@Antoine: I've updated the PR with a prose description of the C Device Data interface. Sorry for the lack of that in the first place.
--Matt On Tue, May 23, 2023 at 10:34 AM Antoine Pitrou <anto...@python.org> wrote: > > Also, I forgot to say, but thanks a lot for doing this! We can hope this > will drastically improve interoperability between non-CPU data > frameworks and libraries. > > Regards > > Antoine. > > > Le 23/05/2023 à 16:32, Antoine Pitrou a écrit : > > > > Depends on what we're voting on? > > > > The C declarations seem fine to me (I'm a bit lukewarm on the reserved > > bits, but I understand the motivation), however I've posted comments as > > to how to document the interface. The current PR entirely lacks a prose > > description of the C Device Data Interface. > > > > Regards > > > > Antoine. > > > > > > Le 22/05/2023 à 18:02, Matt Topol a écrit : > >> Hello, > >> > >> Now that there's a rough consensus and a toy example POC[1], I would > like > >> to propose an official enhancement to the Arrow C-Data API > specification as > >> described in the PR[2]. The new ArrowDeviceArray/ArrowDeviceArrayStream > >> structs would be considered "experimental" and the documentation would > >> label them as such for the time being. > >> > >> Please comment, ask questions, and look at the PR and toy example POC as > >> needed. > >> > >> The vote will be open for at least 72 hours. > >> > >> [ ] +1 Add this to the C-Data API > >> [ ] +0 > >> [ ] -1 Do not add this to the C-Data API because... > >> > >> Thank you very much everyone! > >> -- Matt > >> > >> [1]: https://github.com/zeroshade/arrow-non-cpu > >> [2]: https://github.com/apache/arrow/pull/34972 > >> >
