09:29 <ctismer> Hi guys! I adapted PySide to nogil Python 3.9 . How is PyPy’s status concerning nogil? Do you have a nogil branch?

09:40 <cfbolz> not at this point, no

09:43 <ctismer> ah, that's sad. It would be such a great boost to PyPy, and I wanted to be early, this time :D

09:44 <cfbolz> ctismer: yeah, it's a bit sad, but we aren't enough people to do something that is a pretty complicated change at this point

09:51 <ctismer> Is there maybe a blog where the problem is explained?

09:59 <cfbolz> ctismer: which problem?

10:00 <cfbolz> arigato, ctismer: btw, pypy is turning 20 on friday (first day of the first sprint in hildesheim

10:01 <Hodgestar> ctismer: How does PySide do on nogil Python? Is there a performance benefit of some sort? Any bugs being exposed (either in nogil Python or PySide)?

10:01 <Hodgestar> cfbolz: That is a big milestone! :D

10:12 <ctismer> cfbolz: Hurray!!! And I turned from pa into grandpa.

10:12 <cfbolz> oh wow, congratulations :-)

10:15 <ctismer> Hodgestar: well I did not measure, yet (and I‘m afraid of a bulk of errors). I only worked on the necessary patches (with Sam Gross) to get all tests to pass.

10:19 <ctismer> arigato: what are you now working on, a different project?

10:25 <cfbolz> ctismer: armin works on https://vrsketch.eu/ mostly I think

10:27 <ctismer> Nice!

10:45 <ctismer> Hodgestar: But I assume using the main thread for the GUI like always, and using some worker threads that are Python only might work.

10:47 <ctismer> Will try to accelerate the Mandelbrot example. Too sad that it is not on NoGIL PyPy...

14:04 <lesshaste> pip install scipy fails. Is that to be expected?

14:04 <lesshaste> ERROR: Dependency "OpenBLAS" not found, tried pkgconfig

14:04 <lesshaste>

14:48 <mattip> lesshaste: if you use conda you will get a binary package already built, no need to compile

15:21 <ctismer> cfbolz: the problem why adoption of NoGIL is complex. I mean, I'm sure it is complex, since this makes huge efforts like STM no more necessary. I would need to analyse NoGIL itself what was essential, and then find out how this maps to PyPy, if at all. That's why I hoped there was something written, already.

20:07 <mattip> PR https://github.com/python/cpython/pull/24298 added a func_builtins field to the PyFuncObject

20:08 <mattip> then in response to issue https://bugs.python.org/issue42990, that field was exposed as func.__builtins__ for python 3.10

20:08 <mattip> Is that part of the python func object model or an implementation detail?

20:09 <mattip> if it is part of the model, we need to implement it. If only an implementation detail we can ignore it

20:13 <mattip> it seems like a change to the language model

20:30 <mattip> https://github.com/python/cpython/issues/101888

21:06 <cfbolz> mattip: we have kind of always ignored the ability to override the builtins

21:06 <cfbolz> it's even documented I think