On 4/26/23 10:05, Paolo Bonzini wrote:
>>
>
> Thanks, this looks pretty good. Some changes I'd make for the non-RFC
> version:
>
> - I think we should just check in the meson wheel (which also removes
> the need for patch 12, so it can be done in its stead) and remove the
> submodule
OK, if there's no objections.
As a consequence, meson will likely never be downloaded from PyPI with the patch set written as-is.
No real problem with that, just a difference.
>
> - The verbosity of mkvenv.py can be tuned down and most prints replaced
> with logger.info() or logger.debug()
😅
>
> - While I agree with keeping patch 18 separate, I would move it earlier
> so that patch 19 can be squashed into patch 14
>
Agree. It was newer / more experimental so I kept it unsquash'd. If you like it enough, asquashing I shall go.
> - I am ambivalent about keeping --enable/--disable-pypi in the first
> committed patchset, but in any case I would move patches 16 and 20
> before patch 15
I might be stubborn but I think I want to keep it in for now. If it needs redesigned to fit with the other flags you want to add, I think that's OK.
if we vendor the whl directly in qemu.git we won't need PyPI for meson, but it's still useful for Sphinx so I think I'm still leaning towards keeping it.
I'll try to refactor to keep it at the tail end of the series.
Just one extra thing, since we're changing so much of Python handling
and since the code is written, I would keep the Debian 10 workarounds
for now, and only drop them after we drop support for 3.6.
This series was written assuming we get to drop 3.6 as a prereq. Is that not the case?
Or did you mean to write 3.7 there?