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 2026/06/11 13:06:08 UTC
(iceberg-rust) branch gh-pages updated: deploy: 80a30d3440928e1456fd116db934d829844a1a42
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 12aa00206 deploy: 80a30d3440928e1456fd116db934d829844a1a42
12aa00206 is described below
commit 12aa0020665d221f6035a6b35b25bf15d8b8eb38
Author: blackmwk <bl...@users.noreply.github.com>
AuthorDate: Thu Jun 11 13:06:02 2026 +0000
deploy: 80a30d3440928e1456fd116db934d829844a1a42
---
api/src/iceberg/arrow/reader/row_filter.rs.html | 1055 ++++++++++++++---------
1 file changed, 638 insertions(+), 417 deletions(-)
diff --git a/api/src/iceberg/arrow/reader/row_filter.rs.html b/api/src/iceberg/arrow/reader/row_filter.rs.html
index c98e82de6..d4b83210a 100644
--- a/api/src/iceberg/arrow/reader/row_filter.rs.html
+++ b/api/src/iceberg/arrow/reader/row_filter.rs.html
@@ -160,449 +160,670 @@
<a href=#160 id=160 data-nosnippet>160</a>
<a href=#161 id=161 data-nosnippet>161</a> <span class="doccomment">/// Filters row groups by byte range to support Iceberg's file splitting.
<a href=#162 id=162 data-nosnippet>162</a> ///
-<a href=#163 id=163 data-nosnippet>163</a> /// Iceberg splits large files at row group boundaries, so we only read row groups
-<a href=#164 id=164 data-nosnippet>164</a> /// whose byte ranges overlap with [start, start+length).
-<a href=#165 id=165 data-nosnippet>165</a> </span><span class="kw">pub</span>(<span class="kw">super</span>) <span class="kw">fn </span>filter_row_groups_by_byte_range(
-<a href=#166 id=166 data-nosnippet>166</a> parquet_metadata: <span class="kw-2">&</span>Arc<ParquetMetaData>,
-<a href=#167 id=167 data-nosnippet>167</a> start: u64,
-<a href=#168 id=168 data-nosnippet>168</a> length: u64,
-<a href=#169 id=169 data-nosnippet>169</a> ) -> <span class="prelude-ty">Result</span><Vec<usize>> {
-<a href=#170 id=170 data-nosnippet>170</a> <span class="kw">let </span>row_groups = parquet_metadata.row_groups();
-<a href=#171 id=171 data-nosnippet>171</a> <span class="kw">let </span><span class="kw-2">mut </span>selected = Vec::new();
-<a href=#172 id=172 data-nosnippet>172</a> <span class="kw">let </span>end = start + length;
-<a href=#173 id=173 data-nosnippet>173</a>
-<a href=#174 id=174 data-nosnippet>174</a> <span class="comment">// Row groups are stored sequentially after the 4-byte magic header.
-<a href=#175 id=175 data-nosnippet>175</a> </span><span class="kw">let </span><span class="kw-2">mut </span>current_byte_offset = <span class="number">4u64</span>;
-<a href=#176 id=176 data-nosnippet>176</a>
-<a href=#177 id=177 data-nosnippet>177</a> <span class="kw">for </span>(idx, row_group) <span class="kw">in </span>row_groups.iter().enumerate() {
-<a href=#178 id=178 data-nosnippet>178</a> <span class="kw">let </span>row_group_size = row_group.compressed_size() <span class="kw">as </span>u64;
-<a href=#179 id=179 data-nosnippet>179</a> <span class="kw">let </span>row_group_end = current_byte_offset + row_group_size;
-<a href=#180 id=180 data-nosnippet>180</a>
-<a href=#181 id=181 data-nosnippet>181</a> <span class="kw">if </span>current_byte_offset < end && start < row_group_end {
-<a href=#182 id=182 data-nosnippet>182</a> selected.push(idx);
-<a href=#183 id=183 data-nosnippet>183</a> }
+<a href=#163 id=163 data-nosnippet>163</a> /// Engines split a data file into multiple scan tasks, each covering a byte range
+<a href=#164 id=164 data-nosnippet>164</a> /// `[start, start+length)`. Normally Iceberg planning aligns these splits to row group
+<a href=#165 id=165 data-nosnippet>165</a> /// boundaries using the data file's `split_offsets` metadata, so a task's range never
+<a href=#166 id=166 data-nosnippet>166</a> /// bisects a row group. But when `split_offsets` is missing (e.g. a manually written or
+<a href=#167 id=167 data-nosnippet>167</a> /// non-conforming file), planning falls back to tiling the file at the requested split
+<a href=#168 id=168 data-nosnippet>168</a> /// size, and a task's range can land in the middle of a row group.
+<a href=#169 id=169 data-nosnippet>169</a> ///
+<a href=#170 id=170 data-nosnippet>170</a> /// A row group must be read by exactly one task, otherwise its rows are duplicated. We
+<a href=#171 id=171 data-nosnippet>171</a> /// assign ownership by the row group's midpoint: a task owns a row group only if its range
+<a href=#172 id=172 data-nosnippet>172</a> /// contains that midpoint. Because the tasks tile the file contiguously and disjointly,
+<a href=#173 id=173 data-nosnippet>173</a> /// each midpoint falls in exactly one task. This matches parquet-mr's `BlockMetaData`
+<a href=#174 id=174 data-nosnippet>174</a> /// midpoint semantics. For a whole-file task (`start=0, length=fileSize`, as iceberg-rust's
+<a href=#175 id=175 data-nosnippet>175</a> /// own planner emits) every midpoint lies in range, so all row groups are selected.
+<a href=#176 id=176 data-nosnippet>176</a> </span><span class="kw">pub</span>(<span class="kw">super</span>) <span class="kw">fn </span>filter_row_groups_by_byte_range(
+<a href=#177 id=177 data-nosnippet>177</a> parquet_metadata: <span class="kw-2">&</span>Arc<ParquetMetaData>,
+<a href=#178 id=178 data-nosnippet>178</a> start: u64,
+<a href=#179 id=179 data-nosnippet>179</a> length: u64,
+<a href=#180 id=180 data-nosnippet>180</a> ) -> <span class="prelude-ty">Result</span><Vec<usize>> {
+<a href=#181 id=181 data-nosnippet>181</a> <span class="kw">let </span>row_groups = parquet_metadata.row_groups();
+<a href=#182 id=182 data-nosnippet>182</a> <span class="kw">let </span><span class="kw-2">mut </span>selected = Vec::new();
+<a href=#183 id=183 data-nosnippet>183</a> <span class="kw">let </span>end = start + length;
<a href=#184 id=184 data-nosnippet>184</a>
-<a href=#185 id=185 data-nosnippet>185</a> current_byte_offset = row_group_end;
-<a href=#186 id=186 data-nosnippet>186</a> }
+<a href=#185 id=185 data-nosnippet>185</a> <span class="comment">// Row groups are stored sequentially after the 4-byte magic header.
+<a href=#186 id=186 data-nosnippet>186</a> </span><span class="kw">let </span><span class="kw-2">mut </span>current_byte_offset = <span class="number">4u64</span>;
<a href=#187 id=187 data-nosnippet>187</a>
-<a href=#188 id=188 data-nosnippet>188</a> <span class="prelude-val">Ok</span>(selected)
-<a href=#189 id=189 data-nosnippet>189</a> }
-<a href=#190 id=190 data-nosnippet>190</a>}
+<a href=#188 id=188 data-nosnippet>188</a> <span class="kw">for </span>(idx, row_group) <span class="kw">in </span>row_groups.iter().enumerate() {
+<a href=#189 id=189 data-nosnippet>189</a> <span class="kw">let </span>row_group_size = row_group.compressed_size() <span class="kw">as </span>u64;
+<a href=#190 id=190 data-nosnippet>190</a> <span class="kw">let </span>row_group_midpoint = current_byte_offset + row_group_size / <span class="number">2</span>;
<a href=#191 id=191 data-nosnippet>191</a>
-<a href=#192 id=192 data-nosnippet>192</a><span class="attr">#[cfg(test)]
-<a href=#193 id=193 data-nosnippet>193</a></span><span class="kw">mod </span>tests {
-<a href=#194 id=194 data-nosnippet>194</a> <span class="kw">use </span>std::collections::HashMap;
-<a href=#195 id=195 data-nosnippet>195</a> <span class="kw">use </span>std::fs::File;
-<a href=#196 id=196 data-nosnippet>196</a> <span class="kw">use </span>std::sync::Arc;
+<a href=#192 id=192 data-nosnippet>192</a> <span class="comment">// Half-open ownership: a midpoint on a task boundary belongs to the upper task,
+<a href=#193 id=193 data-nosnippet>193</a> // so exactly one task ever claims a given row group.
+<a href=#194 id=194 data-nosnippet>194</a> </span><span class="kw">if </span>start <= row_group_midpoint && row_group_midpoint < end {
+<a href=#195 id=195 data-nosnippet>195</a> selected.push(idx);
+<a href=#196 id=196 data-nosnippet>196</a> }
<a href=#197 id=197 data-nosnippet>197</a>
-<a href=#198 id=198 data-nosnippet>198</a> <span class="kw">use </span>arrow_array::cast::AsArray;
-<a href=#199 id=199 data-nosnippet>199</a> <span class="kw">use </span>arrow_array::{ArrayRef, LargeStringArray, RecordBatch, StringArray};
-<a href=#200 id=200 data-nosnippet>200</a> <span class="kw">use </span>arrow_schema::{DataType, Field, Schema <span class="kw">as </span>ArrowSchema};
-<a href=#201 id=201 data-nosnippet>201</a> <span class="kw">use </span>futures::TryStreamExt;
-<a href=#202 id=202 data-nosnippet>202</a> <span class="kw">use </span>parquet::arrow::{ArrowWriter, PARQUET_FIELD_ID_META_KEY};
-<a href=#203 id=203 data-nosnippet>203</a> <span class="kw">use </span>parquet::basic::Compression;
-<a href=#204 id=204 data-nosnippet>204</a> <span class="kw">use </span>parquet::file::properties::WriterProperties;
-<a href=#205 id=205 data-nosnippet>205</a> <span class="kw">use </span>tempfile::TempDir;
-<a href=#206 id=206 data-nosnippet>206</a>
-<a href=#207 id=207 data-nosnippet>207</a> <span class="kw">use </span><span class="kw">crate</span>::Runtime;
-<a href=#208 id=208 data-nosnippet>208</a> <span class="kw">use </span><span class="kw">crate</span>::arrow::{ArrowReader, ArrowReaderBuilder};
-<a href=#209 id=209 data-nosnippet>209</a> <span class="kw">use </span><span class="kw">crate</span>::expr::{Bind, Predicate, Reference};
-<a href=#210 id=210 data-nosnippet>210</a> <span class="kw">use </span><span class="kw">crate</span>::io::FileIO;
-<a href=#211 id=211 data-nosnippet>211</a> <span class="kw">use </span><span class="kw">crate</span>::scan::{FileScanTask, FileScanTaskStream};
-<a href=#212 id=212 data-nosnippet>212</a> <span class="kw">use </span><span class="kw">crate</span>::spec::{DataFileFormat, Datum, NestedField, PrimitiveType, Schema, SchemaRef, Type};
-<a href=#213 id=213 data-nosnippet>213</a>
-<a href=#214 id=214 data-nosnippet>214</a> <span class="kw">async fn </span>test_perform_read(
-<a href=#215 id=215 data-nosnippet>215</a> predicate: Predicate,
-<a href=#216 id=216 data-nosnippet>216</a> schema: SchemaRef,
-<a href=#217 id=217 data-nosnippet>217</a> table_location: String,
-<a href=#218 id=218 data-nosnippet>218</a> reader: ArrowReader,
-<a href=#219 id=219 data-nosnippet>219</a> ) -> Vec<<span class="prelude-ty">Option</span><String>> {
-<a href=#220 id=220 data-nosnippet>220</a> <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(
-<a href=#221 id=221 data-nosnippet>221</a> <span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(FileScanTask::builder()
-<a href=#222 id=222 data-nosnippet>222</a> .with_file_size_in_bytes(
-<a href=#223 id=223 data-nosnippet>223</a> std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>))
-<a href=#224 id=224 data-nosnippet>224</a> .unwrap()
-<a href=#225 id=225 data-nosnippet>225</a> .len(),
-<a href=#226 id=226 data-nosnippet>226</a> )
-<a href=#227 id=227 data-nosnippet>227</a> .with_start(<span class="number">0</span>)
-<a href=#228 id=228 data-nosnippet>228</a> .with_length(<span class="number">0</span>)
-<a href=#229 id=229 data-nosnippet>229</a> .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>))
-<a href=#230 id=230 data-nosnippet>230</a> .with_data_file_format(DataFileFormat::Parquet)
-<a href=#231 id=231 data-nosnippet>231</a> .with_schema(schema.clone())
-<a href=#232 id=232 data-nosnippet>232</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#233 id=233 data-nosnippet>233</a> .with_predicate(<span class="prelude-val">Some</span>(predicate.bind(schema, <span class="bool-val">true</span>).unwrap()))
-<a href=#234 id=234 data-nosnippet>234</a> .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#235 id=235 data-nosnippet>235</a> .build())]
-<a href=#236 id=236 data-nosnippet>236</a> .into_iter(),
-<a href=#237 id=237 data-nosnippet>237</a> )) <span class="kw">as </span>FileScanTaskStream;
-<a href=#238 id=238 data-nosnippet>238</a>
-<a href=#239 id=239 data-nosnippet>239</a> <span class="kw">let </span>result = reader
-<a href=#240 id=240 data-nosnippet>240</a> .read(tasks)
-<a href=#241 id=241 data-nosnippet>241</a> .unwrap()
-<a href=#242 id=242 data-nosnippet>242</a> .stream()
-<a href=#243 id=243 data-nosnippet>243</a> .try_collect::<Vec<RecordBatch>>()
-<a href=#244 id=244 data-nosnippet>244</a> .<span class="kw">await
-<a href=#245 id=245 data-nosnippet>245</a> </span>.unwrap();
-<a href=#246 id=246 data-nosnippet>246</a>
-<a href=#247 id=247 data-nosnippet>247</a> result[<span class="number">0</span>].columns()[<span class="number">0</span>]
-<a href=#248 id=248 data-nosnippet>248</a> .as_string_opt::<i32>()
-<a href=#249 id=249 data-nosnippet>249</a> .unwrap()
-<a href=#250 id=250 data-nosnippet>250</a> .iter()
-<a href=#251 id=251 data-nosnippet>251</a> .map(|v| v.map(ToOwned::to_owned))
-<a href=#252 id=252 data-nosnippet>252</a> .collect::<Vec<<span class="kw">_</span>>>()
-<a href=#253 id=253 data-nosnippet>253</a> }
-<a href=#254 id=254 data-nosnippet>254</a>
-<a href=#255 id=255 data-nosnippet>255</a> <span class="kw">fn </span>setup_kleene_logic(
-<a href=#256 id=256 data-nosnippet>256</a> data_for_col_a: Vec<<span class="prelude-ty">Option</span><String>>,
-<a href=#257 id=257 data-nosnippet>257</a> col_a_type: DataType,
-<a href=#258 id=258 data-nosnippet>258</a> ) -> (FileIO, SchemaRef, String, TempDir) {
-<a href=#259 id=259 data-nosnippet>259</a> <span class="kw">let </span>schema = Arc::new(
-<a href=#260 id=260 data-nosnippet>260</a> Schema::builder()
-<a href=#261 id=261 data-nosnippet>261</a> .with_schema_id(<span class="number">1</span>)
-<a href=#262 id=262 data-nosnippet>262</a> .with_fields(<span class="macro">vec!</span>[
-<a href=#263 id=263 data-nosnippet>263</a> NestedField::optional(<span class="number">1</span>, <span class="string">"a"</span>, Type::Primitive(PrimitiveType::String)).into(),
-<a href=#264 id=264 data-nosnippet>264</a> ])
-<a href=#265 id=265 data-nosnippet>265</a> .build()
-<a href=#266 id=266 data-nosnippet>266</a> .unwrap(),
-<a href=#267 id=267 data-nosnippet>267</a> );
-<a href=#268 id=268 data-nosnippet>268</a>
-<a href=#269 id=269 data-nosnippet>269</a> <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#270 id=270 data-nosnippet>270</a> Field::new(<span class="string">"a"</span>, col_a_type.clone(), <span class="bool-val">true</span>).with_metadata(HashMap::from([(
-<a href=#271 id=271 data-nosnippet>271</a> PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#272 id=272 data-nosnippet>272</a> <span class="string">"1"</span>.to_string(),
-<a href=#273 id=273 data-nosnippet>273</a> )])),
-<a href=#274 id=274 data-nosnippet>274</a> ]));
-<a href=#275 id=275 data-nosnippet>275</a>
-<a href=#276 id=276 data-nosnippet>276</a> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#277 id=277 data-nosnippet>277</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#278 id=278 data-nosnippet>278</a>
-<a href=#279 id=279 data-nosnippet>279</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#280 id=280 data-nosnippet>280</a>
-<a href=#281 id=281 data-nosnippet>281</a> <span class="kw">let </span>col = <span class="kw">match </span>col_a_type {
-<a href=#282 id=282 data-nosnippet>282</a> DataType::Utf8 => Arc::new(StringArray::from(data_for_col_a)) <span class="kw">as </span>ArrayRef,
-<a href=#283 id=283 data-nosnippet>283</a> DataType::LargeUtf8 => Arc::new(LargeStringArray::from(data_for_col_a)) <span class="kw">as </span>ArrayRef,
-<a href=#284 id=284 data-nosnippet>284</a> <span class="kw">_ </span>=> <span class="macro">panic!</span>(<span class="string">"unexpected col_a_type"</span>),
-<a href=#285 id=285 data-nosnippet>285</a> };
-<a href=#286 id=286 data-nosnippet>286</a>
-<a href=#287 id=287 data-nosnippet>287</a> <span class="kw">let </span>to_write = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[col]).unwrap();
+<a href=#198 id=198 data-nosnippet>198</a> current_byte_offset += row_group_size;
+<a href=#199 id=199 data-nosnippet>199</a> }
+<a href=#200 id=200 data-nosnippet>200</a>
+<a href=#201 id=201 data-nosnippet>201</a> <span class="prelude-val">Ok</span>(selected)
+<a href=#202 id=202 data-nosnippet>202</a> }
+<a href=#203 id=203 data-nosnippet>203</a>}
+<a href=#204 id=204 data-nosnippet>204</a>
+<a href=#205 id=205 data-nosnippet>205</a><span class="attr">#[cfg(test)]
+<a href=#206 id=206 data-nosnippet>206</a></span><span class="kw">mod </span>tests {
+<a href=#207 id=207 data-nosnippet>207</a> <span class="kw">use </span>std::collections::HashMap;
+<a href=#208 id=208 data-nosnippet>208</a> <span class="kw">use </span>std::fs::File;
+<a href=#209 id=209 data-nosnippet>209</a> <span class="kw">use </span>std::sync::Arc;
+<a href=#210 id=210 data-nosnippet>210</a>
+<a href=#211 id=211 data-nosnippet>211</a> <span class="kw">use </span>arrow_array::cast::AsArray;
+<a href=#212 id=212 data-nosnippet>212</a> <span class="kw">use </span>arrow_array::{ArrayRef, LargeStringArray, RecordBatch, StringArray};
+<a href=#213 id=213 data-nosnippet>213</a> <span class="kw">use </span>arrow_schema::{DataType, Field, Schema <span class="kw">as </span>ArrowSchema};
+<a href=#214 id=214 data-nosnippet>214</a> <span class="kw">use </span>futures::TryStreamExt;
+<a href=#215 id=215 data-nosnippet>215</a> <span class="kw">use </span>parquet::arrow::{ArrowWriter, PARQUET_FIELD_ID_META_KEY};
+<a href=#216 id=216 data-nosnippet>216</a> <span class="kw">use </span>parquet::basic::Compression;
+<a href=#217 id=217 data-nosnippet>217</a> <span class="kw">use </span>parquet::file::properties::WriterProperties;
+<a href=#218 id=218 data-nosnippet>218</a> <span class="kw">use </span>tempfile::TempDir;
+<a href=#219 id=219 data-nosnippet>219</a>
+<a href=#220 id=220 data-nosnippet>220</a> <span class="kw">use </span><span class="kw">crate</span>::Runtime;
+<a href=#221 id=221 data-nosnippet>221</a> <span class="kw">use </span><span class="kw">crate</span>::arrow::{ArrowReader, ArrowReaderBuilder};
+<a href=#222 id=222 data-nosnippet>222</a> <span class="kw">use </span><span class="kw">crate</span>::expr::{Bind, Predicate, Reference};
+<a href=#223 id=223 data-nosnippet>223</a> <span class="kw">use </span><span class="kw">crate</span>::io::FileIO;
+<a href=#224 id=224 data-nosnippet>224</a> <span class="kw">use </span><span class="kw">crate</span>::scan::{FileScanTask, FileScanTaskStream};
+<a href=#225 id=225 data-nosnippet>225</a> <span class="kw">use </span><span class="kw">crate</span>::spec::{DataFileFormat, Datum, NestedField, PrimitiveType, Schema, SchemaRef, Type};
+<a href=#226 id=226 data-nosnippet>226</a>
+<a href=#227 id=227 data-nosnippet>227</a> <span class="kw">async fn </span>test_perform_read(
+<a href=#228 id=228 data-nosnippet>228</a> predicate: Predicate,
+<a href=#229 id=229 data-nosnippet>229</a> schema: SchemaRef,
+<a href=#230 id=230 data-nosnippet>230</a> table_location: String,
+<a href=#231 id=231 data-nosnippet>231</a> reader: ArrowReader,
+<a href=#232 id=232 data-nosnippet>232</a> ) -> Vec<<span class="prelude-ty">Option</span><String>> {
+<a href=#233 id=233 data-nosnippet>233</a> <span class="kw">let </span>tasks = Box::pin(futures::stream::iter(
+<a href=#234 id=234 data-nosnippet>234</a> <span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(FileScanTask::builder()
+<a href=#235 id=235 data-nosnippet>235</a> .with_file_size_in_bytes(
+<a href=#236 id=236 data-nosnippet>236</a> std::fs::metadata(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>))
+<a href=#237 id=237 data-nosnippet>237</a> .unwrap()
+<a href=#238 id=238 data-nosnippet>238</a> .len(),
+<a href=#239 id=239 data-nosnippet>239</a> )
+<a href=#240 id=240 data-nosnippet>240</a> .with_start(<span class="number">0</span>)
+<a href=#241 id=241 data-nosnippet>241</a> .with_length(<span class="number">0</span>)
+<a href=#242 id=242 data-nosnippet>242</a> .with_data_file_path(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>))
+<a href=#243 id=243 data-nosnippet>243</a> .with_data_file_format(DataFileFormat::Parquet)
+<a href=#244 id=244 data-nosnippet>244</a> .with_schema(schema.clone())
+<a href=#245 id=245 data-nosnippet>245</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#246 id=246 data-nosnippet>246</a> .with_predicate(<span class="prelude-val">Some</span>(predicate.bind(schema, <span class="bool-val">true</span>).unwrap()))
+<a href=#247 id=247 data-nosnippet>247</a> .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#248 id=248 data-nosnippet>248</a> .build())]
+<a href=#249 id=249 data-nosnippet>249</a> .into_iter(),
+<a href=#250 id=250 data-nosnippet>250</a> )) <span class="kw">as </span>FileScanTaskStream;
+<a href=#251 id=251 data-nosnippet>251</a>
+<a href=#252 id=252 data-nosnippet>252</a> <span class="kw">let </span>result = reader
+<a href=#253 id=253 data-nosnippet>253</a> .read(tasks)
+<a href=#254 id=254 data-nosnippet>254</a> .unwrap()
+<a href=#255 id=255 data-nosnippet>255</a> .stream()
+<a href=#256 id=256 data-nosnippet>256</a> .try_collect::<Vec<RecordBatch>>()
+<a href=#257 id=257 data-nosnippet>257</a> .<span class="kw">await
+<a href=#258 id=258 data-nosnippet>258</a> </span>.unwrap();
+<a href=#259 id=259 data-nosnippet>259</a>
+<a href=#260 id=260 data-nosnippet>260</a> result[<span class="number">0</span>].columns()[<span class="number">0</span>]
+<a href=#261 id=261 data-nosnippet>261</a> .as_string_opt::<i32>()
+<a href=#262 id=262 data-nosnippet>262</a> .unwrap()
+<a href=#263 id=263 data-nosnippet>263</a> .iter()
+<a href=#264 id=264 data-nosnippet>264</a> .map(|v| v.map(ToOwned::to_owned))
+<a href=#265 id=265 data-nosnippet>265</a> .collect::<Vec<<span class="kw">_</span>>>()
+<a href=#266 id=266 data-nosnippet>266</a> }
+<a href=#267 id=267 data-nosnippet>267</a>
+<a href=#268 id=268 data-nosnippet>268</a> <span class="kw">fn </span>setup_kleene_logic(
+<a href=#269 id=269 data-nosnippet>269</a> data_for_col_a: Vec<<span class="prelude-ty">Option</span><String>>,
+<a href=#270 id=270 data-nosnippet>270</a> col_a_type: DataType,
+<a href=#271 id=271 data-nosnippet>271</a> ) -> (FileIO, SchemaRef, String, TempDir) {
+<a href=#272 id=272 data-nosnippet>272</a> <span class="kw">let </span>schema = Arc::new(
+<a href=#273 id=273 data-nosnippet>273</a> Schema::builder()
+<a href=#274 id=274 data-nosnippet>274</a> .with_schema_id(<span class="number">1</span>)
+<a href=#275 id=275 data-nosnippet>275</a> .with_fields(<span class="macro">vec!</span>[
+<a href=#276 id=276 data-nosnippet>276</a> NestedField::optional(<span class="number">1</span>, <span class="string">"a"</span>, Type::Primitive(PrimitiveType::String)).into(),
+<a href=#277 id=277 data-nosnippet>277</a> ])
+<a href=#278 id=278 data-nosnippet>278</a> .build()
+<a href=#279 id=279 data-nosnippet>279</a> .unwrap(),
+<a href=#280 id=280 data-nosnippet>280</a> );
+<a href=#281 id=281 data-nosnippet>281</a>
+<a href=#282 id=282 data-nosnippet>282</a> <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#283 id=283 data-nosnippet>283</a> Field::new(<span class="string">"a"</span>, col_a_type.clone(), <span class="bool-val">true</span>).with_metadata(HashMap::from([(
+<a href=#284 id=284 data-nosnippet>284</a> PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#285 id=285 data-nosnippet>285</a> <span class="string">"1"</span>.to_string(),
+<a href=#286 id=286 data-nosnippet>286</a> )])),
+<a href=#287 id=287 data-nosnippet>287</a> ]));
<a href=#288 id=288 data-nosnippet>288</a>
-<a href=#289 id=289 data-nosnippet>289</a> <span class="comment">// Write the Parquet files
-<a href=#290 id=290 data-nosnippet>290</a> </span><span class="kw">let </span>props = WriterProperties::builder()
-<a href=#291 id=291 data-nosnippet>291</a> .set_compression(Compression::SNAPPY)
-<a href=#292 id=292 data-nosnippet>292</a> .build();
+<a href=#289 id=289 data-nosnippet>289</a> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#290 id=290 data-nosnippet>290</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#291 id=291 data-nosnippet>291</a>
+<a href=#292 id=292 data-nosnippet>292</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
<a href=#293 id=293 data-nosnippet>293</a>
-<a href=#294 id=294 data-nosnippet>294</a> <span class="kw">let </span>file = File::create(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>)).unwrap();
-<a href=#295 id=295 data-nosnippet>295</a> <span class="kw">let </span><span class="kw-2">mut </span>writer =
-<a href=#296 id=296 data-nosnippet>296</a> ArrowWriter::try_new(file, to_write.schema(), <span class="prelude-val">Some</span>(props.clone())).unwrap();
-<a href=#297 id=297 data-nosnippet>297</a>
-<a href=#298 id=298 data-nosnippet>298</a> writer.write(<span class="kw-2">&</span>to_write).expect(<span class="string">"Writing batch"</span>);
+<a href=#294 id=294 data-nosnippet>294</a> <span class="kw">let </span>col = <span class="kw">match </span>col_a_type {
+<a href=#295 id=295 data-nosnippet>295</a> DataType::Utf8 => Arc::new(StringArray::from(data_for_col_a)) <span class="kw">as </span>ArrayRef,
+<a href=#296 id=296 data-nosnippet>296</a> DataType::LargeUtf8 => Arc::new(LargeStringArray::from(data_for_col_a)) <span class="kw">as </span>ArrayRef,
+<a href=#297 id=297 data-nosnippet>297</a> <span class="kw">_ </span>=> <span class="macro">panic!</span>(<span class="string">"unexpected col_a_type"</span>),
+<a href=#298 id=298 data-nosnippet>298</a> };
<a href=#299 id=299 data-nosnippet>299</a>
-<a href=#300 id=300 data-nosnippet>300</a> <span class="comment">// writer must be closed to write footer
-<a href=#301 id=301 data-nosnippet>301</a> </span>writer.close().unwrap();
-<a href=#302 id=302 data-nosnippet>302</a>
-<a href=#303 id=303 data-nosnippet>303</a> (file_io, schema, table_location, tmp_dir)
-<a href=#304 id=304 data-nosnippet>304</a> }
-<a href=#305 id=305 data-nosnippet>305</a>
-<a href=#306 id=306 data-nosnippet>306</a> <span class="attr">#[tokio::test]
-<a href=#307 id=307 data-nosnippet>307</a> </span><span class="kw">async fn </span>test_kleene_logic_or_behaviour() {
-<a href=#308 id=308 data-nosnippet>308</a> <span class="comment">// a IS NULL OR a = 'foo'
-<a href=#309 id=309 data-nosnippet>309</a> </span><span class="kw">let </span>predicate = Reference::new(<span class="string">"a"</span>)
-<a href=#310 id=310 data-nosnippet>310</a> .is_null()
-<a href=#311 id=311 data-nosnippet>311</a> .or(Reference::new(<span class="string">"a"</span>).equal_to(Datum::string(<span class="string">"foo"</span>)));
+<a href=#300 id=300 data-nosnippet>300</a> <span class="kw">let </span>to_write = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[col]).unwrap();
+<a href=#301 id=301 data-nosnippet>301</a>
+<a href=#302 id=302 data-nosnippet>302</a> <span class="comment">// Write the Parquet files
+<a href=#303 id=303 data-nosnippet>303</a> </span><span class="kw">let </span>props = WriterProperties::builder()
+<a href=#304 id=304 data-nosnippet>304</a> .set_compression(Compression::SNAPPY)
+<a href=#305 id=305 data-nosnippet>305</a> .build();
+<a href=#306 id=306 data-nosnippet>306</a>
+<a href=#307 id=307 data-nosnippet>307</a> <span class="kw">let </span>file = File::create(<span class="macro">format!</span>(<span class="string">"{table_location}/1.parquet"</span>)).unwrap();
+<a href=#308 id=308 data-nosnippet>308</a> <span class="kw">let </span><span class="kw-2">mut </span>writer =
+<a href=#309 id=309 data-nosnippet>309</a> ArrowWriter::try_new(file, to_write.schema(), <span class="prelude-val">Some</span>(props.clone())).unwrap();
+<a href=#310 id=310 data-nosnippet>310</a>
+<a href=#311 id=311 data-nosnippet>311</a> writer.write(<span class="kw-2">&</span>to_write).expect(<span class="string">"Writing batch"</span>);
<a href=#312 id=312 data-nosnippet>312</a>
-<a href=#313 id=313 data-nosnippet>313</a> <span class="comment">// Table data: [NULL, "foo", "bar"]
-<a href=#314 id=314 data-nosnippet>314</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#313 id=313 data-nosnippet>313</a> <span class="comment">// writer must be closed to write footer
+<a href=#314 id=314 data-nosnippet>314</a> </span>writer.close().unwrap();
<a href=#315 id=315 data-nosnippet>315</a>
-<a href=#316 id=316 data-nosnippet>316</a> <span class="comment">// Expected: [NULL, "foo"].
-<a href=#317 id=317 data-nosnippet>317</a> </span><span class="kw">let </span>expected = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())];
+<a href=#316 id=316 data-nosnippet>316</a> (file_io, schema, table_location, tmp_dir)
+<a href=#317 id=317 data-nosnippet>317</a> }
<a href=#318 id=318 data-nosnippet>318</a>
-<a href=#319 id=319 data-nosnippet>319</a> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
-<a href=#320 id=320 data-nosnippet>320</a> setup_kleene_logic(data_for_col_a, DataType::Utf8);
-<a href=#321 id=321 data-nosnippet>321</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#322 id=322 data-nosnippet>322</a>
-<a href=#323 id=323 data-nosnippet>323</a> <span class="kw">let </span>result_data = test_perform_read(predicate, schema, table_location, reader).<span class="kw">await</span>;
-<a href=#324 id=324 data-nosnippet>324</a>
-<a href=#325 id=325 data-nosnippet>325</a> <span class="macro">assert_eq!</span>(result_data, expected);
-<a href=#326 id=326 data-nosnippet>326</a> }
-<a href=#327 id=327 data-nosnippet>327</a>
-<a href=#328 id=328 data-nosnippet>328</a> <span class="attr">#[tokio::test]
-<a href=#329 id=329 data-nosnippet>329</a> </span><span class="kw">async fn </span>test_kleene_logic_and_behaviour() {
-<a href=#330 id=330 data-nosnippet>330</a> <span class="comment">// a IS NOT NULL AND a != 'foo'
-<a href=#331 id=331 data-nosnippet>331</a> </span><span class="kw">let </span>predicate = Reference::new(<span class="string">"a"</span>)
-<a href=#332 id=332 data-nosnippet>332</a> .is_not_null()
-<a href=#333 id=333 data-nosnippet>333</a> .and(Reference::new(<span class="string">"a"</span>).not_equal_to(Datum::string(<span class="string">"foo"</span>)));
-<a href=#334 id=334 data-nosnippet>334</a>
-<a href=#335 id=335 data-nosnippet>335</a> <span class="comment">// Table data: [NULL, "foo", "bar"]
-<a href=#336 id=336 data-nosnippet>336</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#319 id=319 data-nosnippet>319</a> <span class="attr">#[tokio::test]
+<a href=#320 id=320 data-nosnippet>320</a> </span><span class="kw">async fn </span>test_kleene_logic_or_behaviour() {
+<a href=#321 id=321 data-nosnippet>321</a> <span class="comment">// a IS NULL OR a = 'foo'
+<a href=#322 id=322 data-nosnippet>322</a> </span><span class="kw">let </span>predicate = Reference::new(<span class="string">"a"</span>)
+<a href=#323 id=323 data-nosnippet>323</a> .is_null()
+<a href=#324 id=324 data-nosnippet>324</a> .or(Reference::new(<span class="string">"a"</span>).equal_to(Datum::string(<span class="string">"foo"</span>)));
+<a href=#325 id=325 data-nosnippet>325</a>
+<a href=#326 id=326 data-nosnippet>326</a> <span class="comment">// Table data: [NULL, "foo", "bar"]
+<a href=#327 id=327 data-nosnippet>327</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#328 id=328 data-nosnippet>328</a>
+<a href=#329 id=329 data-nosnippet>329</a> <span class="comment">// Expected: [NULL, "foo"].
+<a href=#330 id=330 data-nosnippet>330</a> </span><span class="kw">let </span>expected = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())];
+<a href=#331 id=331 data-nosnippet>331</a>
+<a href=#332 id=332 data-nosnippet>332</a> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
+<a href=#333 id=333 data-nosnippet>333</a> setup_kleene_logic(data_for_col_a, DataType::Utf8);
+<a href=#334 id=334 data-nosnippet>334</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#335 id=335 data-nosnippet>335</a>
+<a href=#336 id=336 data-nosnippet>336</a> <span class="kw">let </span>result_data = test_perform_read(predicate, schema, table_location, reader).<span class="kw">await</span>;
<a href=#337 id=337 data-nosnippet>337</a>
-<a href=#338 id=338 data-nosnippet>338</a> <span class="comment">// Expected: ["bar"].
-<a href=#339 id=339 data-nosnippet>339</a> </span><span class="kw">let </span>expected = <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#338 id=338 data-nosnippet>338</a> <span class="macro">assert_eq!</span>(result_data, expected);
+<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> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
-<a href=#342 id=342 data-nosnippet>342</a> setup_kleene_logic(data_for_col_a, DataType::Utf8);
-<a href=#343 id=343 data-nosnippet>343</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#344 id=344 data-nosnippet>344</a>
-<a href=#345 id=345 data-nosnippet>345</a> <span class="kw">let </span>result_data = test_perform_read(predicate, schema, table_location, reader).<span class="kw">await</span>;
-<a href=#346 id=346 data-nosnippet>346</a>
-<a href=#347 id=347 data-nosnippet>347</a> <span class="macro">assert_eq!</span>(result_data, expected);
-<a href=#348 id=348 data-nosnippet>348</a> }
-<a href=#349 id=349 data-nosnippet>349</a>
-<a href=#350 id=350 data-nosnippet>350</a> <span class="attr">#[tokio::test]
-<a href=#351 id=351 data-nosnippet>351</a> </span><span class="kw">async fn </span>test_predicate_cast_literal() {
-<a href=#352 id=352 data-nosnippet>352</a> <span class="kw">let </span>predicates = <span class="macro">vec!</span>[
-<a href=#353 id=353 data-nosnippet>353</a> <span class="comment">// a == 'foo'
-<a href=#354 id=354 data-nosnippet>354</a> </span>(Reference::new(<span class="string">"a"</span>).equal_to(Datum::string(<span class="string">"foo"</span>)), <span class="macro">vec!</span>[
-<a href=#355 id=355 data-nosnippet>355</a> <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()),
-<a href=#356 id=356 data-nosnippet>356</a> ]),
-<a href=#357 id=357 data-nosnippet>357</a> <span class="comment">// a != 'foo'
-<a href=#358 id=358 data-nosnippet>358</a> </span>(
-<a href=#359 id=359 data-nosnippet>359</a> Reference::new(<span class="string">"a"</span>).not_equal_to(Datum::string(<span class="string">"foo"</span>)),
-<a href=#360 id=360 data-nosnippet>360</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
-<a href=#361 id=361 data-nosnippet>361</a> ),
-<a href=#362 id=362 data-nosnippet>362</a> <span class="comment">// STARTS_WITH(a, 'foo')
-<a href=#363 id=363 data-nosnippet>363</a> </span>(Reference::new(<span class="string">"a"</span>).starts_with(Datum::string(<span class="string">"f"</span>)), <span class="macro">vec!</span>[
-<a href=#364 id=364 data-nosnippet>364</a> <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()),
-<a href=#365 id=365 data-nosnippet>365</a> ]),
-<a href=#366 id=366 data-nosnippet>366</a> <span class="comment">// NOT STARTS_WITH(a, 'foo')
-<a href=#367 id=367 data-nosnippet>367</a> </span>(
-<a href=#368 id=368 data-nosnippet>368</a> Reference::new(<span class="string">"a"</span>).not_starts_with(Datum::string(<span class="string">"f"</span>)),
-<a href=#369 id=369 data-nosnippet>369</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
-<a href=#370 id=370 data-nosnippet>370</a> ),
-<a href=#371 id=371 data-nosnippet>371</a> <span class="comment">// a < 'foo'
-<a href=#372 id=372 data-nosnippet>372</a> </span>(Reference::new(<span class="string">"a"</span>).less_than(Datum::string(<span class="string">"foo"</span>)), <span class="macro">vec!</span>[
-<a href=#373 id=373 data-nosnippet>373</a> <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string()),
-<a href=#374 id=374 data-nosnippet>374</a> ]),
-<a href=#375 id=375 data-nosnippet>375</a> <span class="comment">// a <= 'foo'
-<a href=#376 id=376 data-nosnippet>376</a> </span>(
-<a href=#377 id=377 data-nosnippet>377</a> Reference::new(<span class="string">"a"</span>).less_than_or_equal_to(Datum::string(<span class="string">"foo"</span>)),
-<a href=#378 id=378 data-nosnippet>378</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
-<a href=#379 id=379 data-nosnippet>379</a> ),
-<a href=#380 id=380 data-nosnippet>380</a> <span class="comment">// a > 'foo'
-<a href=#381 id=381 data-nosnippet>381</a> </span>(
-<a href=#382 id=382 data-nosnippet>382</a> Reference::new(<span class="string">"a"</span>).greater_than(Datum::string(<span class="string">"bar"</span>)),
-<a href=#383 id=383 data-nosnippet>383</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
-<a href=#384 id=384 data-nosnippet>384</a> ),
-<a href=#385 id=385 data-nosnippet>385</a> <span class="comment">// a >= 'foo'
-<a href=#386 id=386 data-nosnippet>386</a> </span>(
-<a href=#387 id=387 data-nosnippet>387</a> Reference::new(<span class="string">"a"</span>).greater_than_or_equal_to(Datum::string(<span class="string">"foo"</span>)),
-<a href=#388 id=388 data-nosnippet>388</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
-<a href=#389 id=389 data-nosnippet>389</a> ),
-<a href=#390 id=390 data-nosnippet>390</a> <span class="comment">// a IN ('foo', 'bar')
-<a href=#391 id=391 data-nosnippet>391</a> </span>(
-<a href=#392 id=392 data-nosnippet>392</a> Reference::new(<span class="string">"a"</span>).is_in([Datum::string(<span class="string">"foo"</span>), Datum::string(<span class="string">"baz"</span>)]),
-<a href=#393 id=393 data-nosnippet>393</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
-<a href=#394 id=394 data-nosnippet>394</a> ),
-<a href=#395 id=395 data-nosnippet>395</a> <span class="comment">// a NOT IN ('foo', 'bar')
-<a href=#396 id=396 data-nosnippet>396</a> </span>(
-<a href=#397 id=397 data-nosnippet>397</a> Reference::new(<span class="string">"a"</span>).is_not_in([Datum::string(<span class="string">"foo"</span>), Datum::string(<span class="string">"baz"</span>)]),
-<a href=#398 id=398 data-nosnippet>398</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
-<a href=#399 id=399 data-nosnippet>399</a> ),
-<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> <span class="comment">// Table data: ["foo", "bar"]
-<a href=#403 id=403 data-nosnippet>403</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
-<a href=#404 id=404 data-nosnippet>404</a>
-<a href=#405 id=405 data-nosnippet>405</a> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
-<a href=#406 id=406 data-nosnippet>406</a> setup_kleene_logic(data_for_col_a, DataType::LargeUtf8);
-<a href=#407 id=407 data-nosnippet>407</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#408 id=408 data-nosnippet>408</a>
-<a href=#409 id=409 data-nosnippet>409</a> <span class="kw">for </span>(predicate, expected) <span class="kw">in </span>predicates {
-<a href=#410 id=410 data-nosnippet>410</a> <span class="macro">println!</span>(<span class="string">"testing predicate {predicate}"</span>);
-<a href=#411 id=411 data-nosnippet>411</a> <span class="kw">let </span>result_data = test_perform_read(
-<a href=#412 id=412 data-nosnippet>412</a> predicate.clone(),
-<a href=#413 id=413 data-nosnippet>413</a> schema.clone(),
-<a href=#414 id=414 data-nosnippet>414</a> table_location.clone(),
-<a href=#415 id=415 data-nosnippet>415</a> reader.clone(),
-<a href=#416 id=416 data-nosnippet>416</a> )
-<a href=#417 id=417 data-nosnippet>417</a> .<span class="kw">await</span>;
-<a href=#418 id=418 data-nosnippet>418</a>
-<a href=#419 id=419 data-nosnippet>419</a> <span class="macro">assert_eq!</span>(result_data, expected, <span class="string">"predicate={predicate}"</span>);
-<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> <span class="doccomment">/// Verifies that file splits respect byte ranges and only read specific row groups.
-<a href=#424 id=424 data-nosnippet>424</a> </span><span class="attr">#[tokio::test]
-<a href=#425 id=425 data-nosnippet>425</a> </span><span class="kw">async fn </span>test_file_splits_respect_byte_ranges() {
-<a href=#426 id=426 data-nosnippet>426</a> <span class="kw">use </span>arrow_array::Int32Array;
-<a href=#427 id=427 data-nosnippet>427</a> <span class="kw">use </span>parquet::file::reader::{FileReader, SerializedFileReader};
-<a href=#428 id=428 data-nosnippet>428</a>
-<a href=#429 id=429 data-nosnippet>429</a> <span class="kw">let </span>schema = Arc::new(
-<a href=#430 id=430 data-nosnippet>430</a> Schema::builder()
-<a href=#431 id=431 data-nosnippet>431</a> .with_schema_id(<span class="number">1</span>)
-<a href=#432 id=432 data-nosnippet>432</a> .with_fields(<span class="macro">vec!</span>[
-<a href=#433 id=433 data-nosnippet>433</a> NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
-<a href=#434 id=434 data-nosnippet>434</a> ])
-<a href=#435 id=435 data-nosnippet>435</a> .build()
-<a href=#436 id=436 data-nosnippet>436</a> .unwrap(),
-<a href=#437 id=437 data-nosnippet>437</a> );
-<a href=#438 id=438 data-nosnippet>438</a>
-<a href=#439 id=439 data-nosnippet>439</a> <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
-<a href=#440 id=440 data-nosnippet>440</a> Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
-<a href=#441 id=441 data-nosnippet>441</a> PARQUET_FIELD_ID_META_KEY.to_string(),
-<a href=#442 id=442 data-nosnippet>442</a> <span class="string">"1"</span>.to_string(),
-<a href=#443 id=443 data-nosnippet>443</a> )])),
-<a href=#444 id=444 data-nosnippet>444</a> ]));
-<a href=#445 id=445 data-nosnippet>445</a>
-<a href=#446 id=446 data-nosnippet>446</a> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
-<a href=#447 id=447 data-nosnippet>447</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
-<a href=#448 id=448 data-nosnippet>448</a> <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/multi_row_group.parquet"</span>);
-<a href=#449 id=449 data-nosnippet>449</a>
-<a href=#450 id=450 data-nosnippet>450</a> <span class="comment">// Force each batch into its own row group for testing byte range filtering.
-<a href=#451 id=451 data-nosnippet>451</a> </span><span class="kw">let </span>batch1 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
-<a href=#452 id=452 data-nosnippet>452</a> (<span class="number">0</span>..<span class="number">100</span>).collect::<Vec<i32>>(),
-<a href=#453 id=453 data-nosnippet>453</a> ))])
-<a href=#454 id=454 data-nosnippet>454</a> .unwrap();
-<a href=#455 id=455 data-nosnippet>455</a> <span class="kw">let </span>batch2 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
-<a href=#456 id=456 data-nosnippet>456</a> (<span class="number">100</span>..<span class="number">200</span>).collect::<Vec<i32>>(),
-<a href=#457 id=457 data-nosnippet>457</a> ))])
-<a href=#458 id=458 data-nosnippet>458</a> .unwrap();
-<a href=#459 id=459 data-nosnippet>459</a> <span class="kw">let </span>batch3 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
-<a href=#460 id=460 data-nosnippet>460</a> (<span class="number">200</span>..<span class="number">300</span>).collect::<Vec<i32>>(),
-<a href=#461 id=461 data-nosnippet>461</a> ))])
-<a href=#462 id=462 data-nosnippet>462</a> .unwrap();
-<a href=#463 id=463 data-nosnippet>463</a>
-<a href=#464 id=464 data-nosnippet>464</a> <span class="kw">let </span>props = WriterProperties::builder()
-<a href=#465 id=465 data-nosnippet>465</a> .set_compression(Compression::SNAPPY)
-<a href=#466 id=466 data-nosnippet>466</a> .set_max_row_group_row_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
-<a href=#467 id=467 data-nosnippet>467</a> .build();
-<a href=#468 id=468 data-nosnippet>468</a>
-<a href=#469 id=469 data-nosnippet>469</a> <span class="kw">let </span>file = File::create(<span class="kw-2">&</span>file_path).unwrap();
-<a href=#470 id=470 data-nosnippet>470</a> <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema.clone(), <span class="prelude-val">Some</span>(props)).unwrap();
-<a href=#471 id=471 data-nosnippet>471</a> writer.write(<span class="kw-2">&</span>batch1).expect(<span class="string">"Writing batch 1"</span>);
-<a href=#472 id=472 data-nosnippet>472</a> writer.write(<span class="kw-2">&</span>batch2).expect(<span class="string">"Writing batch 2"</span>);
-<a href=#473 id=473 data-nosnippet>473</a> writer.write(<span class="kw-2">&</span>batch3).expect(<span class="string">"Writing batch 3"</span>);
-<a href=#474 id=474 data-nosnippet>474</a> writer.close().unwrap();
-<a href=#475 id=475 data-nosnippet>475</a>
-<a href=#476 id=476 data-nosnippet>476</a> <span class="comment">// Read the file metadata to get row group byte positions
-<a href=#477 id=477 data-nosnippet>477</a> </span><span class="kw">let </span>file = File::open(<span class="kw-2">&</span>file_path).unwrap();
-<a href=#478 id=478 data-nosnippet>478</a> <span class="kw">let </span>reader = SerializedFileReader::new(file).unwrap();
-<a href=#479 id=479 data-nosnippet>479</a> <span class="kw">let </span>metadata = reader.metadata();
-<a href=#480 id=480 data-nosnippet>480</a>
-<a href=#481 id=481 data-nosnippet>481</a> <span class="macro">println!</span>(<span class="string">"File has {} row groups"</span>, metadata.num_row_groups());
-<a href=#482 id=482 data-nosnippet>482</a> <span class="macro">assert_eq!</span>(metadata.num_row_groups(), <span class="number">3</span>, <span class="string">"Expected 3 row groups"</span>);
-<a href=#483 id=483 data-nosnippet>483</a>
-<a href=#484 id=484 data-nosnippet>484</a> <span class="comment">// Get byte positions for each row group
-<a href=#485 id=485 data-nosnippet>485</a> </span><span class="kw">let </span>row_group_0 = metadata.row_group(<span class="number">0</span>);
-<a href=#486 id=486 data-nosnippet>486</a> <span class="kw">let </span>row_group_1 = metadata.row_group(<span class="number">1</span>);
-<a href=#487 id=487 data-nosnippet>487</a> <span class="kw">let </span>row_group_2 = metadata.row_group(<span class="number">2</span>);
+<a href=#341 id=341 data-nosnippet>341</a> <span class="attr">#[tokio::test]
+<a href=#342 id=342 data-nosnippet>342</a> </span><span class="kw">async fn </span>test_kleene_logic_and_behaviour() {
+<a href=#343 id=343 data-nosnippet>343</a> <span class="comment">// a IS NOT NULL AND a != 'foo'
+<a href=#344 id=344 data-nosnippet>344</a> </span><span class="kw">let </span>predicate = Reference::new(<span class="string">"a"</span>)
+<a href=#345 id=345 data-nosnippet>345</a> .is_not_null()
+<a href=#346 id=346 data-nosnippet>346</a> .and(Reference::new(<span class="string">"a"</span>).not_equal_to(Datum::string(<span class="string">"foo"</span>)));
+<a href=#347 id=347 data-nosnippet>347</a>
+<a href=#348 id=348 data-nosnippet>348</a> <span class="comment">// Table data: [NULL, "foo", "bar"]
+<a href=#349 id=349 data-nosnippet>349</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">None</span>, <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#350 id=350 data-nosnippet>350</a>
+<a href=#351 id=351 data-nosnippet>351</a> <span class="comment">// Expected: ["bar"].
+<a href=#352 id=352 data-nosnippet>352</a> </span><span class="kw">let </span>expected = <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#353 id=353 data-nosnippet>353</a>
+<a href=#354 id=354 data-nosnippet>354</a> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
+<a href=#355 id=355 data-nosnippet>355</a> setup_kleene_logic(data_for_col_a, DataType::Utf8);
+<a href=#356 id=356 data-nosnippet>356</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#357 id=357 data-nosnippet>357</a>
+<a href=#358 id=358 data-nosnippet>358</a> <span class="kw">let </span>result_data = test_perform_read(predicate, schema, table_location, reader).<span class="kw">await</span>;
+<a href=#359 id=359 data-nosnippet>359</a>
+<a href=#360 id=360 data-nosnippet>360</a> <span class="macro">assert_eq!</span>(result_data, expected);
+<a href=#361 id=361 data-nosnippet>361</a> }
+<a href=#362 id=362 data-nosnippet>362</a>
+<a href=#363 id=363 data-nosnippet>363</a> <span class="attr">#[tokio::test]
+<a href=#364 id=364 data-nosnippet>364</a> </span><span class="kw">async fn </span>test_predicate_cast_literal() {
+<a href=#365 id=365 data-nosnippet>365</a> <span class="kw">let </span>predicates = <span class="macro">vec!</span>[
+<a href=#366 id=366 data-nosnippet>366</a> <span class="comment">// a == 'foo'
+<a href=#367 id=367 data-nosnippet>367</a> </span>(Reference::new(<span class="string">"a"</span>).equal_to(Datum::string(<span class="string">"foo"</span>)), <span class="macro">vec!</span>[
+<a href=#368 id=368 data-nosnippet>368</a> <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()),
+<a href=#369 id=369 data-nosnippet>369</a> ]),
+<a href=#370 id=370 data-nosnippet>370</a> <span class="comment">// a != 'foo'
+<a href=#371 id=371 data-nosnippet>371</a> </span>(
+<a href=#372 id=372 data-nosnippet>372</a> Reference::new(<span class="string">"a"</span>).not_equal_to(Datum::string(<span class="string">"foo"</span>)),
+<a href=#373 id=373 data-nosnippet>373</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
+<a href=#374 id=374 data-nosnippet>374</a> ),
+<a href=#375 id=375 data-nosnippet>375</a> <span class="comment">// STARTS_WITH(a, 'foo')
+<a href=#376 id=376 data-nosnippet>376</a> </span>(Reference::new(<span class="string">"a"</span>).starts_with(Datum::string(<span class="string">"f"</span>)), <span class="macro">vec!</span>[
+<a href=#377 id=377 data-nosnippet>377</a> <span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()),
+<a href=#378 id=378 data-nosnippet>378</a> ]),
+<a href=#379 id=379 data-nosnippet>379</a> <span class="comment">// NOT STARTS_WITH(a, 'foo')
+<a href=#380 id=380 data-nosnippet>380</a> </span>(
+<a href=#381 id=381 data-nosnippet>381</a> Reference::new(<span class="string">"a"</span>).not_starts_with(Datum::string(<span class="string">"f"</span>)),
+<a href=#382 id=382 data-nosnippet>382</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
+<a href=#383 id=383 data-nosnippet>383</a> ),
+<a href=#384 id=384 data-nosnippet>384</a> <span class="comment">// a < 'foo'
+<a href=#385 id=385 data-nosnippet>385</a> </span>(Reference::new(<span class="string">"a"</span>).less_than(Datum::string(<span class="string">"foo"</span>)), <span class="macro">vec!</span>[
+<a href=#386 id=386 data-nosnippet>386</a> <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string()),
+<a href=#387 id=387 data-nosnippet>387</a> ]),
+<a href=#388 id=388 data-nosnippet>388</a> <span class="comment">// a <= 'foo'
+<a href=#389 id=389 data-nosnippet>389</a> </span>(
+<a href=#390 id=390 data-nosnippet>390</a> Reference::new(<span class="string">"a"</span>).less_than_or_equal_to(Datum::string(<span class="string">"foo"</span>)),
+<a href=#391 id=391 data-nosnippet>391</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
+<a href=#392 id=392 data-nosnippet>392</a> ),
+<a href=#393 id=393 data-nosnippet>393</a> <span class="comment">// a > 'foo'
+<a href=#394 id=394 data-nosnippet>394</a> </span>(
+<a href=#395 id=395 data-nosnippet>395</a> Reference::new(<span class="string">"a"</span>).greater_than(Datum::string(<span class="string">"bar"</span>)),
+<a href=#396 id=396 data-nosnippet>396</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
+<a href=#397 id=397 data-nosnippet>397</a> ),
+<a href=#398 id=398 data-nosnippet>398</a> <span class="comment">// a >= 'foo'
+<a href=#399 id=399 data-nosnippet>399</a> </span>(
+<a href=#400 id=400 data-nosnippet>400</a> Reference::new(<span class="string">"a"</span>).greater_than_or_equal_to(Datum::string(<span class="string">"foo"</span>)),
+<a href=#401 id=401 data-nosnippet>401</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
+<a href=#402 id=402 data-nosnippet>402</a> ),
+<a href=#403 id=403 data-nosnippet>403</a> <span class="comment">// a IN ('foo', 'bar')
+<a href=#404 id=404 data-nosnippet>404</a> </span>(
+<a href=#405 id=405 data-nosnippet>405</a> Reference::new(<span class="string">"a"</span>).is_in([Datum::string(<span class="string">"foo"</span>), Datum::string(<span class="string">"baz"</span>)]),
+<a href=#406 id=406 data-nosnippet>406</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string())],
+<a href=#407 id=407 data-nosnippet>407</a> ),
+<a href=#408 id=408 data-nosnippet>408</a> <span class="comment">// a NOT IN ('foo', 'bar')
+<a href=#409 id=409 data-nosnippet>409</a> </span>(
+<a href=#410 id=410 data-nosnippet>410</a> Reference::new(<span class="string">"a"</span>).is_not_in([Datum::string(<span class="string">"foo"</span>), Datum::string(<span class="string">"baz"</span>)]),
+<a href=#411 id=411 data-nosnippet>411</a> <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())],
+<a href=#412 id=412 data-nosnippet>412</a> ),
+<a href=#413 id=413 data-nosnippet>413</a> ];
+<a href=#414 id=414 data-nosnippet>414</a>
+<a href=#415 id=415 data-nosnippet>415</a> <span class="comment">// Table data: ["foo", "bar"]
+<a href=#416 id=416 data-nosnippet>416</a> </span><span class="kw">let </span>data_for_col_a = <span class="macro">vec!</span>[<span class="prelude-val">Some</span>(<span class="string">"foo"</span>.to_string()), <span class="prelude-val">Some</span>(<span class="string">"bar"</span>.to_string())];
+<a href=#417 id=417 data-nosnippet>417</a>
+<a href=#418 id=418 data-nosnippet>418</a> <span class="kw">let </span>(file_io, schema, table_location, _temp_dir) =
+<a href=#419 id=419 data-nosnippet>419</a> setup_kleene_logic(data_for_col_a, DataType::LargeUtf8);
+<a href=#420 id=420 data-nosnippet>420</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#421 id=421 data-nosnippet>421</a>
+<a href=#422 id=422 data-nosnippet>422</a> <span class="kw">for </span>(predicate, expected) <span class="kw">in </span>predicates {
+<a href=#423 id=423 data-nosnippet>423</a> <span class="macro">println!</span>(<span class="string">"testing predicate {predicate}"</span>);
+<a href=#424 id=424 data-nosnippet>424</a> <span class="kw">let </span>result_data = test_perform_read(
+<a href=#425 id=425 data-nosnippet>425</a> predicate.clone(),
+<a href=#426 id=426 data-nosnippet>426</a> schema.clone(),
+<a href=#427 id=427 data-nosnippet>427</a> table_location.clone(),
+<a href=#428 id=428 data-nosnippet>428</a> reader.clone(),
+<a href=#429 id=429 data-nosnippet>429</a> )
+<a href=#430 id=430 data-nosnippet>430</a> .<span class="kw">await</span>;
+<a href=#431 id=431 data-nosnippet>431</a>
+<a href=#432 id=432 data-nosnippet>432</a> <span class="macro">assert_eq!</span>(result_data, expected, <span class="string">"predicate={predicate}"</span>);
+<a href=#433 id=433 data-nosnippet>433</a> }
+<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="doccomment">/// Verifies that file splits respect byte ranges and only read specific row groups.
+<a href=#437 id=437 data-nosnippet>437</a> </span><span class="attr">#[tokio::test]
+<a href=#438 id=438 data-nosnippet>438</a> </span><span class="kw">async fn </span>test_file_splits_respect_byte_ranges() {
+<a href=#439 id=439 data-nosnippet>439</a> <span class="kw">use </span>arrow_array::Int32Array;
+<a href=#440 id=440 data-nosnippet>440</a> <span class="kw">use </span>parquet::file::reader::{FileReader, SerializedFileReader};
+<a href=#441 id=441 data-nosnippet>441</a>
+<a href=#442 id=442 data-nosnippet>442</a> <span class="kw">let </span>schema = Arc::new(
+<a href=#443 id=443 data-nosnippet>443</a> Schema::builder()
+<a href=#444 id=444 data-nosnippet>444</a> .with_schema_id(<span class="number">1</span>)
+<a href=#445 id=445 data-nosnippet>445</a> .with_fields(<span class="macro">vec!</span>[
+<a href=#446 id=446 data-nosnippet>446</a> NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#447 id=447 data-nosnippet>447</a> ])
+<a href=#448 id=448 data-nosnippet>448</a> .build()
+<a href=#449 id=449 data-nosnippet>449</a> .unwrap(),
+<a href=#450 id=450 data-nosnippet>450</a> );
+<a href=#451 id=451 data-nosnippet>451</a>
+<a href=#452 id=452 data-nosnippet>452</a> <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#453 id=453 data-nosnippet>453</a> Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#454 id=454 data-nosnippet>454</a> PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#455 id=455 data-nosnippet>455</a> <span class="string">"1"</span>.to_string(),
+<a href=#456 id=456 data-nosnippet>456</a> )])),
+<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> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#460 id=460 data-nosnippet>460</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#461 id=461 data-nosnippet>461</a> <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/multi_row_group.parquet"</span>);
+<a href=#462 id=462 data-nosnippet>462</a>
+<a href=#463 id=463 data-nosnippet>463</a> <span class="comment">// Force each batch into its own row group for testing byte range filtering.
+<a href=#464 id=464 data-nosnippet>464</a> </span><span class="kw">let </span>batch1 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
+<a href=#465 id=465 data-nosnippet>465</a> (<span class="number">0</span>..<span class="number">100</span>).collect::<Vec<i32>>(),
+<a href=#466 id=466 data-nosnippet>466</a> ))])
+<a href=#467 id=467 data-nosnippet>467</a> .unwrap();
+<a href=#468 id=468 data-nosnippet>468</a> <span class="kw">let </span>batch2 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
+<a href=#469 id=469 data-nosnippet>469</a> (<span class="number">100</span>..<span class="number">200</span>).collect::<Vec<i32>>(),
+<a href=#470 id=470 data-nosnippet>470</a> ))])
+<a href=#471 id=471 data-nosnippet>471</a> .unwrap();
+<a href=#472 id=472 data-nosnippet>472</a> <span class="kw">let </span>batch3 = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(Int32Array::from(
+<a href=#473 id=473 data-nosnippet>473</a> (<span class="number">200</span>..<span class="number">300</span>).collect::<Vec<i32>>(),
+<a href=#474 id=474 data-nosnippet>474</a> ))])
+<a href=#475 id=475 data-nosnippet>475</a> .unwrap();
+<a href=#476 id=476 data-nosnippet>476</a>
+<a href=#477 id=477 data-nosnippet>477</a> <span class="kw">let </span>props = WriterProperties::builder()
+<a href=#478 id=478 data-nosnippet>478</a> .set_compression(Compression::SNAPPY)
+<a href=#479 id=479 data-nosnippet>479</a> .set_max_row_group_row_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#480 id=480 data-nosnippet>480</a> .build();
+<a href=#481 id=481 data-nosnippet>481</a>
+<a href=#482 id=482 data-nosnippet>482</a> <span class="kw">let </span>file = File::create(<span class="kw-2">&</span>file_path).unwrap();
+<a href=#483 id=483 data-nosnippet>483</a> <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema.clone(), <span class="prelude-val">Some</span>(props)).unwrap();
+<a href=#484 id=484 data-nosnippet>484</a> writer.write(<span class="kw-2">&</span>batch1).expect(<span class="string">"Writing batch 1"</span>);
+<a href=#485 id=485 data-nosnippet>485</a> writer.write(<span class="kw-2">&</span>batch2).expect(<span class="string">"Writing batch 2"</span>);
+<a href=#486 id=486 data-nosnippet>486</a> writer.write(<span class="kw-2">&</span>batch3).expect(<span class="string">"Writing batch 3"</span>);
+<a href=#487 id=487 data-nosnippet>487</a> writer.close().unwrap();
<a href=#488 id=488 data-nosnippet>488</a>
-<a href=#489 id=489 data-nosnippet>489</a> <span class="kw">let </span>rg0_start = <span class="number">4u64</span>; <span class="comment">// Parquet files start with 4-byte magic "PAR1"
-<a href=#490 id=490 data-nosnippet>490</a> </span><span class="kw">let </span>rg1_start = rg0_start + row_group_0.compressed_size() <span class="kw">as </span>u64;
-<a href=#491 id=491 data-nosnippet>491</a> <span class="kw">let </span>rg2_start = rg1_start + row_group_1.compressed_size() <span class="kw">as </span>u64;
-<a href=#492 id=492 data-nosnippet>492</a> <span class="kw">let </span>file_end = rg2_start + row_group_2.compressed_size() <span class="kw">as </span>u64;
+<a href=#489 id=489 data-nosnippet>489</a> <span class="comment">// Read the file metadata to get row group byte positions
+<a href=#490 id=490 data-nosnippet>490</a> </span><span class="kw">let </span>file = File::open(<span class="kw-2">&</span>file_path).unwrap();
+<a href=#491 id=491 data-nosnippet>491</a> <span class="kw">let </span>reader = SerializedFileReader::new(file).unwrap();
+<a href=#492 id=492 data-nosnippet>492</a> <span class="kw">let </span>metadata = reader.metadata();
<a href=#493 id=493 data-nosnippet>493</a>
-<a href=#494 id=494 data-nosnippet>494</a> <span class="macro">println!</span>(
-<a href=#495 id=495 data-nosnippet>495</a> <span class="string">"Row group 0: {} rows, starts at byte {}, {} bytes compressed"</span>,
-<a href=#496 id=496 data-nosnippet>496</a> row_group_0.num_rows(),
-<a href=#497 id=497 data-nosnippet>497</a> rg0_start,
-<a href=#498 id=498 data-nosnippet>498</a> row_group_0.compressed_size()
-<a href=#499 id=499 data-nosnippet>499</a> );
-<a href=#500 id=500 data-nosnippet>500</a> <span class="macro">println!</span>(
-<a href=#501 id=501 data-nosnippet>501</a> <span class="string">"Row group 1: {} rows, starts at byte {}, {} bytes compressed"</span>,
-<a href=#502 id=502 data-nosnippet>502</a> row_group_1.num_rows(),
-<a href=#503 id=503 data-nosnippet>503</a> rg1_start,
-<a href=#504 id=504 data-nosnippet>504</a> row_group_1.compressed_size()
-<a href=#505 id=505 data-nosnippet>505</a> );
-<a href=#506 id=506 data-nosnippet>506</a> <span class="macro">println!</span>(
-<a href=#507 id=507 data-nosnippet>507</a> <span class="string">"Row group 2: {} rows, starts at byte {}, {} bytes compressed"</span>,
-<a href=#508 id=508 data-nosnippet>508</a> row_group_2.num_rows(),
-<a href=#509 id=509 data-nosnippet>509</a> rg2_start,
-<a href=#510 id=510 data-nosnippet>510</a> row_group_2.compressed_size()
-<a href=#511 id=511 data-nosnippet>511</a> );
-<a href=#512 id=512 data-nosnippet>512</a>
-<a href=#513 id=513 data-nosnippet>513</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
-<a href=#514 id=514 data-nosnippet>514</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
-<a href=#515 id=515 data-nosnippet>515</a>
-<a href=#516 id=516 data-nosnippet>516</a> <span class="comment">// Task 1: read only the first row group
-<a href=#517 id=517 data-nosnippet>517</a> </span><span class="kw">let </span>task1 = FileScanTask::builder()
-<a href=#518 id=518 data-nosnippet>518</a> .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&</span>file_path).unwrap().len())
-<a href=#519 id=519 data-nosnippet>519</a> .with_start(rg0_start)
-<a href=#520 id=520 data-nosnippet>520</a> .with_length(row_group_0.compressed_size() <span class="kw">as </span>u64)
-<a href=#521 id=521 data-nosnippet>521</a> .with_data_file_path(file_path.clone())
-<a href=#522 id=522 data-nosnippet>522</a> .with_data_file_format(DataFileFormat::Parquet)
-<a href=#523 id=523 data-nosnippet>523</a> .with_schema(schema.clone())
-<a href=#524 id=524 data-nosnippet>524</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#525 id=525 data-nosnippet>525</a> .with_record_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
-<a href=#526 id=526 data-nosnippet>526</a> .with_case_sensitive(<span class="bool-val">false</span>)
-<a href=#527 id=527 data-nosnippet>527</a> .build();
+<a href=#494 id=494 data-nosnippet>494</a> <span class="macro">println!</span>(<span class="string">"File has {} row groups"</span>, metadata.num_row_groups());
+<a href=#495 id=495 data-nosnippet>495</a> <span class="macro">assert_eq!</span>(metadata.num_row_groups(), <span class="number">3</span>, <span class="string">"Expected 3 row groups"</span>);
+<a href=#496 id=496 data-nosnippet>496</a>
+<a href=#497 id=497 data-nosnippet>497</a> <span class="comment">// Get byte positions for each row group
+<a href=#498 id=498 data-nosnippet>498</a> </span><span class="kw">let </span>row_group_0 = metadata.row_group(<span class="number">0</span>);
+<a href=#499 id=499 data-nosnippet>499</a> <span class="kw">let </span>row_group_1 = metadata.row_group(<span class="number">1</span>);
+<a href=#500 id=500 data-nosnippet>500</a> <span class="kw">let </span>row_group_2 = metadata.row_group(<span class="number">2</span>);
+<a href=#501 id=501 data-nosnippet>501</a>
+<a href=#502 id=502 data-nosnippet>502</a> <span class="kw">let </span>rg0_start = <span class="number">4u64</span>; <span class="comment">// Parquet files start with 4-byte magic "PAR1"
+<a href=#503 id=503 data-nosnippet>503</a> </span><span class="kw">let </span>rg1_start = rg0_start + row_group_0.compressed_size() <span class="kw">as </span>u64;
+<a href=#504 id=504 data-nosnippet>504</a> <span class="kw">let </span>rg2_start = rg1_start + row_group_1.compressed_size() <span class="kw">as </span>u64;
+<a href=#505 id=505 data-nosnippet>505</a> <span class="kw">let </span>file_end = rg2_start + row_group_2.compressed_size() <span class="kw">as </span>u64;
+<a href=#506 id=506 data-nosnippet>506</a>
+<a href=#507 id=507 data-nosnippet>507</a> <span class="macro">println!</span>(
+<a href=#508 id=508 data-nosnippet>508</a> <span class="string">"Row group 0: {} rows, starts at byte {}, {} bytes compressed"</span>,
+<a href=#509 id=509 data-nosnippet>509</a> row_group_0.num_rows(),
+<a href=#510 id=510 data-nosnippet>510</a> rg0_start,
+<a href=#511 id=511 data-nosnippet>511</a> row_group_0.compressed_size()
+<a href=#512 id=512 data-nosnippet>512</a> );
+<a href=#513 id=513 data-nosnippet>513</a> <span class="macro">println!</span>(
+<a href=#514 id=514 data-nosnippet>514</a> <span class="string">"Row group 1: {} rows, starts at byte {}, {} bytes compressed"</span>,
+<a href=#515 id=515 data-nosnippet>515</a> row_group_1.num_rows(),
+<a href=#516 id=516 data-nosnippet>516</a> rg1_start,
+<a href=#517 id=517 data-nosnippet>517</a> row_group_1.compressed_size()
+<a href=#518 id=518 data-nosnippet>518</a> );
+<a href=#519 id=519 data-nosnippet>519</a> <span class="macro">println!</span>(
+<a href=#520 id=520 data-nosnippet>520</a> <span class="string">"Row group 2: {} rows, starts at byte {}, {} bytes compressed"</span>,
+<a href=#521 id=521 data-nosnippet>521</a> row_group_2.num_rows(),
+<a href=#522 id=522 data-nosnippet>522</a> rg2_start,
+<a href=#523 id=523 data-nosnippet>523</a> row_group_2.compressed_size()
+<a href=#524 id=524 data-nosnippet>524</a> );
+<a href=#525 id=525 data-nosnippet>525</a>
+<a href=#526 id=526 data-nosnippet>526</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#527 id=527 data-nosnippet>527</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
<a href=#528 id=528 data-nosnippet>528</a>
-<a href=#529 id=529 data-nosnippet>529</a> <span class="comment">// Task 2: read the second and third row groups
-<a href=#530 id=530 data-nosnippet>530</a> </span><span class="kw">let </span>task2 = FileScanTask::builder()
+<a href=#529 id=529 data-nosnippet>529</a> <span class="comment">// Task 1: read only the first row group
+<a href=#530 id=530 data-nosnippet>530</a> </span><span class="kw">let </span>task1 = FileScanTask::builder()
<a href=#531 id=531 data-nosnippet>531</a> .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&</span>file_path).unwrap().len())
-<a href=#532 id=532 data-nosnippet>532</a> .with_start(rg1_start)
-<a href=#533 id=533 data-nosnippet>533</a> .with_length(file_end - rg1_start)
+<a href=#532 id=532 data-nosnippet>532</a> .with_start(rg0_start)
+<a href=#533 id=533 data-nosnippet>533</a> .with_length(row_group_0.compressed_size() <span class="kw">as </span>u64)
<a href=#534 id=534 data-nosnippet>534</a> .with_data_file_path(file_path.clone())
<a href=#535 id=535 data-nosnippet>535</a> .with_data_file_format(DataFileFormat::Parquet)
<a href=#536 id=536 data-nosnippet>536</a> .with_schema(schema.clone())
<a href=#537 id=537 data-nosnippet>537</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
-<a href=#538 id=538 data-nosnippet>538</a> .with_record_count(<span class="prelude-val">Some</span>(<span class="number">200</span>))
+<a href=#538 id=538 data-nosnippet>538</a> .with_record_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
<a href=#539 id=539 data-nosnippet>539</a> .with_case_sensitive(<span class="bool-val">false</span>)
<a href=#540 id=540 data-nosnippet>540</a> .build();
<a href=#541 id=541 data-nosnippet>541</a>
-<a href=#542 id=542 data-nosnippet>542</a> <span class="kw">let </span>tasks1 = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task1)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#543 id=543 data-nosnippet>543</a> <span class="kw">let </span>result1 = reader
-<a href=#544 id=544 data-nosnippet>544</a> .clone()
-<a href=#545 id=545 data-nosnippet>545</a> .read(tasks1)
-<a href=#546 id=546 data-nosnippet>546</a> .unwrap()
-<a href=#547 id=547 data-nosnippet>547</a> .stream()
-<a href=#548 id=548 data-nosnippet>548</a> .try_collect::<Vec<RecordBatch>>()
-<a href=#549 id=549 data-nosnippet>549</a> .<span class="kw">await
-<a href=#550 id=550 data-nosnippet>550</a> </span>.unwrap();
-<a href=#551 id=551 data-nosnippet>551</a>
-<a href=#552 id=552 data-nosnippet>552</a> <span class="kw">let </span>total_rows_task1: usize = result1.iter().map(|b| b.num_rows()).sum();
-<a href=#553 id=553 data-nosnippet>553</a> <span class="macro">println!</span>(
-<a href=#554 id=554 data-nosnippet>554</a> <span class="string">"Task 1 (bytes {}-{}) returned {} rows"</span>,
-<a href=#555 id=555 data-nosnippet>555</a> rg0_start,
-<a href=#556 id=556 data-nosnippet>556</a> rg0_start + row_group_0.compressed_size() <span class="kw">as </span>u64,
-<a href=#557 id=557 data-nosnippet>557</a> total_rows_task1
-<a href=#558 id=558 data-nosnippet>558</a> );
-<a href=#559 id=559 data-nosnippet>559</a>
-<a href=#560 id=560 data-nosnippet>560</a> <span class="kw">let </span>tasks2 = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task2)])) <span class="kw">as </span>FileScanTaskStream;
-<a href=#561 id=561 data-nosnippet>561</a> <span class="kw">let </span>result2 = reader
-<a href=#562 id=562 data-nosnippet>562</a> .read(tasks2)
-<a href=#563 id=563 data-nosnippet>563</a> .unwrap()
-<a href=#564 id=564 data-nosnippet>564</a> .stream()
-<a href=#565 id=565 data-nosnippet>565</a> .try_collect::<Vec<RecordBatch>>()
-<a href=#566 id=566 data-nosnippet>566</a> .<span class="kw">await
-<a href=#567 id=567 data-nosnippet>567</a> </span>.unwrap();
-<a href=#568 id=568 data-nosnippet>568</a>
-<a href=#569 id=569 data-nosnippet>569</a> <span class="kw">let </span>total_rows_task2: usize = result2.iter().map(|b| b.num_rows()).sum();
-<a href=#570 id=570 data-nosnippet>570</a> <span class="macro">println!</span>(<span class="string">"Task 2 (bytes {rg1_start}-{file_end}) returned {total_rows_task2} rows"</span>);
-<a href=#571 id=571 data-nosnippet>571</a>
-<a href=#572 id=572 data-nosnippet>572</a> <span class="macro">assert_eq!</span>(
-<a href=#573 id=573 data-nosnippet>573</a> total_rows_task1, <span class="number">100</span>,
-<a href=#574 id=574 data-nosnippet>574</a> <span class="string">"Task 1 should read only the first row group (100 rows), but got {total_rows_task1} rows"
-<a href=#575 id=575 data-nosnippet>575</a> </span>);
-<a href=#576 id=576 data-nosnippet>576</a>
-<a href=#577 id=577 data-nosnippet>577</a> <span class="macro">assert_eq!</span>(
-<a href=#578 id=578 data-nosnippet>578</a> total_rows_task2, <span class="number">200</span>,
-<a href=#579 id=579 data-nosnippet>579</a> <span class="string">"Task 2 should read only the second+third row groups (200 rows), but got {total_rows_task2} rows"
-<a href=#580 id=580 data-nosnippet>580</a> </span>);
+<a href=#542 id=542 data-nosnippet>542</a> <span class="comment">// Task 2: read the second and third row groups
+<a href=#543 id=543 data-nosnippet>543</a> </span><span class="kw">let </span>task2 = FileScanTask::builder()
+<a href=#544 id=544 data-nosnippet>544</a> .with_file_size_in_bytes(std::fs::metadata(<span class="kw-2">&</span>file_path).unwrap().len())
+<a href=#545 id=545 data-nosnippet>545</a> .with_start(rg1_start)
+<a href=#546 id=546 data-nosnippet>546</a> .with_length(file_end - rg1_start)
+<a href=#547 id=547 data-nosnippet>547</a> .with_data_file_path(file_path.clone())
+<a href=#548 id=548 data-nosnippet>548</a> .with_data_file_format(DataFileFormat::Parquet)
+<a href=#549 id=549 data-nosnippet>549</a> .with_schema(schema.clone())
+<a href=#550 id=550 data-nosnippet>550</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#551 id=551 data-nosnippet>551</a> .with_record_count(<span class="prelude-val">Some</span>(<span class="number">200</span>))
+<a href=#552 id=552 data-nosnippet>552</a> .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#553 id=553 data-nosnippet>553</a> .build();
+<a href=#554 id=554 data-nosnippet>554</a>
+<a href=#555 id=555 data-nosnippet>555</a> <span class="kw">let </span>tasks1 = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task1)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#556 id=556 data-nosnippet>556</a> <span class="kw">let </span>result1 = reader
+<a href=#557 id=557 data-nosnippet>557</a> .clone()
+<a href=#558 id=558 data-nosnippet>558</a> .read(tasks1)
+<a href=#559 id=559 data-nosnippet>559</a> .unwrap()
+<a href=#560 id=560 data-nosnippet>560</a> .stream()
+<a href=#561 id=561 data-nosnippet>561</a> .try_collect::<Vec<RecordBatch>>()
+<a href=#562 id=562 data-nosnippet>562</a> .<span class="kw">await
+<a href=#563 id=563 data-nosnippet>563</a> </span>.unwrap();
+<a href=#564 id=564 data-nosnippet>564</a>
+<a href=#565 id=565 data-nosnippet>565</a> <span class="kw">let </span>total_rows_task1: usize = result1.iter().map(|b| b.num_rows()).sum();
+<a href=#566 id=566 data-nosnippet>566</a> <span class="macro">println!</span>(
+<a href=#567 id=567 data-nosnippet>567</a> <span class="string">"Task 1 (bytes {}-{}) returned {} rows"</span>,
+<a href=#568 id=568 data-nosnippet>568</a> rg0_start,
+<a href=#569 id=569 data-nosnippet>569</a> rg0_start + row_group_0.compressed_size() <span class="kw">as </span>u64,
+<a href=#570 id=570 data-nosnippet>570</a> total_rows_task1
+<a href=#571 id=571 data-nosnippet>571</a> );
+<a href=#572 id=572 data-nosnippet>572</a>
+<a href=#573 id=573 data-nosnippet>573</a> <span class="kw">let </span>tasks2 = Box::pin(futures::stream::iter(<span class="macro">vec!</span>[<span class="prelude-val">Ok</span>(task2)])) <span class="kw">as </span>FileScanTaskStream;
+<a href=#574 id=574 data-nosnippet>574</a> <span class="kw">let </span>result2 = reader
+<a href=#575 id=575 data-nosnippet>575</a> .read(tasks2)
+<a href=#576 id=576 data-nosnippet>576</a> .unwrap()
+<a href=#577 id=577 data-nosnippet>577</a> .stream()
+<a href=#578 id=578 data-nosnippet>578</a> .try_collect::<Vec<RecordBatch>>()
+<a href=#579 id=579 data-nosnippet>579</a> .<span class="kw">await
+<a href=#580 id=580 data-nosnippet>580</a> </span>.unwrap();
<a href=#581 id=581 data-nosnippet>581</a>
-<a href=#582 id=582 data-nosnippet>582</a> <span class="comment">// Verify the actual data values are correct (not just the row count)
-<a href=#583 id=583 data-nosnippet>583</a> </span><span class="kw">if </span>total_rows_task1 > <span class="number">0 </span>{
-<a href=#584 id=584 data-nosnippet>584</a> <span class="kw">let </span>first_batch = <span class="kw-2">&</span>result1[<span class="number">0</span>];
-<a href=#585 id=585 data-nosnippet>585</a> <span class="kw">let </span>id_col = first_batch
-<a href=#586 id=586 data-nosnippet>586</a> .column(<span class="number">0</span>)
-<a href=#587 id=587 data-nosnippet>587</a> .as_primitive::<arrow_array::types::Int32Type>();
-<a href=#588 id=588 data-nosnippet>588</a> <span class="kw">let </span>first_val = id_col.value(<span class="number">0</span>);
-<a href=#589 id=589 data-nosnippet>589</a> <span class="kw">let </span>last_val = id_col.value(id_col.len() - <span class="number">1</span>);
-<a href=#590 id=590 data-nosnippet>590</a> <span class="macro">println!</span>(<span class="string">"Task 1 data range: {first_val} to {last_val}"</span>);
-<a href=#591 id=591 data-nosnippet>591</a>
-<a href=#592 id=592 data-nosnippet>592</a> <span class="macro">assert_eq!</span>(first_val, <span class="number">0</span>, <span class="string">"Task 1 should start with id=0"</span>);
-<a href=#593 id=593 data-nosnippet>593</a> <span class="macro">assert_eq!</span>(last_val, <span class="number">99</span>, <span class="string">"Task 1 should end with id=99"</span>);
-<a href=#594 id=594 data-nosnippet>594</a> }
-<a href=#595 id=595 data-nosnippet>595</a>
-<a href=#596 id=596 data-nosnippet>596</a> <span class="kw">if </span>total_rows_task2 > <span class="number">0 </span>{
-<a href=#597 id=597 data-nosnippet>597</a> <span class="kw">let </span>first_batch = <span class="kw-2">&</span>result2[<span class="number">0</span>];
+<a href=#582 id=582 data-nosnippet>582</a> <span class="kw">let </span>total_rows_task2: usize = result2.iter().map(|b| b.num_rows()).sum();
+<a href=#583 id=583 data-nosnippet>583</a> <span class="macro">println!</span>(<span class="string">"Task 2 (bytes {rg1_start}-{file_end}) returned {total_rows_task2} rows"</span>);
+<a href=#584 id=584 data-nosnippet>584</a>
+<a href=#585 id=585 data-nosnippet>585</a> <span class="macro">assert_eq!</span>(
+<a href=#586 id=586 data-nosnippet>586</a> total_rows_task1, <span class="number">100</span>,
+<a href=#587 id=587 data-nosnippet>587</a> <span class="string">"Task 1 should read only the first row group (100 rows), but got {total_rows_task1} rows"
+<a href=#588 id=588 data-nosnippet>588</a> </span>);
+<a href=#589 id=589 data-nosnippet>589</a>
+<a href=#590 id=590 data-nosnippet>590</a> <span class="macro">assert_eq!</span>(
+<a href=#591 id=591 data-nosnippet>591</a> total_rows_task2, <span class="number">200</span>,
+<a href=#592 id=592 data-nosnippet>592</a> <span class="string">"Task 2 should read only the second+third row groups (200 rows), but got {total_rows_task2} rows"
+<a href=#593 id=593 data-nosnippet>593</a> </span>);
+<a href=#594 id=594 data-nosnippet>594</a>
+<a href=#595 id=595 data-nosnippet>595</a> <span class="comment">// Verify the actual data values are correct (not just the row count)
+<a href=#596 id=596 data-nosnippet>596</a> </span><span class="kw">if </span>total_rows_task1 > <span class="number">0 </span>{
+<a href=#597 id=597 data-nosnippet>597</a> <span class="kw">let </span>first_batch = <span class="kw-2">&</span>result1[<span class="number">0</span>];
<a href=#598 id=598 data-nosnippet>598</a> <span class="kw">let </span>id_col = first_batch
<a href=#599 id=599 data-nosnippet>599</a> .column(<span class="number">0</span>)
<a href=#600 id=600 data-nosnippet>600</a> .as_primitive::<arrow_array::types::Int32Type>();
<a href=#601 id=601 data-nosnippet>601</a> <span class="kw">let </span>first_val = id_col.value(<span class="number">0</span>);
-<a href=#602 id=602 data-nosnippet>602</a> <span class="macro">println!</span>(<span class="string">"Task 2 first value: {first_val}"</span>);
-<a href=#603 id=603 data-nosnippet>603</a>
-<a href=#604 id=604 data-nosnippet>604</a> <span class="macro">assert_eq!</span>(first_val, <span class="number">100</span>, <span class="string">"Task 2 should start with id=100, not id=0"</span>);
-<a href=#605 id=605 data-nosnippet>605</a> }
-<a href=#606 id=606 data-nosnippet>606</a> }
-<a href=#607 id=607 data-nosnippet>607</a>}
+<a href=#602 id=602 data-nosnippet>602</a> <span class="kw">let </span>last_val = id_col.value(id_col.len() - <span class="number">1</span>);
+<a href=#603 id=603 data-nosnippet>603</a> <span class="macro">println!</span>(<span class="string">"Task 1 data range: {first_val} to {last_val}"</span>);
+<a href=#604 id=604 data-nosnippet>604</a>
+<a href=#605 id=605 data-nosnippet>605</a> <span class="macro">assert_eq!</span>(first_val, <span class="number">0</span>, <span class="string">"Task 1 should start with id=0"</span>);
+<a href=#606 id=606 data-nosnippet>606</a> <span class="macro">assert_eq!</span>(last_val, <span class="number">99</span>, <span class="string">"Task 1 should end with id=99"</span>);
+<a href=#607 id=607 data-nosnippet>607</a> }
+<a href=#608 id=608 data-nosnippet>608</a>
+<a href=#609 id=609 data-nosnippet>609</a> <span class="kw">if </span>total_rows_task2 > <span class="number">0 </span>{
+<a href=#610 id=610 data-nosnippet>610</a> <span class="kw">let </span>first_batch = <span class="kw-2">&</span>result2[<span class="number">0</span>];
+<a href=#611 id=611 data-nosnippet>611</a> <span class="kw">let </span>id_col = first_batch
+<a href=#612 id=612 data-nosnippet>612</a> .column(<span class="number">0</span>)
+<a href=#613 id=613 data-nosnippet>613</a> .as_primitive::<arrow_array::types::Int32Type>();
+<a href=#614 id=614 data-nosnippet>614</a> <span class="kw">let </span>first_val = id_col.value(<span class="number">0</span>);
+<a href=#615 id=615 data-nosnippet>615</a> <span class="macro">println!</span>(<span class="string">"Task 2 first value: {first_val}"</span>);
+<a href=#616 id=616 data-nosnippet>616</a>
+<a href=#617 id=617 data-nosnippet>617</a> <span class="macro">assert_eq!</span>(first_val, <span class="number">100</span>, <span class="string">"Task 2 should start with id=100, not id=0"</span>);
+<a href=#618 id=618 data-nosnippet>618</a> }
+<a href=#619 id=619 data-nosnippet>619</a> }
+<a href=#620 id=620 data-nosnippet>620</a>
+<a href=#621 id=621 data-nosnippet>621</a> <span class="doccomment">/// A single data file split into multiple sub-row-group byte ranges (as Spark/Iceberg
+<a href=#622 id=622 data-nosnippet>622</a> /// planning produces when split-size is smaller than a row group) must still yield each
+<a href=#623 id=623 data-nosnippet>623</a> /// row exactly once. The previous overlap-based selection let every split whose byte range
+<a href=#624 id=624 data-nosnippet>624</a> /// touched a row group read it, duplicating rows; ownership by midpoint reads each row group
+<a href=#625 id=625 data-nosnippet>625</a> /// from exactly one split.
+<a href=#626 id=626 data-nosnippet>626</a> </span><span class="attr">#[tokio::test]
+<a href=#627 id=627 data-nosnippet>627</a> </span><span class="kw">async fn </span>test_sub_row_group_splits_do_not_duplicate_rows() {
+<a href=#628 id=628 data-nosnippet>628</a> <span class="kw">use </span>arrow_array::Int32Array;
+<a href=#629 id=629 data-nosnippet>629</a>
+<a href=#630 id=630 data-nosnippet>630</a> <span class="kw">let </span>schema = Arc::new(
+<a href=#631 id=631 data-nosnippet>631</a> Schema::builder()
+<a href=#632 id=632 data-nosnippet>632</a> .with_schema_id(<span class="number">1</span>)
+<a href=#633 id=633 data-nosnippet>633</a> .with_fields(<span class="macro">vec!</span>[
+<a href=#634 id=634 data-nosnippet>634</a> NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#635 id=635 data-nosnippet>635</a> ])
+<a href=#636 id=636 data-nosnippet>636</a> .build()
+<a href=#637 id=637 data-nosnippet>637</a> .unwrap(),
+<a href=#638 id=638 data-nosnippet>638</a> );
+<a href=#639 id=639 data-nosnippet>639</a>
+<a href=#640 id=640 data-nosnippet>640</a> <span class="kw">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#641 id=641 data-nosnippet>641</a> Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#642 id=642 data-nosnippet>642</a> PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#643 id=643 data-nosnippet>643</a> <span class="string">"1"</span>.to_string(),
+<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>
+<a href=#647 id=647 data-nosnippet>647</a> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#648 id=648 data-nosnippet>648</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#649 id=649 data-nosnippet>649</a> <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/sub_split.parquet"</span>);
+<a href=#650 id=650 data-nosnippet>650</a>
+<a href=#651 id=651 data-nosnippet>651</a> <span class="comment">// Three row groups of 100 rows each (ids 0..300).
+<a href=#652 id=652 data-nosnippet>652</a> </span><span class="kw">let </span>props = WriterProperties::builder()
+<a href=#653 id=653 data-nosnippet>653</a> .set_compression(Compression::SNAPPY)
+<a href=#654 id=654 data-nosnippet>654</a> .set_max_row_group_row_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#655 id=655 data-nosnippet>655</a> .build();
+<a href=#656 id=656 data-nosnippet>656</a> <span class="kw">let </span>file = File::create(<span class="kw-2">&</span>file_path).unwrap();
+<a href=#657 id=657 data-nosnippet>657</a> <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema.clone(), <span class="prelude-val">Some</span>(props)).unwrap();
+<a href=#658 id=658 data-nosnippet>658</a> <span class="kw">for </span>chunk <span class="kw">in </span>[<span class="number">0</span>..<span class="number">100</span>, <span class="number">100</span>..<span class="number">200</span>, <span class="number">200</span>..<span class="number">300</span>] {
+<a href=#659 id=659 data-nosnippet>659</a> <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(
+<a href=#660 id=660 data-nosnippet>660</a> Int32Array::from(chunk.collect::<Vec<i32>>()),
+<a href=#661 id=661 data-nosnippet>661</a> )])
+<a href=#662 id=662 data-nosnippet>662</a> .unwrap();
+<a href=#663 id=663 data-nosnippet>663</a> writer.write(<span class="kw-2">&</span>batch).expect(<span class="string">"Writing batch"</span>);
+<a href=#664 id=664 data-nosnippet>664</a> }
+<a href=#665 id=665 data-nosnippet>665</a> writer.close().unwrap();
+<a href=#666 id=666 data-nosnippet>666</a>
+<a href=#667 id=667 data-nosnippet>667</a> <span class="kw">let </span>file_size = std::fs::metadata(<span class="kw-2">&</span>file_path).unwrap().len();
+<a href=#668 id=668 data-nosnippet>668</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#669 id=669 data-nosnippet>669</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#670 id=670 data-nosnippet>670</a>
+<a href=#671 id=671 data-nosnippet>671</a> <span class="comment">// Tile the whole file into 64-byte splits, mirroring Spark's split-size planning, and
+<a href=#672 id=672 data-nosnippet>672</a> // read every split. A 64-byte split is far smaller than a row group, so each row group
+<a href=#673 id=673 data-nosnippet>673</a> // is touched by several splits but must be owned (read) by exactly one.
+<a href=#674 id=674 data-nosnippet>674</a> </span><span class="kw">let </span><span class="kw-2">mut </span>ids = Vec::new();
+<a href=#675 id=675 data-nosnippet>675</a> <span class="kw">let </span>split_size = <span class="number">64u64</span>;
+<a href=#676 id=676 data-nosnippet>676</a> <span class="kw">let </span><span class="kw-2">mut </span>start = <span class="number">0u64</span>;
+<a href=#677 id=677 data-nosnippet>677</a> <span class="kw">while </span>start < file_size {
+<a href=#678 id=678 data-nosnippet>678</a> <span class="kw">let </span>length = split_size.min(file_size - start);
+<a href=#679 id=679 data-nosnippet>679</a> <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#680 id=680 data-nosnippet>680</a> .with_file_size_in_bytes(file_size)
+<a href=#681 id=681 data-nosnippet>681</a> .with_start(start)
+<a href=#682 id=682 data-nosnippet>682</a> .with_length(length)
+<a href=#683 id=683 data-nosnippet>683</a> .with_data_file_path(file_path.clone())
+<a href=#684 id=684 data-nosnippet>684</a> .with_data_file_format(DataFileFormat::Parquet)
+<a href=#685 id=685 data-nosnippet>685</a> .with_schema(schema.clone())
+<a href=#686 id=686 data-nosnippet>686</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#687 id=687 data-nosnippet>687</a> .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#688 id=688 data-nosnippet>688</a> .build();
+<a href=#689 id=689 data-nosnippet>689</a>
+<a href=#690 id=690 data-nosnippet>690</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=#691 id=691 data-nosnippet>691</a> <span class="kw">let </span>batches = reader
+<a href=#692 id=692 data-nosnippet>692</a> .clone()
+<a href=#693 id=693 data-nosnippet>693</a> .read(tasks)
+<a href=#694 id=694 data-nosnippet>694</a> .unwrap()
+<a href=#695 id=695 data-nosnippet>695</a> .stream()
+<a href=#696 id=696 data-nosnippet>696</a> .try_collect::<Vec<RecordBatch>>()
+<a href=#697 id=697 data-nosnippet>697</a> .<span class="kw">await
+<a href=#698 id=698 data-nosnippet>698</a> </span>.unwrap();
+<a href=#699 id=699 data-nosnippet>699</a>
+<a href=#700 id=700 data-nosnippet>700</a> <span class="kw">for </span>batch <span class="kw">in </span><span class="kw-2">&</span>batches {
+<a href=#701 id=701 data-nosnippet>701</a> <span class="kw">let </span>col = batch
+<a href=#702 id=702 data-nosnippet>702</a> .column(<span class="number">0</span>)
+<a href=#703 id=703 data-nosnippet>703</a> .as_primitive::<arrow_array::types::Int32Type>();
+<a href=#704 id=704 data-nosnippet>704</a> ids.extend(col.values().iter().copied());
+<a href=#705 id=705 data-nosnippet>705</a> }
+<a href=#706 id=706 data-nosnippet>706</a>
+<a href=#707 id=707 data-nosnippet>707</a> start += length;
+<a href=#708 id=708 data-nosnippet>708</a> }
+<a href=#709 id=709 data-nosnippet>709</a>
+<a href=#710 id=710 data-nosnippet>710</a> ids.sort_unstable();
+<a href=#711 id=711 data-nosnippet>711</a> <span class="macro">assert_eq!</span>(
+<a href=#712 id=712 data-nosnippet>712</a> ids,
+<a href=#713 id=713 data-nosnippet>713</a> (<span class="number">0</span>..<span class="number">300</span>).collect::<Vec<i32>>(),
+<a href=#714 id=714 data-nosnippet>714</a> <span class="string">"each row must be read exactly once across all splits, got {} rows"</span>,
+<a href=#715 id=715 data-nosnippet>715</a> ids.len()
+<a href=#716 id=716 data-nosnippet>716</a> );
+<a href=#717 id=717 data-nosnippet>717</a> }
+<a href=#718 id=718 data-nosnippet>718</a>
+<a href=#719 id=719 data-nosnippet>719</a> <span class="doccomment">/// When a split boundary lands exactly on a row group's midpoint, half-open ownership
+<a href=#720 id=720 data-nosnippet>720</a> /// (`start <= midpoint < end`) must hand that row group to the upper split only: the lower
+<a href=#721 id=721 data-nosnippet>721</a> /// split ends at the midpoint and so excludes it, the upper split starts at the midpoint and
+<a href=#722 id=722 data-nosnippet>722</a> /// so claims it. Two splits meeting exactly at the middle row group's midpoint must therefore
+<a href=#723 id=723 data-nosnippet>723</a> /// read every row once, with the middle row group going to the upper split.
+<a href=#724 id=724 data-nosnippet>724</a> </span><span class="attr">#[tokio::test]
+<a href=#725 id=725 data-nosnippet>725</a> </span><span class="kw">async fn </span>test_split_boundary_on_row_group_midpoint() {
+<a href=#726 id=726 data-nosnippet>726</a> <span class="kw">use </span>arrow_array::Int32Array;
+<a href=#727 id=727 data-nosnippet>727</a> <span class="kw">use </span>parquet::file::reader::{FileReader, SerializedFileReader};
+<a href=#728 id=728 data-nosnippet>728</a>
+<a href=#729 id=729 data-nosnippet>729</a> <span class="kw">let </span>schema = Arc::new(
+<a href=#730 id=730 data-nosnippet>730</a> Schema::builder()
+<a href=#731 id=731 data-nosnippet>731</a> .with_schema_id(<span class="number">1</span>)
+<a href=#732 id=732 data-nosnippet>732</a> .with_fields(<span class="macro">vec!</span>[
+<a href=#733 id=733 data-nosnippet>733</a> NestedField::required(<span class="number">1</span>, <span class="string">"id"</span>, Type::Primitive(PrimitiveType::Int)).into(),
+<a href=#734 id=734 data-nosnippet>734</a> ])
+<a href=#735 id=735 data-nosnippet>735</a> .build()
+<a href=#736 id=736 data-nosnippet>736</a> .unwrap(),
+<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">let </span>arrow_schema = Arc::new(ArrowSchema::new(<span class="macro">vec!</span>[
+<a href=#740 id=740 data-nosnippet>740</a> Field::new(<span class="string">"id"</span>, DataType::Int32, <span class="bool-val">false</span>).with_metadata(HashMap::from([(
+<a href=#741 id=741 data-nosnippet>741</a> PARQUET_FIELD_ID_META_KEY.to_string(),
+<a href=#742 id=742 data-nosnippet>742</a> <span class="string">"1"</span>.to_string(),
+<a href=#743 id=743 data-nosnippet>743</a> )])),
+<a href=#744 id=744 data-nosnippet>744</a> ]));
+<a href=#745 id=745 data-nosnippet>745</a>
+<a href=#746 id=746 data-nosnippet>746</a> <span class="kw">let </span>tmp_dir = TempDir::new().unwrap();
+<a href=#747 id=747 data-nosnippet>747</a> <span class="kw">let </span>table_location = tmp_dir.path().to_str().unwrap().to_string();
+<a href=#748 id=748 data-nosnippet>748</a> <span class="kw">let </span>file_path = <span class="macro">format!</span>(<span class="string">"{table_location}/midpoint.parquet"</span>);
+<a href=#749 id=749 data-nosnippet>749</a>
+<a href=#750 id=750 data-nosnippet>750</a> <span class="comment">// Three row groups of 100 rows each (ids 0..300).
+<a href=#751 id=751 data-nosnippet>751</a> </span><span class="kw">let </span>props = WriterProperties::builder()
+<a href=#752 id=752 data-nosnippet>752</a> .set_compression(Compression::SNAPPY)
+<a href=#753 id=753 data-nosnippet>753</a> .set_max_row_group_row_count(<span class="prelude-val">Some</span>(<span class="number">100</span>))
+<a href=#754 id=754 data-nosnippet>754</a> .build();
+<a href=#755 id=755 data-nosnippet>755</a> <span class="kw">let </span>file = File::create(<span class="kw-2">&</span>file_path).unwrap();
+<a href=#756 id=756 data-nosnippet>756</a> <span class="kw">let </span><span class="kw-2">mut </span>writer = ArrowWriter::try_new(file, arrow_schema.clone(), <span class="prelude-val">Some</span>(props)).unwrap();
+<a href=#757 id=757 data-nosnippet>757</a> <span class="kw">for </span>chunk <span class="kw">in </span>[<span class="number">0</span>..<span class="number">100</span>, <span class="number">100</span>..<span class="number">200</span>, <span class="number">200</span>..<span class="number">300</span>] {
+<a href=#758 id=758 data-nosnippet>758</a> <span class="kw">let </span>batch = RecordBatch::try_new(arrow_schema.clone(), <span class="macro">vec!</span>[Arc::new(
+<a href=#759 id=759 data-nosnippet>759</a> Int32Array::from(chunk.collect::<Vec<i32>>()),
+<a href=#760 id=760 data-nosnippet>760</a> )])
+<a href=#761 id=761 data-nosnippet>761</a> .unwrap();
+<a href=#762 id=762 data-nosnippet>762</a> writer.write(<span class="kw-2">&</span>batch).expect(<span class="string">"Writing batch"</span>);
+<a href=#763 id=763 data-nosnippet>763</a> }
+<a href=#764 id=764 data-nosnippet>764</a> writer.close().unwrap();
+<a href=#765 id=765 data-nosnippet>765</a>
+<a href=#766 id=766 data-nosnippet>766</a> <span class="comment">// Locate the middle row group's exact midpoint. Row groups are stored back to back after
+<a href=#767 id=767 data-nosnippet>767</a> // the 4-byte magic header.
+<a href=#768 id=768 data-nosnippet>768</a> </span><span class="kw">let </span>metadata = SerializedFileReader::new(File::open(<span class="kw-2">&</span>file_path).unwrap())
+<a href=#769 id=769 data-nosnippet>769</a> .unwrap()
+<a href=#770 id=770 data-nosnippet>770</a> .metadata()
+<a href=#771 id=771 data-nosnippet>771</a> .clone();
+<a href=#772 id=772 data-nosnippet>772</a> <span class="macro">assert_eq!</span>(metadata.num_row_groups(), <span class="number">3</span>);
+<a href=#773 id=773 data-nosnippet>773</a> <span class="kw">let </span>rg1_start = <span class="number">4 </span>+ metadata.row_group(<span class="number">0</span>).compressed_size() <span class="kw">as </span>u64;
+<a href=#774 id=774 data-nosnippet>774</a> <span class="kw">let </span>rg1_size = metadata.row_group(<span class="number">1</span>).compressed_size() <span class="kw">as </span>u64;
+<a href=#775 id=775 data-nosnippet>775</a> <span class="kw">let </span>rg1_midpoint = rg1_start + rg1_size / <span class="number">2</span>;
+<a href=#776 id=776 data-nosnippet>776</a> <span class="kw">let </span>file_end = rg1_start + rg1_size + metadata.row_group(<span class="number">2</span>).compressed_size() <span class="kw">as </span>u64;
+<a href=#777 id=777 data-nosnippet>777</a>
+<a href=#778 id=778 data-nosnippet>778</a> <span class="kw">let </span>file_size = std::fs::metadata(<span class="kw-2">&</span>file_path).unwrap().len();
+<a href=#779 id=779 data-nosnippet>779</a> <span class="kw">let </span>file_io = FileIO::new_with_fs();
+<a href=#780 id=780 data-nosnippet>780</a> <span class="kw">let </span>reader = ArrowReaderBuilder::new(file_io, Runtime::current()).build();
+<a href=#781 id=781 data-nosnippet>781</a>
+<a href=#782 id=782 data-nosnippet>782</a> <span class="comment">// Two splits meeting exactly at rg1's midpoint. The lower split ends there, the upper
+<a href=#783 id=783 data-nosnippet>783</a> // starts there; the middle row group must fall to the upper split alone.
+<a href=#784 id=784 data-nosnippet>784</a> </span><span class="kw">let </span><span class="kw-2">mut </span>per_split = Vec::new();
+<a href=#785 id=785 data-nosnippet>785</a> <span class="kw">for </span>(start, end) <span class="kw">in </span>[(<span class="number">0</span>, rg1_midpoint), (rg1_midpoint, file_end)] {
+<a href=#786 id=786 data-nosnippet>786</a> <span class="kw">let </span>task = FileScanTask::builder()
+<a href=#787 id=787 data-nosnippet>787</a> .with_file_size_in_bytes(file_size)
+<a href=#788 id=788 data-nosnippet>788</a> .with_start(start)
+<a href=#789 id=789 data-nosnippet>789</a> .with_length(end - start)
+<a href=#790 id=790 data-nosnippet>790</a> .with_data_file_path(file_path.clone())
+<a href=#791 id=791 data-nosnippet>791</a> .with_data_file_format(DataFileFormat::Parquet)
+<a href=#792 id=792 data-nosnippet>792</a> .with_schema(schema.clone())
+<a href=#793 id=793 data-nosnippet>793</a> .with_project_field_ids(<span class="macro">vec!</span>[<span class="number">1</span>])
+<a href=#794 id=794 data-nosnippet>794</a> .with_case_sensitive(<span class="bool-val">false</span>)
+<a href=#795 id=795 data-nosnippet>795</a> .build();
+<a href=#796 id=796 data-nosnippet>796</a>
+<a href=#797 id=797 data-nosnippet>797</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=#798 id=798 data-nosnippet>798</a> <span class="kw">let </span>batches = reader
+<a href=#799 id=799 data-nosnippet>799</a> .clone()
+<a href=#800 id=800 data-nosnippet>800</a> .read(tasks)
+<a href=#801 id=801 data-nosnippet>801</a> .unwrap()
+<a href=#802 id=802 data-nosnippet>802</a> .stream()
+<a href=#803 id=803 data-nosnippet>803</a> .try_collect::<Vec<RecordBatch>>()
+<a href=#804 id=804 data-nosnippet>804</a> .<span class="kw">await
+<a href=#805 id=805 data-nosnippet>805</a> </span>.unwrap();
+<a href=#806 id=806 data-nosnippet>806</a>
+<a href=#807 id=807 data-nosnippet>807</a> <span class="kw">let </span><span class="kw-2">mut </span>ids = Vec::new();
+<a href=#808 id=808 data-nosnippet>808</a> <span class="kw">for </span>batch <span class="kw">in </span><span class="kw-2">&</span>batches {
+<a href=#809 id=809 data-nosnippet>809</a> <span class="kw">let </span>col = batch
+<a href=#810 id=810 data-nosnippet>810</a> .column(<span class="number">0</span>)
+<a href=#811 id=811 data-nosnippet>811</a> .as_primitive::<arrow_array::types::Int32Type>();
+<a href=#812 id=812 data-nosnippet>812</a> ids.extend(col.values().iter().copied());
+<a href=#813 id=813 data-nosnippet>813</a> }
+<a href=#814 id=814 data-nosnippet>814</a> per_split.push(ids);
+<a href=#815 id=815 data-nosnippet>815</a> }
+<a href=#816 id=816 data-nosnippet>816</a>
+<a href=#817 id=817 data-nosnippet>817</a> <span class="macro">assert_eq!</span>(
+<a href=#818 id=818 data-nosnippet>818</a> per_split[<span class="number">0</span>],
+<a href=#819 id=819 data-nosnippet>819</a> (<span class="number">0</span>..<span class="number">100</span>).collect::<Vec<i32>>(),
+<a href=#820 id=820 data-nosnippet>820</a> <span class="string">"lower split, ending at rg1's midpoint, must read only rg0"
+<a href=#821 id=821 data-nosnippet>821</a> </span>);
+<a href=#822 id=822 data-nosnippet>822</a> <span class="macro">assert_eq!</span>(
+<a href=#823 id=823 data-nosnippet>823</a> per_split[<span class="number">1</span>],
+<a href=#824 id=824 data-nosnippet>824</a> (<span class="number">100</span>..<span class="number">300</span>).collect::<Vec<i32>>(),
+<a href=#825 id=825 data-nosnippet>825</a> <span class="string">"upper split, starting at rg1's midpoint, must read rg1 and rg2"
+<a href=#826 id=826 data-nosnippet>826</a> </span>);
+<a href=#827 id=827 data-nosnippet>827</a> }
+<a href=#828 id=828 data-nosnippet>828</a>}
</code></pre></div></section></main></body></html>
\ No newline at end of file