alamb commented on PR #9778: URL: https://github.com/apache/arrow-rs/pull/9778#issuecomment-4316927101
> Yes @alamb , w.r.t the alignment concern. I checked this specifically before adding the fix. > > In `arrow-ipc/src/reader.rs`, the only direct `ScalarBuffer<...>` conversions are the two inside the `Union(fields, mode)` branch. Other array paths go through `ArrayDataBuilder::align_buffers(...)`, so this is the only place in the IPC reader where the Vec-backed buffer can bypass the usual alignment handling before typed access. Got it -- this makes sense However, that would then suggest to me that there will be cases where the current code on main does not copy the data (because the initial allocation was correctly aligned) but the new code *DOES* copy the data (because the allocation `read_buffer` chose was not aligned) 🤔 This may well not be covered in our benchmarks. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
