet::data_type::ByteArray::from(<span class="string">"event_time"</span>)],
+<a href=#1989 id=1989 data-nosnippet>1989</a>                    <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">2</span>]),
+<a href=#1990 id=1990 data-nosnippet>1990</a>                    <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]),
+<a href=#1991 id=1991 data-nosnippet>1991</a>                )
+<a href=#1992 id=1992 data-nosnippet>1992</a>                .unwrap();
+<a href=#1993 id=1993 data-nosnippet>1993</a>            col.close().unwrap();
+<a href=#1994 id=1994 data-nosnippet>1994</a>        }
+<a href=#1995 id=1995 data-nosnippet>1995</a>        {
+<a href=#1996 id=1996 data-nosnippet>1996</a>            <span class="kw">let </span><span class="kw-2">mut </span>col = row_group.next_column().unwrap().unwrap();
+<a href=#1997 id=1997 data-nosnippet>1997</a>            col.typed::&lt;Int96Type&gt;()
+<a href=#1998 id=1998 data-nosnippet>1998</a>                .write_batch(<span class="kw-2">&amp;</span>[int96_val], <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">3</span>]), <span class="prelude-val">Some</span>(<span class="kw-2">&amp;</span>[<span class="number">0</span>]))
+<a href=#1999 id=1999 data-nosnippet>1999</a>                .unwrap();
+<a href=#2000 id=2000 data-nosnippet>2000</a>            col.close().unwrap();
+<a href=#2001 id=2001 data-nosnippet>2001</a>        }
+<a href=#2002 id=2002 data-nosnippet>2002</a>        row_group.close().unwrap();
+<a href=#2003 id=2003 data-nosnippet>2003</a>        writer.close().unwrap();
+<a href=#2004 id=2004 data-nosnippet>2004</a>
+<a href=#2005 id=2005 data-nosnippet>2005</a>        <span class="kw">let </span>iceberg_schema = Arc::new(
+<a href=#2006 id=2006 data-nosnippet>2006</a>            Schema::builder()
+<a href=#2007 id=2007 data-nosnippet>2007</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#2008 id=2008 data-nosnippet>2008</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#2009 id=2009 data-nosnippet>2009</a>                    NestedField::optional(
+<a href=#2010 id=2010 data-nosnippet>2010</a>                        <span class="number">1</span>,
+<a href=#2011 id=2011 data-nosnippet>2011</a>                        <span class="string">"ts_map"</span>,
+<a href=#2012 id=2012 data-nosnippet>2012</a>                        Type::Map(<span class="kw">crate</span>::spec::MapType {
+<a href=#2013 id=2013 data-nosnippet>2013</a>                            key_field: NestedField::required(
+<a href=#2014 id=2014 data-nosnippet>2014</a>                                <span class="number">2</span>,
+<a href=#2015 id=2015 data-nosnippet>2015</a>                                <span class="string">"key"</span>,
+<a href=#2016 id=2016 data-nosnippet>2016</a>                                Type::Primitive(PrimitiveType::String),
+<a href=#2017 id=2017 data-nosnippet>2017</a>                            )
+<a href=#2018 id=2018 data-nosnippet>2018</a>                            .into(),
+<a href=#2019 id=2019 data-nosnippet>2019</a>                            value_field: NestedField::optional(
+<a href=#2020 id=2020 data-nosnippet>2020</a>                                <span class="number">3</span>,
+<a href=#2021 id=2021 data-nosnippet>2021</a>                                <span class="string">"value"</span>,
+<a href=#2022 id=2022 data-nosnippet>2022</a>                                Type::Primitive(PrimitiveType::Timestamp),
+<a href=#2023 id=2023 data-nosnippet>2023</a>                            )
+<a href=#2024 id=2024 data-nosnippet>2024</a>                            .into(),
+<a href=#2025 id=2025 data-nosnippet>2025</a>                        }),
+<a href=#2026 id=2026 data-nosnippet>2026</a>                    )
+<a href=#2027 id=2027 data-nosnippet>2027</a>                    .into(),
+<a href=#2028 id=2028 data-nosnippet>2028</a>                ])
+<a href=#2029 id=2029 data-nosnippet>2029</a>                .build()
+<a href=#2030 id=2030 data-nosnippet>2030</a>                .unwrap(),
+<a href=#2031 id=2031 data-nosnippet>2031</a>        );
+<a href=#2032 id=2032 data-nosnippet>2032</a>
+<a href=#2033 id=2033 data-nosnippet>2033</a>        <span class="kw">let </span>batches = read_int96_batches(<span class="kw-2">&amp;</span>file_path, iceberg_schema, <span class="macro">vec!</span>[<span class="number">1</span>]).<span class="kw">await</span>;
+<a href=#2034 id=2034 data-nosnippet>2034</a>
+<a href=#2035 id=2035 data-nosnippet>2035</a>        <span class="macro">assert_eq!</span>(batches.len(), <span class="number">1</span>);
+<a href=#2036 id=2036 data-nosnippet>2036</a>        <span class="kw">let </span>map_array = batches[<span class="number">0</span>]
+<a href=#2037 id=2037 data-nosnippet>2037</a>            .column(<span class="number">0</span>)
+<a href=#2038 id=2038 data-nosnippet>2038</a>            .as_any()
+<a href=#2039 id=2039 data-nosnippet>2039</a>            .downcast_ref::&lt;MapArray&gt;()
+<a href=#2040 id=2040 data-nosnippet>2040</a>            .expect(<span class="string">"Expected MapArray"</span>);
+<a href=#2041 id=2041 data-nosnippet>2041</a>        <span class="kw">let </span>ts_array = map_array
+<a href=#2042 id=2042 data-nosnippet>2042</a>            .values()
+<a href=#2043 id=2043 data-nosnippet>2043</a>            .as_any()
+<a href=#2044 id=2044 data-nosnippet>2044</a>            .downcast_ref::&lt;TimestampMicrosecondArray&gt;()
+<a href=#2045 id=2045 data-nosnippet>2045</a>            .expect(<span class="string">"Expected TimestampMicrosecondArray as map values"</span>);
+<a href=#2046 id=2046 data-nosnippet>2046</a>
+<a href=#2047 id=2047 data-nosnippet>2047</a>        <span class="macro">assert_eq!</span>(
+<a href=#2048 id=2048 data-nosnippet>2048</a>            ts_array.value(<span class="number">0</span>),
+<a href=#2049 id=2049 data-nosnippet>2049</a>            expected_micros,
+<a href=#2050 id=2050 data-nosnippet>2050</a>            <span class="string">"INT96 in map: got {}, expected {expected_micros}"</span>,
+<a href=#2051 id=2051 data-nosnippet>2051</a>            ts_array.value(<span class="number">0</span>)
+<a href=#2052 id=2052 data-nosnippet>2052</a>        );
+<a href=#2053 id=2053 data-nosnippet>2053</a>    }
+<a href=#2054 id=2054 data-nosnippet>2054</a>
+<a href=#2055 id=2055 data-nosnippet>2055</a>    <span class="doccomment">/// Writes `id` (Int32) plus a wide string column (field id 2) whose bytes dominate
+<a href=#2056 id=2056 data-nosnippet>2056</a>    /// the file, so that reading it is visible in `bytes_read`.
+<a href=#2057 id=2057 data-nosnippet>2057</a>    ///
+<a href=#2058 id=2058 data-nosnippet>2058</a>    /// `extra_fields`/`extra_columns` (e.g. a physical metadata leaf) are appended after
+<a href=#2059 id=2059 data-nosnippet>2059</a>    /// the `id` and wide columns, mirroring `write_plain_parquet`'s shape.
+<a href=#2060 id=2060 data-nosnippet>2060</a>    </span><span class="kw">fn </span>write_parquet_with_wide_column(
+<a href=#2061 id=2061 data-nosnippet>2061</a>        dir: <span class="kw-2">&amp;</span>str,
+<a href=#2062 id=2062 data-nosnippet>2062</a>        name: <span class="kw-2">&amp;</span>str,
+<a href=#2063 id=2063 data-nosnippet>2063</a>        extra_fields: Vec&lt;Field&gt;,
+<a href=#2064 id=2064 data-nosnippet>2064</a>        extra_columns: Vec&lt;ArrayRef&gt;,
+<a href=#2065 id=2065 data-nosnippet>2065</a>    ) -&gt; String {
+<a href=#2066 id=2066 data-nosnippet>2066</a>        <span class="kw">let </span>wide_field =
+<a href=#2067 id=2067 data-nosnippet>2067</a>            Field::new(<span class="string">"wide"</span>, DataType::Utf8, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#2068 id=2068 data-nosnippet>2068</a>                PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#2069 id=2069 data-nosnippet>2069</a>                <span class="string">"2"</span>.to_string(),
+<a href=#2070 id=2070 data-nosnippet>2070</a>            )]));
+<a href=#2071 id=2071 data-nosnippet>2071</a>        <span class="comment">// Varied bytes so the column chunk does not compress away under SNAPPY, keeping
+<a href=#2072 id=2072 data-nosnippet>2072</a>        // the `bytes_read` difference between projecting it and not unambiguous.
+<a href=#2073 id=2073 data-nosnippet>2073</a>        </span><span class="kw">let </span>wide_values: Vec&lt;String&gt; = (<span class="number">0</span>..<span class="number">3</span>)
+<a href=#2074 id=2074 data-nosnippet>2074</a>            .map(|i| {
+<a href=#2075 id=2075 data-nosnippet>2075</a>                (<span class="number">0</span>..<span class="number">2048</span>)
+<a href=#2076 id=2076 data-nosnippet>2076</a>                    .map(|j| ((i * <span class="number">2048 </span>+ j) % <span class="number">251</span>) <span class="kw">as </span>u8 <span class="kw">as </span>char)
+<a href=#2077 id=2077 data-nosnippet>2077</a>                    .collect()
+<a href=#2078 id=2078 data-nosnippet>2078</a>            })
+<a href=#2079 id=2079 data-nosnippet>2079</a>            .collect();
+<a href=#2080 id=2080 data-nosnippet>2080</a>
+<a href=#2081 id=2081 data-nosnippet>2081</a>        <span class="kw">let </span><span class="kw-2">mut </span>fields = <span class="macro">vec!</span>[wide_field];
+<a href=#2082 id=2082 data-nosnippet>2082</a>        fields.extend(extra_fields);
+<a href=#2083 id=2083 data-nosnippet>2083</a>        <span class="kw">let </span><span class="kw-2">mut </span>columns: Vec&lt;ArrayRef&gt; = <span class="macro">vec!</span>[Arc::new(StringArray::from(wide_values))];
+<a href=#2084 id=2084 data-nosnippet>2084</a>        columns.extend(extra_columns);
+<a href=#2085 id=2085 data-nosnippet>2085</a>        write_plain_parquet(dir, name, fields, columns)
+<a href=#2086 id=2086 data-nosnippet>2086</a>    }
+<a href=#2087 id=2087 data-nosnippet>2087</a>
+<a href=#2088 id=2088 data-nosnippet>2088</a>    <span class="doccomment">/// Schema with `id` (field 1, Int) and `wide` (field 2, String), matching
+<a href=#2089 id=2089 data-nosnippet>2089</a>    /// `write_parquet_with_wide_column`.
+<a href=#2090 id=2090 data-nosnippet>2090</a>    </span><span class="kw">fn </span>id_and_wide_schema() -&gt; SchemaRef {
+<a href=#2091 id=2091 data-nosnippet>2091</a>        Arc::new(
+<a href=#2092 id=2092 data-nosnippet>2092</a>            Schema::builder()
+<a href=#2093 id=2093 data-nosnippet>2093</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#2094 id=2094 data-nosnippet>2094</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#2095 id=2095 data-nosnippet>2095</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#2096 id=2096 data-nosnippet>2096</a>                    NestedField::required(<span class="number">2</span>, <span class="string">"wide"</span>, Type::Primitive(PrimitiveType::String)).into(),
+<a href=#2097 id=2097 data-nosnippet>2097</a>                ])
+<a href=#2098 id=2098 data-nosnippet>2098</a>                .build()
+<a href=#2099 id=2099 data-nosnippet>2099</a>                .unwrap(),
+<a href=#2100 id=2100 data-nosnippet>2100</a>        )
+<a href=#2101 id=2101 data-nosnippet>2101</a>    }
+<a href=#2102 id=2102 data-nosnippet>2102</a>
+<a href=#2103 id=2103 data-nosnippet>2103</a>    <span class="doccomment">/// Builds a scan task over `file_path` projecting `project_field_ids`.
+<a href=#2104 id=2104 data-nosnippet>2104</a>    </span><span class="kw">fn </span>metadata_projection_task(
+<a href=#2105 id=2105 data-nosnippet>2105</a>        file_path: String,
+<a href=#2106 id=2106 data-nosnippet>2106</a>        schema: SchemaRef,
+<a href=#2107 id=2107 data-nosnippet>2107</a>        project_field_ids: Vec&lt;i32&gt;,
+<a href=#2108 id=2108 data-nosnippet>2108</a>    ) -&gt; FileScanTask {
+<a href=#2109 id=2109 data-nosnippet>2109</a>        FileScanTask::builder()
+<a href=#2110 id=2110 data-nosnippet>2110</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#2111 id=2111 data-nosnippet>2111</a>            .with_start(<span class="number">0</span>)
+<a href=#2112 id=2112 data-nosnippet>2112</a>            .with_length(<span class="number">0</span>)
+<a href=#2113 id=2113 data-nosnippet>2113</a>            .with_data_file_path(file_path)
+<a href=#2114 id=2114 data-nosnippet>2114</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#2115 id=2115 data-nosnippet>2115</a>            .with_schema(schema)
+<a href=#2116 id=2116 data-nosnippet>2116</a>            .with_project_field_ids(project_field_ids)
+<a href=#2117 id=2117 data-nosnippet>2117</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#2118 id=2118 data-nosnippet>2118</a>            .build()
+<a href=#2119 id=2119 data-nosnippet>2119</a>    }
+<a href=#2120 id=2120 data-nosnippet>2120</a>
+<a href=#2121 id=2121 data-nosnippet>2121</a>    <span class="doccomment">/// Runs a single-task scan and returns the batches plus the bytes read from storage.
+<a href=#2122 id=2122 data-nosnippet>2122</a>    </span><span class="kw">async fn </span>scan_task(task: FileScanTask) -&gt; (Vec&lt;RecordBatch&gt;, u64) {
+<a href=#2123 id=2123 data-nosnippet>2123</a>        <span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current()).build();
+<a href=#2124 id=2124 data-nosnippet>2124</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#2125 id=2125 data-nosnippet>2125</a>        <span class="kw">let </span>scan = reader.read(tasks).unwrap();
+<a href=#2126 id=2126 data-nosnippet>2126</a>        <span class="kw">let </span>metrics = scan.metrics().clone();
+<a href=#2127 id=2127 data-nosnippet>2127</a>        <span class="kw">let </span>batches = scan.stream().try_collect().<span class="kw">await</span>.unwrap();
+<a href=#2128 id=2128 data-nosnippet>2128</a>        (batches, metrics.bytes_read())
+<a href=#2129 id=2129 data-nosnippet>2129</a>    }
+<a href=#2130 id=2130 data-nosnippet>2130</a>
+<a href=#2131 id=2131 data-nosnippet>2131</a>    <span class="attr">#[tokio::test]
+<a href=#2132 id=2132 data-nosnippet>2132</a>    </span><span class="kw">async fn </span>test_pos_only_projection_reads_no_data_columns() {
+<a href=#2133 id=2133 data-nosnippet>2133</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2134 id=2134 data-nosnippet>2134</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2135 id=2135 data-nosnippet>2135</a>
+<a href=#2136 id=2136 data-nosnippet>2136</a>        <span class="kw">let </span>pos_only = metadata_projection_task(
+<a href=#2137 id=2137 data-nosnippet>2137</a>            write_parquet_with_wide_column(dir, <span class="string">"pos_only.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
+<a href=#2138 id=2138 data-nosnippet>2138</a>            id_and_wide_schema(),
+<a href=#2139 id=2139 data-nosnippet>2139</a>            <span class="macro">vec!</span>[RESERVED_FIELD_ID_POS],
+<a href=#2140 id=2140 data-nosnippet>2140</a>        );
+<a href=#2141 id=2141 data-nosnippet>2141</a>        <span class="kw">let </span>(batches, pos_only_bytes) = scan_task(pos_only).<span class="kw">await</span>;
+<a href=#2142 id=2142 data-nosnippet>2142</a>
+<a href=#2143 id=2143 data-nosnippet>2143</a>        <span class="comment">// Only `_pos` is materialized -- no data columns.
+<a href=#2144 id=2144 data-nosnippet>2144</a>        </span><span class="macro">assert_eq!</span>(batches[<span class="number">0</span>].num_columns(), <span class="number">1</span>);
+<a href=#2145 id=2145 data-nosnippet>2145</a>        <span class="kw">let </span>pos_col = batches[<span class="number">0</span>]
+<a href=#2146 id=2146 data-nosnippet>2146</a>            .column_by_name(RESERVED_COL_NAME_POS)
+<a href=#2147 id=2147 data-nosnippet>2147</a>            .expect(<span class="string">"_pos column should be present"</span>)
+<a href=#2148 id=2148 data-nosnippet>2148</a>            .as_primitive::&lt;arrow_array::types::Int64Type&gt;();
+<a href=#2149 id=2149 data-nosnippet>2149</a>        <span class="macro">assert_eq!</span>(pos_col.values(), <span class="kw-2">&amp;</span>[<span class="number">0</span>, <span class="number">1</span>, <span class="number">2</span>]);
+<a href=#2150 id=2150 data-nosnippet>2150</a>
+<a href=#2151 id=2151 data-nosnippet>2151</a>        <span class="comment">// A scan of the same-shaped file that also projects the wide data column must read
+<a href=#2152 id=2152 data-nosnippet>2152</a>        // materially more, proving the wide column chunk was not fetched above.
+<a href=#2153 id=2153 data-nosnippet>2153</a>        </span><span class="kw">let </span>with_data = metadata_projection_task(
+<a href=#2154 id=2154 data-nosnippet>2154</a>            write_parquet_with_wide_column(dir, <span class="string">"pos_only_ref.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]),
+<a href=#2155 id=2155 data-nosnippet>2155</a>            id_and_wide_schema(),
+<a href=#2156 id=2156 data-nosnippet>2156</a>            <span class="macro">vec!</span>[<span class="number">2</span>, RESERVED_FIELD_ID_POS],
+<a href=#2157 id=2157 data-nosnippet>2157</a>        );
+<a href=#2158 id=2158 data-nosnippet>2158</a>        <span class="kw">let </span>(<span class="kw">_</span>, with_data_bytes) = scan_task(with_data).<span class="kw">await</span>;
+<a href=#2159 id=2159 data-nosnippet>2159</a>
+<a href=#2160 id=2160 data-nosnippet>2160</a>        <span class="macro">assert!</span>(
+<a href=#2161 id=2161 data-nosnippet>2161</a>            pos_only_bytes &lt; with_data_bytes,
+<a href=#2162 id=2162 data-nosnippet>2162</a>            <span class="string">"_pos-only scan should read fewer bytes than a scan of the wide column: \
+<a href=#2163 id=2163 data-nosnippet>2163</a>             {pos_only_bytes} vs {with_data_bytes}"
+<a href=#2164 id=2164 data-nosnippet>2164</a>        </span>);
+<a href=#2165 id=2165 data-nosnippet>2165</a>    }
+<a href=#2166 id=2166 data-nosnippet>2166</a>
+<a href=#2167 id=2167 data-nosnippet>2167</a>    <span class="attr">#[tokio::test]
+<a href=#2168 id=2168 data-nosnippet>2168</a>    </span><span class="kw">async fn </span>test_pos_only_projection_keeps_absolute_pos_under_predicate() {
+<a href=#2169 id=2169 data-nosnippet>2169</a>        <span class="kw">use </span><span class="kw">crate</span>::expr::{Bind, Reference};
+<a href=#2170 id=2170 data-nosnippet>2170</a>        <span class="kw">use </span><span class="kw">crate</span>::spec::Datum;
+<a href=#2171 id=2171 data-nosnippet>2171</a>
+<a href=#2172 id=2172 data-nosnippet>2172</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2173 id=2173 data-nosnippet>2173</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2174 id=2174 data-nosnippet>2174</a>        <span class="comment">// id = [1, 2, 3]; drop the middle physical row via a predicate + row selection.
+<a href=#2175 id=2175 data-nosnippet>2175</a>        </span><span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"pos_only_predicate.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#2176 id=2176 data-nosnippet>2176</a>
+<a href=#2177 id=2177 data-nosnippet>2177</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#2178 id=2178 data-nosnippet>2178</a>            Schema::builder()
+<a href=#2179 id=2179 data-nosnippet>2179</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#2180 id=2180 data-nosnippet>2180</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#2181 id=2181 data-nosnippet>2181</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#2182 id=2182 data-nosnippet>2182</a>                ])
+<a href=#2183 id=2183 data-nosnippet>2183</a>                .build()
+<a href=#2184 id=2184 data-nosnippet>2184</a>                .unwrap(),
+<a href=#2185 id=2185 data-nosnippet>2185</a>        );
+<a href=#2186 id=2186 data-nosnippet>2186</a>        <span class="kw">let </span>bound = Reference::new(<span class="string">"id"</span>)
+<a href=#2187 id=2187 data-nosnippet>2187</a>            .not_equal_to(Datum::int(<span class="number">2</span>))
+<a href=#2188 id=2188 data-nosnippet>2188</a>            .bind(Arc::clone(<span class="kw-2">&amp;</span>schema), <span class="bool-val">false</span>)
+<a href=#2189 id=2189 data-nosnippet>2189</a>            .unwrap();
+<a href=#2190 id=2190 data-nosnippet>2190</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#2191 id=2191 data-nosnippet>2191</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#2192 id=2192 data-nosnippet>2192</a>            .with_start(<span class="number">0</span>)
+<a href=#2193 id=2193 data-nosnippet>2193</a>            .with_length(<span class="number">0</span>)
+<a href=#2194 id=2194 data-nosnippet>2194</a>            .with_data_file_path(file_path)
+<a href=#2195 id=2195 data-nosnippet>2195</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#2196 id=2196 data-nosnippet>2196</a>            .with_schema(schema)
+<a href=#2197 id=2197 data-nosnippet>2197</a>            .with_project_field_ids(<span class="macro">vec!</span>[RESERVED_FIELD_ID_POS])
+<a href=#2198 id=2198 data-nosnippet>2198</a>            .with_predicate(<span class="prelude-val">Some</span>(bound))
+<a href=#2199 id=2199 data-nosnippet>2199</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#2200 id=2200 data-nosnippet>2200</a>            .build();
+<a href=#2201 id=2201 data-nosnippet>2201</a>
+<a href=#2202 id=2202 data-nosnippet>2202</a>        <span class="comment">// Row selection must be enabled for the predicate to filter rows. The row filter
+<a href=#2203 id=2203 data-nosnippet>2203</a>        // reads `id` for its own evaluation even though `id` is not projected; the surviving
+<a href=#2204 id=2204 data-nosnippet>2204</a>        // rows must keep their ABSOLUTE positions (0 and 2), not renumbered (0 and 1).
+<a href=#2205 id=2205 data-nosnippet>2205</a>        </span><span class="kw">let </span>reader = ArrowReaderBuilder::new(FileIO::new_with_fs(), Runtime::current())
+<a href=#2206 id=2206 data-nosnippet>2206</a>            .with_row_selection_enabled(<span class="bool-val">true</span>)
+<a href=#2207 id=2207 data-nosnippet>2207</a>            .build();
+<a href=#2208 id=2208 data-nosnippet>2208</a>        <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#2209 id=2209 data-nosnippet>2209</a>        <span class="kw">let </span>batches: Vec&lt;RecordBatch&gt; = reader
+<a href=#2210 id=2210 data-nosnippet>2210</a>            .read(tasks)
+<a href=#2211 id=2211 data-nosnippet>2211</a>            .unwrap()
+<a href=#2212 id=2212 data-nosnippet>2212</a>            .stream()
+<a href=#2213 id=2213 data-nosnippet>2213</a>            .try_collect()
+<a href=#2214 id=2214 data-nosnippet>2214</a>            .<span class="kw">await
+<a href=#2215 id=2215 data-nosnippet>2215</a>            </span>.unwrap();
+<a href=#2216 id=2216 data-nosnippet>2216</a>
+<a href=#2217 id=2217 data-nosnippet>2217</a>        <span class="kw">let </span>pos: Vec&lt;i64&gt; = batches
+<a href=#2218 id=2218 data-nosnippet>2218</a>            .iter()
+<a href=#2219 id=2219 data-nosnippet>2219</a>            .flat_map(|b| {
+<a href=#2220 id=2220 data-nosnippet>2220</a>                b.column_by_name(RESERVED_COL_NAME_POS)
+<a href=#2221 id=2221 data-nosnippet>2221</a>                    .expect(<span class="string">"_pos column should be present"</span>)
+<a href=#2222 id=2222 data-nosnippet>2222</a>                    .as_primitive::&lt;arrow_array::types::Int64Type&gt;()
+<a href=#2223 id=2223 data-nosnippet>2223</a>                    .values()
+<a href=#2224 id=2224 data-nosnippet>2224</a>                    .to_vec()
+<a href=#2225 id=2225 data-nosnippet>2225</a>            })
+<a href=#2226 id=2226 data-nosnippet>2226</a>            .collect();
+<a href=#2227 id=2227 data-nosnippet>2227</a>        <span class="macro">assert_eq!</span>(pos, <span class="macro">vec!</span>[<span class="number">0</span>, <span class="number">2</span>]);
+<a href=#2228 id=2228 data-nosnippet>2228</a>    }
+<a href=#2229 id=2229 data-nosnippet>2229</a>
+<a href=#2230 id=2230 data-nosnippet>2230</a>    <span class="attr">#[tokio::test]
+<a href=#2231 id=2231 data-nosnippet>2231</a>    </span><span class="kw">async fn </span>test_pos_and_file_projection() {
+<a href=#2232 id=2232 data-nosnippet>2232</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_FILE;
+<a href=#2233 id=2233 data-nosnippet>2233</a>
+<a href=#2234 id=2234 data-nosnippet>2234</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2235 id=2235 data-nosnippet>2235</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2236 id=2236 data-nosnippet>2236</a>        <span class="comment">// The motivating row-lineage shape: a synthesized position column (mask -&gt; none)
+<a href=#2237 id=2237 data-nosnippet>2237</a>        // alongside a materialized per-file constant.
+<a href=#2238 id=2238 data-nosnippet>2238</a>        </span><span class="kw">let </span>file_path = write_parquet_with_wide_column(dir, <span class="string">"pos_and_file.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#2239 id=2239 data-nosnippet>2239</a>        <span class="kw">let </span>task = metadata_projection_task(file_path.clone(), id_and_wide_schema(), <span class="macro">vec!</span>[
+<a href=#2240 id=2240 data-nosnippet>2240</a>            RESERVED_FIELD_ID_POS,
+<a href=#2241 id=2241 data-nosnippet>2241</a>            RESERVED_FIELD_ID_FILE,
+<a href=#2242 id=2242 data-nosnippet>2242</a>        ]);
+<a href=#2243 id=2243 data-nosnippet>2243</a>        <span class="kw">let </span>(batches, <span class="kw">_</span>) = scan_task(task).<span class="kw">await</span>;
+<a href=#2244 id=2244 data-nosnippet>2244</a>
+<a href=#2245 id=2245 data-nosnippet>2245</a>        <span class="comment">// Both metadata columns materialize; no data column is read.
+<a href=#2246 id=2246 data-nosnippet>2246</a>        </span><span class="macro">assert_eq!</span>(batches[<span class="number">0</span>].num_columns(), <span class="number">2</span>);
+<a href=#2247 id=2247 data-nosnippet>2247</a>        <span class="kw">let </span>pos_col = batches[<span class="number">0</span>]
+<a href=#2248 id=2248 data-nosnippet>2248</a>            .column_by_name(RESERVED_COL_NAME_POS)
+<a href=#2249 id=2249 data-nosnippet>2249</a>            .expect(<span class="string">"_pos column should be present"</span>)
+<a href=#2250 id=2250 data-nosnippet>2250</a>            .as_primitive::&lt;arrow_array::types::Int64Type&gt;();
+<a href=#2251 id=2251 data-nosnippet>2251</a>        <span class="macro">assert_eq!</span>(pos_col.values(), <span class="kw-2">&amp;</span>[<span class="number">0</span>, <span class="number">1</span>, <span class="number">2</span>]);
+<a href=#2252 id=2252 data-nosnippet>2252</a>        <span class="kw">let </span>file_col = batches[<span class="number">0</span>]
+<a href=#2253 id=2253 data-nosnippet>2253</a>            .column_by_name(RESERVED_COL_NAME_FILE)
+<a href=#2254 id=2254 data-nosnippet>2254</a>            .expect(<span class="string">"_file column should be present"</span>);
+<a href=#2255 id=2255 data-nosnippet>2255</a>        <span class="kw">let </span>file_col = cast(file_col, <span class="kw-2">&amp;</span>DataType::Utf8).unwrap();
+<a href=#2256 id=2256 data-nosnippet>2256</a>        <span class="kw">let </span>file_col = file_col.as_any().downcast_ref::&lt;StringArray&gt;().unwrap();
+<a href=#2257 id=2257 data-nosnippet>2257</a>        <span class="macro">assert_eq!</span>(file_col.value(<span class="number">0</span>), file_path);
+<a href=#2258 id=2258 data-nosnippet>2258</a>    }
+<a href=#2259 id=2259 data-nosnippet>2259</a>
+<a href=#2260 id=2260 data-nosnippet>2260</a>    <span class="attr">#[tokio::test]
+<a href=#2261 id=2261 data-nosnippet>2261</a>    </span><span class="kw">async fn </span>test_pos_and_physical_seq_projection_reads_only_the_leaf() {
+<a href=#2262 id=2262 data-nosnippet>2262</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#2263 id=2263 data-nosnippet>2263</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2264 id=2264 data-nosnippet>2264</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2265 id=2265 data-nosnippet>2265</a>        };
+<a href=#2266 id=2266 data-nosnippet>2266</a>
+<a href=#2267 id=2267 data-nosnippet>2267</a>        <span class="comment">// A v3 rewrite that carried rows forward stores `_last_updated_sequence_number`
+<a href=#2268 id=2268 data-nosnippet>2268</a>        // per-row. Projecting only `_pos` + the sequence column must read just that one
+<a href=#2269 id=2269 data-nosnippet>2269</a>        // physical leaf, not every data column.
+<a href=#2270 id=2270 data-nosnippet>2270</a>        </span><span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2271 id=2271 data-nosnippet>2271</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2272 id=2272 data-nosnippet>2272</a>
+<a href=#2273 id=2273 data-nosnippet>2273</a>        <span class="comment">// File: id (1), wide data column (2), physical _last_updated_sequence_number.
+<a href=#2274 id=2274 data-nosnippet>2274</a>        </span><span class="kw">let </span>write = |name: <span class="kw-2">&amp;</span>str| {
+<a href=#2275 id=2275 data-nosnippet>2275</a>            write_parquet_with_wide_column(
+<a href=#2276 id=2276 data-nosnippet>2276</a>                dir,
+<a href=#2277 id=2277 data-nosnippet>2277</a>                name,
+<a href=#2278 id=2278 data-nosnippet>2278</a>                <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#2279 id=2279 data-nosnippet>2279</a>                <span class="macro">vec!</span>[Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef],
+<a href=#2280 id=2280 data-nosnippet>2280</a>            )
+<a href=#2281 id=2281 data-nosnippet>2281</a>        };
+<a href=#2282 id=2282 data-nosnippet>2282</a>
+<a href=#2283 id=2283 data-nosnippet>2283</a>        <span class="kw">let </span>seq_task = |path: String, ids: Vec&lt;i32&gt;| {
+<a href=#2284 id=2284 data-nosnippet>2284</a>            FileScanTask::builder()
+<a href=#2285 id=2285 data-nosnippet>2285</a>                .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>path).unwrap().len())
+<a href=#2286 id=2286 data-nosnippet>2286</a>                .with_start(<span class="number">0</span>)
+<a href=#2287 id=2287 data-nosnippet>2287</a>                .with_length(<span class="number">0</span>)
+<a href=#2288 id=2288 data-nosnippet>2288</a>                .with_data_file_path(path)
+<a href=#2289 id=2289 data-nosnippet>2289</a>                .with_data_file_format(DataFileFormat::Parquet)
+<a href=#2290 id=2290 data-nosnippet>2290</a>                .with_schema(id_and_wide_schema())
+<a href=#2291 id=2291 data-nosnippet>2291</a>                .with_project_field_ids(ids)
+<a href=#2292 id=2292 data-nosnippet>2292</a>                .with_first_row_id(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#2293 id=2293 data-nosnippet>2293</a>                .with_data_sequence_number(<span class="prelude-val">Some</span>(<span class="number">9</span>))
+<a href=#2294 id=2294 data-nosnippet>2294</a>                .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#2295 id=2295 data-nosnippet>2295</a>                .build()
+<a href=#2296 id=2296 data-nosnippet>2296</a>        };
+<a href=#2297 id=2297 data-nosnippet>2297</a>
+<a href=#2298 id=2298 data-nosnippet>2298</a>        <span class="kw">let </span>meta_only = seq_task(write(<span class="string">"pos_seq.parquet"</span>), <span class="macro">vec!</span>[
+<a href=#2299 id=2299 data-nosnippet>2299</a>            RESERVED_FIELD_ID_POS,
+<a href=#2300 id=2300 data-nosnippet>2300</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2301 id=2301 data-nosnippet>2301</a>        ]);
+<a href=#2302 id=2302 data-nosnippet>2302</a>        <span class="kw">let </span>(batches, meta_only_bytes) = scan_task(meta_only).<span class="kw">await</span>;
+<a href=#2303 id=2303 data-nosnippet>2303</a>
+<a href=#2304 id=2304 data-nosnippet>2304</a>        <span class="comment">// `_pos` and the coalesced sequence column materialize; the wide column does not.
+<a href=#2305 id=2305 data-nosnippet>2305</a>        </span><span class="kw">let </span>pos_col = batches[<span class="number">0</span>]
+<a href=#2306 id=2306 data-nosnippet>2306</a>            .column_by_name(RESERVED_COL_NAME_POS)
+<a href=#2307 id=2307 data-nosnippet>2307</a>            .expect(<span class="string">"_pos column should be present"</span>)
+<a href=#2308 id=2308 data-nosnippet>2308</a>            .as_primitive::&lt;arrow_array::types::Int64Type&gt;();
+<a href=#2309 id=2309 data-nosnippet>2309</a>        <span class="macro">assert_eq!</span>(pos_col.values(), <span class="kw-2">&amp;</span>[<span class="number">0</span>, <span class="number">1</span>, <span class="number">2</span>]);
+<a href=#2310 id=2310 data-nosnippet>2310</a>        <span class="kw">let </span>seq_col = batches[<span class="number">0</span>]
+<a href=#2311 id=2311 data-nosnippet>2311</a>            .column_by_name(RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER)
+<a href=#2312 id=2312 data-nosnippet>2312</a>            .expect(<span class="string">"_last_updated_sequence_number column should be present"</span>);
+<a href=#2313 id=2313 data-nosnippet>2313</a>        <span class="kw">let </span>seq_col = cast(seq_col, <span class="kw-2">&amp;</span>DataType::Int64).unwrap();
+<a href=#2314 id=2314 data-nosnippet>2314</a>        <span class="kw">let </span>seq_col = seq_col.as_any().downcast_ref::&lt;Int64Array&gt;().unwrap();
+<a href=#2315 id=2315 data-nosnippet>2315</a>        <span class="comment">// Per-row stored value where non-null, else the data sequence number (9).
+<a href=#2316 id=2316 data-nosnippet>2316</a>        </span><span class="macro">assert_eq!</span>(seq_col.value(<span class="number">0</span>), <span class="number">5</span>);
+<a href=#2317 id=2317 data-nosnippet>2317</a>        <span class="macro">assert_eq!</span>(seq_col.value(<span class="number">1</span>), <span class="number">9</span>);
+<a href=#2318 id=2318 data-nosnippet>2318</a>        <span class="macro">assert_eq!</span>(seq_col.value(<span class="number">2</span>), <span class="number">8</span>);
+<a href=#2319 id=2319 data-nosnippet>2319</a>        <span class="macro">assert!</span>(batches[<span class="number">0</span>].column_by_name(<span class="string">"wide"</span>).is_none());
+<a href=#2320 id=2320 data-nosnippet>2320</a>
+<a href=#2321 id=2321 data-nosnippet>2321</a>        <span class="comment">// A scan that also projects the wide data column must read materially more,
+<a href=#2322 id=2322 data-nosnippet>2322</a>        // proving the metadata-only scan pruned to just the sequence leaf.
+<a href=#2323 id=2323 data-nosnippet>2323</a>        </span><span class="kw">let </span>with_data = seq_task(write(<span class="string">"pos_seq_ref.parquet"</span>), <span class="macro">vec!</span>[
+<a href=#2324 id=2324 data-nosnippet>2324</a>            <span class="number">2</span>,
+<a href=#2325 id=2325 data-nosnippet>2325</a>            RESERVED_FIELD_ID_POS,
+<a href=#2326 id=2326 data-nosnippet>2326</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2327 id=2327 data-nosnippet>2327</a>        ]);
+<a href=#2328 id=2328 data-nosnippet>2328</a>        <span class="kw">let </span>(<span class="kw">_</span>, with_data_bytes) = scan_task(with_data).<span class="kw">await</span>;
+<a href=#2329 id=2329 data-nosnippet>2329</a>
+<a href=#2330 id=2330 data-nosnippet>2330</a>        <span class="macro">assert!</span>(
+<a href=#2331 id=2331 data-nosnippet>2331</a>            meta_only_bytes &lt; with_data_bytes,
+<a href=#2332 id=2332 data-nosnippet>2332</a>            <span class="string">"_pos + physical sequence scan should read fewer bytes than one that also \
+<a href=#2333 id=2333 data-nosnippet>2333</a>             reads the wide column: {meta_only_bytes} vs {with_data_bytes}"
+<a href=#2334 id=2334 data-nosnippet>2334</a>        </span>);
+<a href=#2335 id=2335 data-nosnippet>2335</a>    }
+<a href=#2336 id=2336 data-nosnippet>2336</a>
+<a href=#2337 id=2337 data-nosnippet>2337</a>    <span class="attr">#[tokio::test]
+<a href=#2338 id=2338 data-nosnippet>2338</a>    </span><span class="kw">async fn </span>test_seq_only_projection_reads_only_the_leaf() {
+<a href=#2339 id=2339 data-nosnippet>2339</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#2340 id=2340 data-nosnippet>2340</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2341 id=2341 data-nosnippet>2341</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2342 id=2342 data-nosnippet>2342</a>        };
+<a href=#2343 id=2343 data-nosnippet>2343</a>
+<a href=#2344 id=2344 data-nosnippet>2344</a>        <span class="comment">// `_last_updated_sequence_number` alone (no `_pos`, no data column). The physical
+<a href=#2345 id=2345 data-nosnippet>2345</a>        // leaf is the sole row source, so it -- not the RowNumber virtual column -- must
+<a href=#2346 id=2346 data-nosnippet>2346</a>        // drive the `none()` downgrade and prune the read to just that leaf.
+<a href=#2347 id=2347 data-nosnippet>2347</a>        </span><span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2348 id=2348 data-nosnippet>2348</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2349 id=2349 data-nosnippet>2349</a>        <span class="kw">let </span>write = |name: <span class="kw-2">&amp;</span>str| {
+<a href=#2350 id=2350 data-nosnippet>2350</a>            write_parquet_with_wide_column(
+<a href=#2351 id=2351 data-nosnippet>2351</a>                dir,
+<a href=#2352 id=2352 data-nosnippet>2352</a>                name,
+<a href=#2353 id=2353 data-nosnippet>2353</a>                <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#2354 id=2354 data-nosnippet>2354</a>                <span class="macro">vec!</span>[Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="number">8</span>)])) <span class="kw">as </span>ArrayRef],
+<a href=#2355 id=2355 data-nosnippet>2355</a>            )
+<a href=#2356 id=2356 data-nosnippet>2356</a>        };
+<a href=#2357 id=2357 data-nosnippet>2357</a>        <span class="kw">let </span>seq_task = |path: String, ids: Vec&lt;i32&gt;| {
+<a href=#2358 id=2358 data-nosnippet>2358</a>            FileScanTask::builder()
+<a href=#2359 id=2359 data-nosnippet>2359</a>                .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>path).unwrap().len())
+<a href=#2360 id=2360 data-nosnippet>2360</a>                .with_start(<span class="number">0</span>)
+<a href=#2361 id=2361 data-nosnippet>2361</a>                .with_length(<span class="number">0</span>)
+<a href=#2362 id=2362 data-nosnippet>2362</a>                .with_data_file_path(path)
+<a href=#2363 id=2363 data-nosnippet>2363</a>                .with_data_file_format(DataFileFormat::Parquet)
+<a href=#2364 id=2364 data-nosnippet>2364</a>                .with_schema(id_and_wide_schema())
+<a href=#2365 id=2365 data-nosnippet>2365</a>                .with_project_field_ids(ids)
+<a href=#2366 id=2366 data-nosnippet>2366</a>                .with_first_row_id(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#2367 id=2367 data-nosnippet>2367</a>                .with_data_sequence_number(<span class="prelude-val">Some</span>(<span class="number">9</span>))
+<a href=#2368 id=2368 data-nosnippet>2368</a>                .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#2369 id=2369 data-nosnippet>2369</a>                .build()
+<a href=#2370 id=2370 data-nosnippet>2370</a>        };
+<a href=#2371 id=2371 data-nosnippet>2371</a>
+<a href=#2372 id=2372 data-nosnippet>2372</a>        <span class="kw">let </span>meta_only = seq_task(write(<span class="string">"seq_only.parquet"</span>), <span class="macro">vec!</span>[
+<a href=#2373 id=2373 data-nosnippet>2373</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2374 id=2374 data-nosnippet>2374</a>        ]);
+<a href=#2375 id=2375 data-nosnippet>2375</a>        <span class="kw">let </span>(batches, meta_only_bytes) = scan_task(meta_only).<span class="kw">await</span>;
+<a href=#2376 id=2376 data-nosnippet>2376</a>
+<a href=#2377 id=2377 data-nosnippet>2377</a>        <span class="kw">let </span>seq_col = batches[<span class="number">0</span>]
+<a href=#2378 id=2378 data-nosnippet>2378</a>            .column_by_name(RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER)
+<a href=#2379 id=2379 data-nosnippet>2379</a>            .expect(<span class="string">"_last_updated_sequence_number column should be present"</span>);
+<a href=#2380 id=2380 data-nosnippet>2380</a>        <span class="kw">let </span>seq_col = cast(seq_col, <span class="kw-2">&amp;</span>DataType::Int64).unwrap();
+<a href=#2381 id=2381 data-nosnippet>2381</a>        <span class="kw">let </span>seq_col = seq_col.as_any().downcast_ref::&lt;Int64Array&gt;().unwrap();
+<a href=#2382 id=2382 data-nosnippet>2382</a>        <span class="macro">assert_eq!</span>(seq_col.value(<span class="number">0</span>), <span class="number">5</span>);
+<a href=#2383 id=2383 data-nosnippet>2383</a>        <span class="macro">assert_eq!</span>(seq_col.value(<span class="number">1</span>), <span class="number">9</span>);
+<a href=#2384 id=2384 data-nosnippet>2384</a>        <span class="macro">assert_eq!</span>(seq_col.value(<span class="number">2</span>), <span class="number">8</span>);
+<a href=#2385 id=2385 data-nosnippet>2385</a>        <span class="kw">let </span>total_rows: usize = batches.iter().map(|b| b.num_rows()).sum();
+<a href=#2386 id=2386 data-nosnippet>2386</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">3</span>);
+<a href=#2387 id=2387 data-nosnippet>2387</a>        <span class="macro">assert!</span>(batches[<span class="number">0</span>].column_by_name(<span class="string">"wide"</span>).is_none());
+<a href=#2388 id=2388 data-nosnippet>2388</a>
+<a href=#2389 id=2389 data-nosnippet>2389</a>        <span class="kw">let </span>with_data = seq_task(write(<span class="string">"seq_only_ref.parquet"</span>), <span class="macro">vec!</span>[
+<a href=#2390 id=2390 data-nosnippet>2390</a>            <span class="number">2</span>,
+<a href=#2391 id=2391 data-nosnippet>2391</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2392 id=2392 data-nosnippet>2392</a>        ]);
+<a href=#2393 id=2393 data-nosnippet>2393</a>        <span class="kw">let </span>(<span class="kw">_</span>, with_data_bytes) = scan_task(with_data).<span class="kw">await</span>;
+<a href=#2394 id=2394 data-nosnippet>2394</a>
+<a href=#2395 id=2395 data-nosnippet>2395</a>        <span class="macro">assert!</span>(
+<a href=#2396 id=2396 data-nosnippet>2396</a>            meta_only_bytes &lt; with_data_bytes,
+<a href=#2397 id=2397 data-nosnippet>2397</a>            <span class="string">"seq-only scan should read fewer bytes than one that also reads the wide \
+<a href=#2398 id=2398 data-nosnippet>2398</a>             column: {meta_only_bytes} vs {with_data_bytes}"
+<a href=#2399 id=2399 data-nosnippet>2399</a>        </span>);
+<a href=#2400 id=2400 data-nosnippet>2400</a>    }
+<a href=#2401 id=2401 data-nosnippet>2401</a>
+<a href=#2402 id=2402 data-nosnippet>2402</a>    <span class="attr">#[tokio::test]
+<a href=#2403 id=2403 data-nosnippet>2403</a>    </span><span class="kw">async fn </span>test_seq_only_projection_null_first_row_id_preserves_row_count() {
+<a href=#2404 id=2404 data-nosnippet>2404</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::{
+<a href=#2405 id=2405 data-nosnippet>2405</a>            RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2406 id=2406 data-nosnippet>2406</a>            RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER,
+<a href=#2407 id=2407 data-nosnippet>2407</a>        };
+<a href=#2408 id=2408 data-nosnippet>2408</a>
+<a href=#2409 id=2409 data-nosnippet>2409</a>        <span class="comment">// Seq-only projection with a null first_row_id: the column is nulled and the
+<a href=#2410 id=2410 data-nosnippet>2410</a>        // physical leaf is NOT read (the gated `coalesce_last_updated_seq_leaf` is None).
+<a href=#2411 id=2411 data-nosnippet>2411</a>        // The downgrade must therefore not fire -- keying off the raw `project_*` flag
+<a href=#2412 id=2412 data-nosnippet>2412</a>        // instead would drop the only readable column and lose the row count.
+<a href=#2413 id=2413 data-nosnippet>2413</a>        </span><span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2414 id=2414 data-nosnippet>2414</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2415 id=2415 data-nosnippet>2415</a>        <span class="kw">let </span>file_path = write_parquet_with_wide_column(
+<a href=#2416 id=2416 data-nosnippet>2416</a>            dir,
+<a href=#2417 id=2417 data-nosnippet>2417</a>            <span class="string">"seq_only_null_first.parquet"</span>,
+<a href=#2418 id=2418 data-nosnippet>2418</a>            <span class="macro">vec!</span>[physical_last_updated_seq_field()],
+<a href=#2419 id=2419 data-nosnippet>2419</a>            <span class="macro">vec!</span>[Arc::new(Int64Array::from(<span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="number">5</span>), <span class="prelude-val">Some</span>(<span class="number">6</span>), <span class="prelude-val">Some</span>(<span class="number">7</span>)])) <span class="kw">as </span>ArrayRef],
+<a href=#2420 id=2420 data-nosnippet>2420</a>        );
+<a href=#2421 id=2421 data-nosnippet>2421</a>        <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#2422 id=2422 data-nosnippet>2422</a>            .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&amp;</span>file_path).unwrap().len())
+<a href=#2423 id=2423 data-nosnippet>2423</a>            .with_start(<span class="number">0</span>)
+<a href=#2424 id=2424 data-nosnippet>2424</a>            .with_length(<span class="number">0</span>)
+<a href=#2425 id=2425 data-nosnippet>2425</a>            .with_data_file_path(file_path)
+<a href=#2426 id=2426 data-nosnippet>2426</a>            .with_data_file_format(DataFileFormat::Parquet)
+<a href=#2427 id=2427 data-nosnippet>2427</a>            .with_schema(id_and_wide_schema())
+<a href=#2428 id=2428 data-nosnippet>2428</a>            .with_project_field_ids(<span class="macro">vec!</span>[RESERVED_FIELD_ID_LAST_UPDATED_SEQUENCE_NUMBER])
+<a href=#2429 id=2429 data-nosnippet>2429</a>            .with_first_row_id(<span class="prelude-val">None</span>)
+<a href=#2430 id=2430 data-nosnippet>2430</a>            .with_data_sequence_number(<span class="prelude-val">Some</span>(<span class="number">9</span>))
+<a href=#2431 id=2431 data-nosnippet>2431</a>            .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#2432 id=2432 data-nosnippet>2432</a>            .build();
+<a href=#2433 id=2433 data-nosnippet>2433</a>        <span class="kw">let </span>(batches, <span class="kw">_</span>) = scan_task(task).<span class="kw">await</span>;
+<a href=#2434 id=2434 data-nosnippet>2434</a>
+<a href=#2435 id=2435 data-nosnippet>2435</a>        <span class="kw">let </span>total_rows: usize = batches.iter().map(|b| b.num_rows()).sum();
+<a href=#2436 id=2436 data-nosnippet>2436</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">3</span>);
+<a href=#2437 id=2437 data-nosnippet>2437</a>        <span class="kw">let </span>seq_col = batches[<span class="number">0</span>]
+<a href=#2438 id=2438 data-nosnippet>2438</a>            .column_by_name(RESERVED_COL_NAME_LAST_UPDATED_SEQUENCE_NUMBER)
+<a href=#2439 id=2439 data-nosnippet>2439</a>            .expect(<span class="string">"_last_updated_sequence_number column should be present"</span>);
+<a href=#2440 id=2440 data-nosnippet>2440</a>        <span class="kw">let </span>seq_col = cast(seq_col, <span class="kw-2">&amp;</span>DataType::Int64).unwrap();
+<a href=#2441 id=2441 data-nosnippet>2441</a>        <span class="kw">let </span>seq_col = seq_col.as_any().downcast_ref::&lt;Int64Array&gt;().unwrap();
+<a href=#2442 id=2442 data-nosnippet>2442</a>        <span class="macro">assert!</span>((<span class="number">0</span>..<span class="number">3</span>).all(|i| seq_col.is_null(i)));
+<a href=#2443 id=2443 data-nosnippet>2443</a>    }
+<a href=#2444 id=2444 data-nosnippet>2444</a>
+<a href=#2445 id=2445 data-nosnippet>2445</a>    <span class="attr">#[tokio::test]
+<a href=#2446 id=2446 data-nosnippet>2446</a>    </span><span class="kw">async fn </span>test_file_only_projection_preserves_row_count() {
+<a href=#2447 id=2447 data-nosnippet>2447</a>        <span class="kw">use </span><span class="kw">crate</span>::metadata_columns::RESERVED_COL_NAME_FILE;
+<a href=#2448 id=2448 data-nosnippet>2448</a>
+<a href=#2449 id=2449 data-nosnippet>2449</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2450 id=2450 data-nosnippet>2450</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2451 id=2451 data-nosnippet>2451</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"file_only.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#2452 id=2452 data-nosnippet>2452</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#2453 id=2453 data-nosnippet>2453</a>            Schema::builder()
+<a href=#2454 id=2454 data-nosnippet>2454</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#2455 id=2455 data-nosnippet>2455</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#2456 id=2456 data-nosnippet>2456</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#2457 id=2457 data-nosnippet>2457</a>                ])
+<a href=#2458 id=2458 data-nosnippet>2458</a>                .build()
+<a href=#2459 id=2459 data-nosnippet>2459</a>                .unwrap(),
+<a href=#2460 id=2460 data-nosnippet>2460</a>        );
+<a href=#2461 id=2461 data-nosnippet>2461</a>        <span class="kw">let </span>task =
+<a href=#2462 id=2462 data-nosnippet>2462</a>            metadata_projection_task(file_path.clone(), schema, <span class="macro">vec!</span>[RESERVED_FIELD_ID_FILE]);
+<a href=#2463 id=2463 data-nosnippet>2463</a>        <span class="kw">let </span>(batches, <span class="kw">_</span>) = scan_task(task).<span class="kw">await</span>;
+<a href=#2464 id=2464 data-nosnippet>2464</a>
+<a href=#2465 id=2465 data-nosnippet>2465</a>        <span class="comment">// A pure-constant projection has no independent row source, so the row count must
+<a href=#2466 id=2466 data-nosnippet>2466</a>        // still come from the file (the `empty -&gt; all()` path is preserved for this case).
+<a href=#2467 id=2467 data-nosnippet>2467</a>        </span><span class="kw">let </span>total_rows: usize = batches.iter().map(|b| b.num_rows()).sum();
+<a href=#2468 id=2468 data-nosnippet>2468</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">3</span>);
+<a href=#2469 id=2469 data-nosnippet>2469</a>        <span class="kw">let </span>file_col = batches[<span class="number">0</span>]
+<a href=#2470 id=2470 data-nosnippet>2470</a>            .column_by_name(RESERVED_COL_NAME_FILE)
+<a href=#2471 id=2471 data-nosnippet>2471</a>            .expect(<span class="string">"_file column should be present"</span>);
+<a href=#2472 id=2472 data-nosnippet>2472</a>        <span class="kw">let </span>file_col = cast(file_col, <span class="kw-2">&amp;</span>DataType::Utf8).unwrap();
+<a href=#2473 id=2473 data-nosnippet>2473</a>        <span class="kw">let </span>file_col = file_col.as_any().downcast_ref::&lt;StringArray&gt;().unwrap();
+<a href=#2474 id=2474 data-nosnippet>2474</a>        <span class="macro">assert_eq!</span>(file_col.value(<span class="number">0</span>), file_path);
+<a href=#2475 id=2475 data-nosnippet>2475</a>    }
+<a href=#2476 id=2476 data-nosnippet>2476</a>
+<a href=#2477 id=2477 data-nosnippet>2477</a>    <span class="attr">#[tokio::test]
+<a href=#2478 id=2478 data-nosnippet>2478</a>    </span><span class="kw">async fn </span>test_empty_projection_preserves_row_count() {
+<a href=#2479 id=2479 data-nosnippet>2479</a>        <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#2480 id=2480 data-nosnippet>2480</a>        <span class="kw">let </span>dir = tmp_dir.path().to_str().unwrap();
+<a href=#2481 id=2481 data-nosnippet>2481</a>        <span class="kw">let </span>file_path = write_plain_parquet(dir, <span class="string">"empty_projection.parquet"</span>, <span class="macro">vec!</span>[], <span class="macro">vec!</span>[]);
+<a href=#2482 id=2482 data-nosnippet>2482</a>        <span class="kw">let </span>schema = Arc::new(
+<a href=#2483 id=2483 data-nosnippet>2483</a>            Schema::builder()
+<a href=#2484 id=2484 data-nosnippet>2484</a>                .with_schema_id(<span class="number">1</span>)
+<a href=#2485 id=2485 data-nosnippet>2485</a>                .with_fields(<span class="macro">vec!</span>[
+<a href=#2486 id=2486 data-nosnippet>2486</a>                    NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#2487 id=2487 data-nosnippet>2487</a>                ])
+<a href=#2488 id=2488 data-nosnippet>2488</a>                .build()
+<a href=#2489 id=2489 data-nosnippet>2489</a>                .unwrap(),
+<a href=#2490 id=2490 data-nosnippet>2490</a>        );
+<a href=#2491 id=2491 data-nosnippet>2491</a>        <span class="kw">let </span>task = metadata_projection_task(file_path, schema, <span class="macro">vec!</span>[]);
+<a href=#2492 id=2492 data-nosnippet>2492</a>        <span class="kw">let </span>(batches, <span class="kw">_</span>) = scan_task(task).<span class="kw">await</span>;
+<a href=#2493 id=2493 data-nosnippet>2493</a>
+<a href=#2494 id=2494 data-nosnippet>2494</a>        <span class="comment">// A bare COUNT(*)-style empty projection must still report the row count.
+<a href=#2495 id=2495 data-nosnippet>2495</a>        </span><span class="kw">let </span>total_rows: usize = batches.iter().map(|b| b.num_rows()).sum();
+<a href=#2496 id=2496 data-nosnippet>2496</a>        <span class="macro">assert_eq!</span>(total_rows, <span class="number">3</span>);
+<a href=#2497 id=2497 data-nosnippet>2497</a>    }
+<a href=#2498 id=2498 data-nosnippet>2498</a>}
 </code></pre></div></section></main></body></html>
\ No newline at end of file