Attention is currently required from: fixeria.
1 comment:
Patchset:
Looks good to me, thank you! […]
Vadim, thanks for feedback.
There are drawbacks of both approaches: if we do it earlier in the patchset, then faketrx cannot work without cython at all. However this fix does not require cython and can be done in pure python. It brings significant benefit just by itself, but unfortunately that speedup is not enough.
Overall the problem here is similar to bootstrapping: you cannot really do via single-step without braking something: do single step in plain py - we break compatibility with py < 3.13. Do single step with upgrading to cython: we break compatibility with current tests because they all do currently fail as it is not programmed there to build fake_trx before running it.
I choosed to do this step in plain py without bringing cython yet, because it looks more logical to me to fix this while being at plain py level.
Reworking the patches to bring cython earlier would be alot of internal inter-patch churn without, in my view, strong benefit. I would appreciate if we can leave the order of steps as they are currently.
Kirill
P.S. However one thing that could be maybe possible to maybe do is to merge the patch of pyx timerfd right after switch to cython. Then the time where fake_trx becomes py3.13-only will be limited.
To view, visit change 40047. To unsubscribe, or for help writing mail filters, visit settings.