GNU bug report logs - #55565
[PATCH] gnu: Add python-blis

Previous Next

Full log

Message #14 received at 55565 <at> debbugs.gnu.org (full text, mbox):

From: Ludovic Courtès <ludo <at> gnu.org>
To: "Artyom V. Poptsov" <poptsov.artyom <at> gmail.com>
Cc: 55565 <at> debbugs.gnu.org
Subject: Re: bug#55565: [PATCH] gnu: Add python-blis
Date: Mon, 13 Jun 2022 12:20:18 +0200

Hi Artyom,

"Artyom V. Poptsov" <poptsov.artyom <at> gmail.com> skribis:

> Basically I patched 'blis/benchmark.py' to use "blas_opt_info" instead
> of "blas_ilp64_opt_info" and tests went fine.  An issue with different
> numpy versions I guess?  I also added "blis" and "python-numpy-next" to
> the "native-inputs".

Unfortunately this is not sufficient: the source bundles a copy of BLIS
under ‘blis/_src’ and it starts by building it (which is why it takes so
long), whether or not BLIS is among the inputs.

Could you (1) add a snippet that removes ‘blis/_src’, and (2) see
whether/how ‘setup.py’ can be patched to not build BLIS?  It might be
that commenting out the line that reads:

    cmdclass={"build_ext": ExtensionBuilder},

would be enough, I don’t know.

Anyhow, given that BLIS is the kind of package that’s highly tuned on
our side for performance and reproducibility configuration, it’s
important to not have several copies around.

> +    (native-inputs (list python-numpy-next
> +                         python-pytest
> +                         python-cython
> +                         blis
> +                         python-hypothesis))

NumPy and BLIS should definitely be ‘inputs’, not ‘native-inputs’; not
sure about ‘hypothesis’.

Also, please include a short comment explaining why numpy-next is used
rather than numpy (like “version >= X.Y required”).

Could you send an updated patch?

Sorry that this is providing trickier than we’d like!

Thanks,
Ludo’.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.