Closed wcy-fdu closed 3 weeks ago
cc @liurenjie1024 @Xuanwo for awareness.
Looks ut failed, please help to fix it.
@liurenjie1024 or @Xuanwo are you able to re-run the tests on this? I can't explain why the test would fail the way that it has in CI and when I reproduce the change from this PR on main
locally, the test passes
@liurenjie1024 or @Xuanwo are you able to re-run the tests on this? I can't explain why the test would fail the way that it has in CI and when I reproduce the change from this PR on
main
locally, the test passes
Hi, @sdd I can reproduce the test failure on my mac. Let me try to rereun with update branch.
Strange, I was running it on M1 Mac too and it all passed! Oh well, glad you can repro
cc @wcy-fdu Would you mind to take a look at the ut failure?
cc @wcy-fdu Would you mind to take a look at the ut failure?
Sure, will fix ut outside of work time.
After investigation, I found that although ProjectionMask::leaves() does not require column_indices to be in order, creating PredicateConverter requires it to be in order, so it has to be sorted here anyway. Therefore, it seems unnecessary to remove sort.
The order of indices passed into
ProjectionMask::leaves
is meaningless, because "repeated or out of order indices will not impact the final mask"(refer to https://docs.rs/crate/parquet/latest/source/src/arrow/mod.rs#179). Therefore, we don't need to perform an extra sort, which increases the overhead.