<Hodgestar>
phlebas: How did you find out about it?
<phlebas>
Hodgestar: Max Bernstein sent an email about it
<Hodgestar>
Ah.
<Hodgestar>
Maybe we can just invite Mark to come talk to us and join our efforts? It's no like we're not open to suggestions.
<Hodgestar>
And we have a working code base that already supports lots of things.
<phlebas>
:+1
<phlebas>
the documents on the repo right now also don't deal with a lot of the issues we already looked at, like inheritance and storing objects in C structs and tp_traverse and such
<Hodgestar>
Indeed.
<phlebas>
i think we should write a PEP about HPy as Python C API soon. it would be a good place to summarize our findings also when porting numpy, kiwi, and matplotlib - because the SC will want to know about the impact and drawbacks for the ecosystem for any new C API PEP
<phlebas>
and we have a lot of that data already, so there's a basis for discussion
<Hodgestar>
And if Mark has a plan to ease adoption / integration of a new C API, I would be very keen to hear that.
<Hodgestar>
I'd like to chat to Mark and suggest writing a joint PEP.
<Hodgestar>
Writing PEPs at each other seems a bit unfriendly, unless the ideas really are too different to reconcile.
<Hodgestar>
We should try see this as a big opportunity to get HPy traction with the team working on improving the CPython interpreter.
<Hodgestar>
And in a way Mark writing down his C API ideas means that that team is encountering the problems with the C API that we wrote about in the HPy manifesto.
<phlebas>
it definitely means there will be a desire to move forward, which is good. i also think if someone (antocuni maybe?) could write mark to see if we could have an actual call about his goals and how we could join forces, things might resolve themselves
<Hodgestar>
+1
<cfbolz>
Hm, I wonder, can you write the old cpython api on top of hpy?
<cfbolz>
Or at least the subset that eg pypy supports
<Hodgestar>
cfbolz: I think on CPython, yes, and in other cases, it's like writing cpyext?
<Hodgestar>
cfbolz: Are you thinking of using this as a gradual porting route for CPython somehow?
<Hodgestar>
I worry that you run into similar problems to Victor Stinner's attempts to fix the C API slowly -- i.e. you annoy everyone with small changes for a long time, and then there has to be a huge change later anyway. Maybe that can be avoided somehow though?
<cfbolz>
Hodgestar: no, basically I was wondering whether this can be used as a legacy compatibility implementation of the current api
<cfbolz>
In case cpython really moves to something new
<Hodgestar>
It seems like quite a bit of work, but maybe it can be done incrementally?
<cfbolz>
Hodgestar: I would eg allow pypy and GraalPy to share efforts for their cpython extension implementation
<Hodgestar>
Could one borrow ideas and bits of code from cpyext?