05:54 <fangerer> GianlucaRizzo: welcome to the team!

05:56 <fangerer> cfbolz: that's a great slide set from Stefan. Didn't know about it. It already contains a lot references to publications of to the Graal project.

06:03 <fangerer> antocuni: Concerning the debug mode: Enabling the debug mode per module is, right now, possible. I'm not sure how it is on PyPy but at least on our side, this causes a significant more complicated implementation.

06:05 <fangerer> If the debug mode is enabled globally, it will also detect problems in used modules. I think that's not necessarily a drawback since those should be fixed anyway, right?

06:06 <fangerer> Anyway, we don't need to decide that now. Stepan prepared one more little PR (whichi we will include in HPy 0.0.4) that allows to enable the debug mode with env variable `HPY_DEBUG_CONTEXT`.

06:07 <fangerer> It possible to enable the debug mode for all modules (i..e `HPY_DEBUG_CONTEXT=1`) or for selected ones (i..e `HPY_DEBUG_CONTEXT=moduleA,moduleB`)

06:08 <fangerer> Please review the PR: https://github.com/hpyproject/hpy/pull/314

08:54 <mattip> This comment about using the limited API mode in Cython and getting a 10% slowdown might be relevant for HPy

08:54 <mattip> https://github.com/cython/cython/issues/4382#issuecomment-1099970367

10:28 <antocuni> fangerer: unless things changed recently, on PyPy the per-module debug is for free. Actually, considering the overall design of HPy, I am curious to know why on GraalPython it's not for free as well

10:28 <antocuni> considering that handles should be used only with the ctx they are generated from

10:29 <antocuni> potentially, we could generate a throw-away ctx for each call and things should still work

14:01 <antocuni> mattip: the last comment of the cython issue you posted, stefan says " I was recently asked what the performance difference is between the Limited API and CPython-optimised Cython". FWIW, the author of the question was me :)

14:01 <mattip> haha

14:03 <antocuni> I and phlebas met Stefan over a lunch during PyCon DE

14:04 <mattip> nice

14:08 <cfbolz> antocuni: ah, cool, what did you talk about?

14:10 <antocuni> in theory it was about HPy+Cython, although we didn't have any specific topic nor agenda to discuss, so the conversation digressed over various topics

14:11 <antocuni> HPy-wise, we talked about the possibility of having an argument-clinic like API to wrap C functions, and have a way to expose the original C function to Cython

14:11 <antocuni> so that Cython can call it without the argument packing/unpackng overhead

14:28 <cfbolz> antocuni: right

14:28 <cfbolz> That would be amazing

14:29 <cfbolz> Cython has all the infrastructure already

14:30 <antocuni> yes

14:30 <antocuni> and generally, Stefan seems to be interested in supporting it

14:30 <cfbolz> Cool

14:31 <cfbolz> I always planned to try to do a one day hack to get a prototype

14:31 <cfbolz> But never made the top of the Todo list