Open ericsnowcurrently opened 2 years ago
I'll start. @zooba wrote down what I think are some important design principles:
https://mail.python.org/archives/list/capi-sig@python.org/thread/B2VDVLABM4RQ4ATEJXFZYWEGTBZPUBKW/
I've opened #7 for a deeper discussion.
Here is the high-level HPy API manifesto and the slightly longer explanation of each item:
And the HPy implementation https://github.com/hpyproject/hpy/ as prior art.
HPy can already run kiwisolver, matplotlib, and enough of numpy to run a simple Game of Life. The links are to the respective ports.
@vstinner has also made detailed proposals in the past and written extensively about it:
(Victor's personal Python notes at https://pythondev.readthedocs.io/ probably also have some related material.)
Here's a discussion I started about having an official cython-like tool:
https://mail.python.org/archives/list/capi-sig@python.org/thread/NSREULQDHSJ6SZTOY5YVSNX4U4AYJOI6/
A discussion started by @hodgestar about replacing the C-API with CFFI:
https://mail.python.org/archives/list/capi-sig@python.org/thread/ZV3IQWW7XL6MHZAJARHW62WNZDCX3A4M/
@vstinner also publishes a backward-compatibility shim header that relates here:
There have been a number of informative discussions over the years (especially the last few years) about the future of the C-API. It would be worth taking advantage those discussions.
We can start by leaving links to those discussions here. Any of them that might warrant a deeper discussion could also get separate issues. (There's already some discussion about HPy.)