Antoine Pitrou
Antoine Pitrou
(I'd also be in favor of not bothering with universal wheels, btw)
Note there are a couple basic benchmarks currently in `src/arrow/json/parser_benchmark.cc`. You can build them by passing `-DARROW_BUILD_BENCHMARKS=ON` to CMake (be sure to build in release mode for optimizations). It will...
Thanks @benibus . I think it would be nice to first submit the benchmark changes as a separate PR. That way, we'll be able to better compare the performance changes...
For the record, I think the benchmarks should have another parameter: the proportion of "present" fields. Currently it's implicitly 1.0, i.e. all fields are present in every row; but we...
> Will do. Should I keep this one open or convert it to a draft until the separate PR is merged? Converting to draft sounds fine. > Also, I reverted...
In bpo-39681 we reestablished the fallback to read() when a file-like object doesn't provide readinto() in \_pickle.c. However, doing so leads to lower performance and all file-like object should nowadays...
I'd like to hear if people have a problem with using Arrow C++ here. Since only a minimal build of Arrow C++ should be required, you can probably easily link...
It seems it is a bug in CloudServer's implementation of `ListObjects` and `ListObjectsV2`. If you have a bucket named `bucket` with a single file named `bucket/foo/bar.baz.txt`, then the following requests...
These requests do return a non-zero number of common prefixes, though, so I can use that. Workaround in https://github.com/apache/arrow/pull/11977