Re: math/py-scipy build failure on 2023Q4 branch

["David H. Gutteridge" <david%gutteridge.ca@localhost>]

On Tue, 2 Jan 2024 at 11:41:15 +0100, Benny Siegert wrote:
> Hi!
>
> This is a failure I saw on a verification build for a pullup on the
> latest branch. NetBSD 9, amd64, no special settings in mk.conf.
>
> The full build log is attached; the failure seems to be related to 
> numpy?
[...]

I'm not able to test in an identical (pure 2023Q4) environment right
now, but I can tell you that py-scipy from 2023Q4 does build for me.
I have py-numpy from current pkgsrc; it differs by one commit from
what I see:

revision 1.117
date: 2024-01-02 14:57:46 -0500;  author: thor;  state: Exp;  lines: +7 
-5;  commitid: CpYPM6mz2jLWCUSE;
math/py-numpy: actually blas=blas (cblas implicit) for more correct 
linkage

Now, when trying to build everything from current pkgsrc as of less
than a day ago, I actually hit an entirely different problem with
py-cython instead:

=> Generating post-install file lists
pkg_create: can't stat 
`/home/disciple/NetBSD/pkgsrc/devel/py-cython/work/.destdir/usr/pkg/lib/python3.11/site-packages/Cython/Shadow.pyi'
=> Checking file-check results for py311-cython-3.0.7
ERROR: ************************************************************
ERROR: The following files are in the PLIST but not in 
/home/disciple/NetBSD/pkgsrc/devel/py-cython/work/.destdir/usr/pkg:
ERROR:         
/home/disciple/NetBSD/pkgsrc/devel/py-cython/work/.destdir/usr/pkg/lib/python3.11/site-packages/Cython/Shadow.pyi
*** Error code 1

Here I'm building on 9.3_STABLE/amd64, also nothing "special" in my
mk.conf.

> There are also warnings about conflicting libfortran versions, which
> worry me.

py-numpy and py-scipy want GCC>=8 while blas wants GCC>=7 (from
pkgsrc), which means we build both in pkgsrc for Fortran. This is
just another example of what happens when various packages pull in
different GCC versions as minimums (usually this comes up with C++).

Regards,

Dave