[GitHub] [arrow] jorisvandenbossche commented on a change in pull request #7526: ARROW-9146: [C++][Dataset] Lazily store fragment physical schema

[GitBox <gi...@apache.org>]

jorisvandenbossche commented on a change in pull request #7526:
URL: https://github.com/apache/arrow/pull/7526#discussion_r444486902



##########
File path: cpp/src/arrow/dataset/file_parquet.cc
##########
@@ -357,13 +355,20 @@ static inline Result<std::vector<RowGroupInfo>> AugmentRowGroups(
   return row_groups;
 }
 
-Result<ScanTaskIterator> ParquetFileFormat::ScanFile(
-    const FileSource& source, std::shared_ptr<ScanOptions> options,
-    std::shared_ptr<ScanContext> context, std::vector<RowGroupInfo> row_groups) const {
+Result<ScanTaskIterator> ParquetFileFormat::ScanFile(std::shared_ptr<ScanOptions> options,
+                                                     std::shared_ptr<ScanContext> context,
+                                                     FileFragment* fragment) const {
+  const auto& source = fragment->source();
+  auto row_groups = checked_cast<const ParquetFileFragment*>(fragment)->row_groups();
+
   bool row_groups_are_complete = RowGroupInfosAreComplete(row_groups);
   // The following block is required to avoid any IO if all RowGroups are
   // excluded due to prior statistics knowledge.
   if (row_groups_are_complete) {
+    // physical_schema should be cached at this point
+    ARROW_ASSIGN_OR_RAISE(auto physical_schema, fragment->ReadPhysicalSchema());
+    RETURN_NOT_OK(options->filter->Validate(*physical_schema));

Review comment:
       Didn't look in detail yet (so might be misreading, it's late here ;)), but a quick note: in general, when Fragments were instantiated with ParquetFactory (from a `_metadata` file) with populated statistics, I think it should be possible to filter out fragments without any additional IO.




----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

For queries about this service, please contact Infrastructure at:
users@infra.apache.org