Re: git: 3f582fcf38bb - main - math/py-numpy: Migrate to PEP517

On 2025-10-14 00:10, Antoine Brodin wrote:
> On Tue, Oct 14, 2025 at 8:20 AM Xin Li <delphij@freebsd.org> wrote:
>> On 2025-10-13 22:21, Antoine Brodin wrote:
>>> On Mon, Oct 13, 2025 at 8:37 AM Xin LI <delphij@freebsd.org> wrote:
>>>>
>>>> The branch main has been updated by delphij:
>>>>
>>>> URL: https://cgit.FreeBSD.org/ports/commit/?id=3f582fcf38bb0cee40f9346683490aceb3bda654
>>>>
>>>> commit 3f582fcf38bb0cee40f9346683490aceb3bda654
>>>> Author:     Matthew Wener <matthew@wener.org>
>>>> AuthorDate: 2025-10-11 17:57:18 +0000
>>>> Commit:     Xin LI <delphij@FreeBSD.org>
>>>> CommitDate: 2025-10-13 06:36:40 +0000
>>>>
>>>>       math/py-numpy: Migrate to PEP517
>>>>
>>>>       PR:     ports/287794
>>>
>>> This broke the ports tree and bulk -a
>>> Please fix asap or revert as this blocks packages builders.
>>>
>>> [00:00:39] Error: compute_deps_pkg failed to lookup pkgname for
>>> devel/meson-python@py312 processing package py312-numpy-1.26.4_8,1
>>> from math/py-numpy@py312 -- Is SUBDIR+=meson-python missing in
>>> devel/Makefile? And does the port provide the 'py312' FLAVOR?
>>> [00:00:39] Error: compute_deps_pkg failed to lookup pkgname for
>>> devel/meson-python@py310 processing package py310-numpy-1.26.4_8,1
>>> from math/py-numpy@py310 -- Is SUBDIR+=meson-python missing in
>>> devel/Makefile? And does the port provide the 'py310' FLAVOR?
>>> [00:00:39] Error: compute_deps_pkg failed to lookup pkgname for
>>> devel/meson-python@py39 processing package py39-numpy-1.26.4_8,1 from
>>> math/py-numpy@py39 -- Is SUBDIR+=meson-python missing in
>>> devel/Makefile? And does the port provide the 'py39' FLAVOR?
>>
>> I don't yet understand why this would cause breakages, but I've reverted
>> the change (with a new PORTREVISION bump) as a precaution for now in the
>> hope to unbreak the tree.
> 
> This was failing because default ports tree attempts to build the
> py311 py312 py310 py39 flavors for math/py-numpy ,  while only
> building the py311 flavor for devel/meson-python
> In  my opinion we should only attempt to build the default python
> flavor (py311 at the moment) for py-numpy.

I see. I'll need to investigate how Poudriere gathers dependency data to 
confidently create a solution.

For now, the attached patch (3-allflavors.patch) appears to fix the 
initial stage of the bulk -a build when the original commit is 
reapplied. This patch modifies the BUILD_DEPENDS of py-numpy to build 
with allflavors (maintainer CC'd). I haven't waited for the full build 
to complete, as it takes a significant amount of time on my machine, but 
the initial check appear to pass.

It appears very few ports use allflavors (I think the value of doing it 
at all is questionable as result). For this approach to be more broadly 
useful, we would likely need to have more Python library packages to 
build for allflavors as well.

Cheers,