00:05 <belm0> re. "an experiment to switch CPython to a register VM", it's interesting, although they claim typical speedup of 12% on Python 3.10, while CPython 3.11 gives about the same speedup.  So the question is whether register VM will give the same size boost to 3.11 or not.

00:08 <belm0> So I tried pyston-lite on my app.  Kudos for achieving ~10% speedup (on python 3.8) with just a pip install.  But we can get the same speedup from CPython 3.11.

00:10 <belm0> Now faced with trying pyston-full, which needs a compile of all app dependencies with C extensions, and hope for 20%+ speedup.  Indeed pyston-full will no longer be maintained, but hopefully this will be a stopgap until future CPython versions can match the performance.

00:16 <belm0> the pyston team's plan:  "We currently are unable to use all of our techniques in Pyston-lite due to having less control over the system, but we are in discussions with the CPython team to add more JIT hooks to Python 3.12. Ideally we will be able to offer an extension module for 3.12 that has the same performance as a full fork of CPython [i.e.

00:16 <belm0> pyston-full]."

00:38 <belm0> actually, for RegCPython, the claimed 12% geometric mean speedup is reduced to 6-9% when PGO is enabled :(

00:38 <belm0> "With PGO enabled, the geometric mean of RegCPython’s speedup over CPython on all benchmarks became 1.056 on x86 and 1.092 on ARM"

08:49 <belm0> running my app on RegCPython:  5% speedup.   Small, but still promising that it can compose with CPython 3.11.  I confirmed with author of the paper, though the rebase will be difficult.

08:49 <belm0> running my app on pyston-full:  20% speedup

09:01 <fijal> belm0: I kinda doubt you are going to get better speedups with pypy and probably not without putting in significant work

15:48 <leshaste> should I open a bug report about getting matplotlib to work? I am guessing this is something a lot of people would want to do