Makes sense. If we were concerned about downstream breaking, we could set an upper limit on feather's pyarrow dependency [1] so that anyone using that package still would get a working setup. Or maybe that's not worth it and we could just worry about that if we start seeing bug reports.
Neal [1]: https://github.com/wesm/feather/blob/master/python/setup.py#L71 On Tue, Jun 3, 2025 at 11:09 AM Jacob Wujciak <assignu...@apache.org> wrote: > +1 I like the idea of keeping the reader around for a bit longer! > > Wes McKinney <wesmck...@gmail.com> schrieb am Di., 3. Juni 2025, 17:02: > > > That sounds fine to me. > > > > On Tue, Jun 3, 2025 at 8:09 AM Antoine Pitrou <anto...@python.org> > wrote: > > > > > > > > Hello > > > > > > Arrow C++ still supports the very old file format "Feather V1" which > was > > > designed in 2016 and is superseded by the Arrow IPC file format. (*) > > > > > > (note: "Feather V2" is a synonym for Arrow IPC to encourage users of > > > "Feather V1" to migrate to IPC) > > > > > > I propose that we deprecate reading and writing legacy "Feather V1" > > > files in Arrow C++. We could then retire the functionality in one year > > > or so (or perhaps we can just retire the writer and keep the reader for > > > a bit longer). What do you think? > > > > > > Regards > > > > > > Antoine. > > > > > > > > > (*) https://github.com/wesm/feather/graphs/contributors > > > > > > > > >
