Antoine Pitrou
Antoine Pitrou
I don't think you need to accumulate CRC, you just have to run the CRC over the entire concatenated data exactly as it's written out on disk (which is rep-levels...
To be clear, I'm ok with doing only v1 here, but I would like to make sure we understand the complexity (or not) of handling v2 pages.
I think I'll do the following adjustments later: * check and clean up CRC32 implementation to follow the code style (should be mostly mechanical) * since we determined that the...
> If there are no objections I'd like to merge this before 12.0.0 window closes tomorrow (Tuesday). There is no need to rush this and I think it can wait...
I mean a more thorough review to check the implementation for potential issues.
@mapleFU Yes, I'll do. Sorry!
@rok Is this ready to review again?
Well, is the PR ready? @rok
> [](https://github.com/apache/arrow/blame/6cfe24633b9fe3c474137571940eca35d7a475dc/cpp/src/parquet/column_reader_test.cc#L1181-L1185) > > And this performance PR is avoiding this interface. So it failed. > > Any suggestions to fix this UT? I'm not sure I understand your question....
Instead of defining entire separate classes for this, why not change `EncodingTraits::Accumulator` to the following: ```c++ template struct EncodingTraits { // ... struct Accumulator { std::unique_ptr offsets_builder; std::unique_ptr data_builder; std::vector...