Antoine Pitrou
Antoine Pitrou
> More generally, is there a way we can flush out these issues more systematically? For instance, running tests under an allocator which intentionally returns unaligned buffers? That's a rather...
It seems we could also use SVE in `src/arrow/util/bpacking.cc`, `src/arrow/util/utf8_internal.h`, as well as `src/arrow/csv/lexing_internal.h` and `src/arrow/csv/writer.cc`?
I'm probably missing something, but does this get turned into a dictionary array at some point? Returning dictionary scalars isn't really useful... @westonpace
While the approach here is sane, perhaps we can try profiling this with TSAN enabled to see if there obvious optimizations to do in our test routines? (another possibility might...
Alternative is to simply bump the timeout a bit: https://github.com/apache/arrow/pull/14502
Now that https://github.com/apache/arrow/pull/14502 is merged, I think we can close this.
Note I also profiled as I suggested above (albeit without TSAN), and unfortunately no particular contributor stood out.
Yes, definitely, you cannot interpret one type of array like this as another.
I'm not sure it's a good idea to do this before the HALF logical type is standardized in Parquet.
* Labels: I think we should review and clean up existing GH labels, which are a bit ad hoc. Any "official" labels should be namespaced so that they can denote...