apache / parquet-cpp / 388

push

travis-ci

PARQUET-533: Add a Buffer abstraction, refactor input/output classes to be simpler using Buffers

I have also removed all RowGroupReader and ColumnReader caching until we have an idea of how the caching will help users while also keeping memory-use under control.

The goal with this patch is to encapsulate the "data pointer and size" concept and lightly abstract away buffer ownership. The particular motivation is being able to deal with both normal files (bytes arriving via `fread`) and memory-mapped files (copying of bytes into new memory not required). It also helps do away with a bunch of functions that write output into a `std::vector<uint8_t>` now in favor of the `OwnedMutableBuffer`. Feedback welcome; plenty more work that can be done here.

Requires PARQUET-457. Will rebase once that's merged.

Author: Wes McKinney <wesm@apache.org>

Closes #59 from wesm/PARQUET-533 and squashes the following commits:

a39a4bc [Wes McKinney] Add Buffer abstraction, refactor input/output interface classes to use it where relevant. Output Buffer from InMemoryOutputStream. Stop caching ColumnReader objects

124 of 124 new or added lines in 15 files covered. (100.0%)

2704 of 2929 relevant lines covered (92.32%)

10484.24 hits per line