Describe the enhancement requested
This is part of a long-term series of work to be able to short-circuit decoding Arrow nullability bitmap from Parquet levels when the bit width is one. Currently the levels are unpacked to int16_t unconditionally before being re-encoded in a bitmap.
Previously, in GH-47112GH-47294 we split the mixed Rle-BitPacked into a parser capable of emitting both type of runs.
Next, we need to introduce a decoder for Rle/BitPacked parquet encoding optimized for decoding directly into a bitmap (to be used when source and target bit_width = 1).
Component(s)
C++, Parquet
Describe the enhancement requested
This is part of a long-term series of work to be able to short-circuit decoding Arrow nullability bitmap from Parquet levels when the bit width is one. Currently the levels are unpacked to
int16_tunconditionally before being re-encoded in a bitmap.Previously, in GH-47112GH-47294 we split the mixed Rle-BitPacked into a parser capable of emitting both type of runs.
Next, we need to introduce a decoder for Rle/BitPacked parquet encoding optimized for decoding directly into a bitmap (to be used when source and target bit_width = 1).
Component(s)
C++, Parquet