openblas fail on build from source
Aaron Lauve
***************************************************************
Error building Sage.
The following package(s) may have failed to build (not necessarily
during this run of 'make all-start'):
* package: openblas-0.3.28
last build time: Jul 10 10:59
log file: /Applications/sage/logs/pkgs/openblas-0.3.28.log
build directory: /Applications/sage/local/var/tmp/sage/build/openblas-0.3.28
lauve@edinburgh sage % brew info openblas
==> openblas: stable 0.3.30 (bottled), HEAD [keg-only]
Optimized BLAS library
Installed
/opt/homebrew/Cellar/openblas/0.3.30 (24 files, 50.4MB)
Poured from bottle using the formulae.brew.sh API on 2025-07-10 at 10:45:39
From: https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/o/openblas.rb
License: BSD-3-Clause AND BSD-2-Clause-Views AND BSD-3-Clause-Open-MPI AND BSD-2-Clause
==> Dependencies
Required: gcc ✔
==> Options
--HEAD
Install HEAD version
==> Caveats
openblas is keg-only, which means it was not symlinked into /opt/homebrew,
because macOS provides BLAS in Accelerate.framework.
For compilers to find openblas you may need to set:
export LDFLAGS="-L/opt/homebrew/opt/openblas/lib"
export CPPFLAGS="-I/opt/homebrew/opt/openblas/include"
For pkg-config to find openblas you may need to set:
export PKG_CONFIG_PATH="/opt/homebrew/opt/openblas/lib/pkgconfig"
Dima Pasechnik
thanks for the report.
The email interface to google groups allows bigger attachments, by the way.
(you can also email these logs directly to me)
There could be several reasons for all this, but first of all, it's
good to understand why the Homebrew's openblas is not accepted.
Can you send us the top-level config.log ?
As far as building from source is concerned, I assume you use gfortran
from Homebrew, right?
It could be that 1.3.28 is too old for your OS/hardware, and we need
to bump our package version.
Last but not the least, I'm at the moment working on allowing the user
to rather use the Apple Accelerate framework instead of openblas, see
https://github.com/dimpase/sage/tree/accel_wip
If you're interested in testing the latter, I can provide more instructions.
Best
Dima
https://pasechnik.info/dima/
On Thu, Jul 10, 2025 at 11:16 AM Aaron Lauve <la...@math.luc.edu> wrote:
>
> oops. forgot to attach the log. and then GoogleGroups complained it was too long.
> I'm attaching 'head -n1000' and 'tail -n1000' instead of the full log. Hope that's helpful.
>
> You received this message because you are subscribed to the Google Groups "sage-devel" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.
> To view this discussion visit https://groups.google.com/d/msgid/sage-devel/f5c7a22b-2dee-425d-bd0b-e540cffbce5fn%40googlegroups.com.
Dima Pasechnik
>
> Thanks, Dima.
>
>
>
> “Yes” to using gfortran provided by homebrew (at least I presume)
>
> Here’s what brew has to say about that:
>
>
>
> lauve@edinburgh sage % brew info gfortran
>
> ==> gcc: stable 15.1.0, HEAD
>
> GNU compiler collection
>
> https://gcc.gnu.org/
>
> Installed
>
> /opt/homebrew/Cellar/gcc/15.1.0 (1,957 files, 492.3MB) *
>
> Poured from bottle using the formulae.brew.sh API on 2025-06-24 at 09:16:18
>
> From: https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/g/gcc.rb
>
> License: GPL-3.0-or-later WITH GCC-exception-3.1
>
> ==> Dependencies
>
> Build: make ✘
>
> Required: gmp ✔, isl ✔, libmpc ✔, mpfr ✔, zstd ✔
> ==> Options
>
> --HEAD
>
> Install HEAD version
>
>
>
14.2.0 is Sage's version of gfortran, but there are lines
real_configure:11002: checking for gfortran
real_configure:11023: found /opt/homebrew/bin/gfortran
real_configure:11035: result: gfortran
real_configure:11061: checking for Fortran compiler version
real_configure:11070: gfortran --version >&5
GNU Fortran (Homebrew GCC 15.1.0) 15.1.0
[...]
real_configure:18788: will use system package and not install SPKG gfortran
which really say what it is, and what's going to be used.
>
> Attaching the toplevel config.log file.
right, I knew about this, I mentioned to Homebrew people that it's a bug:
https://github.com/Homebrew/homebrew-core/issues/209091
but they decided not to do anything about it.
The problem is that `openblas.pc` supplied by them does not work with
Apple's "gcc", i.e. clang.
It only works with "real" gcc, which we'd rather avoid using in order
not to create even more mess than what's our
current build system already is.
A quick an dirty fix is to replace their `openblas.pc` with the
corrected one, where openmp is not mentioned
at all - in the same vein as done in https://github.com/sagemath/sage/pull/40397
(I'll post such a file somewhere)
As OpenBLAS people were more keen to listen at
https://github.com/Homebrew/homebrew-core/issues/209091 they
https://github.com/OpenMathLib/OpenBLAS/pull/5180
and so it won't be hard to create a custom Homebrew tap with the
alternative, working with clang, configuration.
Although I think Accelerate is not the better path to go forward.
>
>
>
>
>
> Also, maybe “yes” to being an Apple Accelerate Guinea pig. Let’s see what transpires…
Thanks! I've just made a draft PR from the branch I mentioned
https://github.com/sagemath/sage/pull/40397
Would most appreciate feedback from you and any other volunteers.
Cheers,
Dima
>
>
>
>
>
>
>
> --
>
> Aaron Lauve (he/him/his)
>
> Drop in for a visit
>
> 773.508.3727 | BVM 501 | ala...@luc.edu
>
> ~~~~~~~~~~
>
> Associate Professor and Graduate Program Director for Mathematics
>
> Department of Mathematics & Statistics
>
> Loyola University Chicago
>
>
>
> From: Dima Pasechnik <dim...@gmail.com>
> Date: Thursday, July 10, 2025 at 11:54 AM
> To: sage-...@googlegroups.com <sage-...@googlegroups.com>, Lauve, Aaron <ala...@luc.edu>
> Subject: Re: [sage-devel] Re: openblas fail on build from source
>
> [You don't often get email from dim...@gmail.com. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
> Hi Aaron,
> thanks for the report.
> The email interface to google groups allows bigger attachments, by the way.
> (you can also email these logs directly to me)
>
> There could be several reasons for all this, but first of all, it's
> good to understand why the Homebrew's openblas is not accepted.
> Can you send us the top-level config.log ?
>
> As far as building from source is concerned, I assume you use gfortran
> from Homebrew, right?
> It could be that 1.3.28 is too old for your OS/hardware, and we need
> to bump our package version.
>
> Last but not the least, I'm at the moment working on allowing the user
> to rather use the Apple Accelerate framework instead of openblas, see
>
> Best
> Dima
>
>
>
>
> On Thu, Jul 10, 2025 at 11:16 AM Aaron Lauve <la...@math.luc.edu> wrote:
> >
> > oops. forgot to attach the log. and then GoogleGroups complained it was too long.
> > I'm attaching 'head -n1000' and 'tail -n1000' instead of the full log. Hope that's helpful.
> >
> > --
> > You received this message because you are subscribed to the Google Groups "sage-devel" group.
> > To unsubscribe from this group and stop receiving emails from it, send an email to sage-devel+...@googlegroups.com.
Aaron Lauve
Dima Pasechnik
./meson.build:80:0: ERROR: Compiler gfortran cannot compile programs.
[scipy-1.15.2] [spkg-install]
[scipy-1.15.2] [spkg-install] A full log can be found at /Applications/sage/local/var/lib/sage/venv-python3.13/var/tmp/sage/build/scipy-1.15.2/src/.mesonpy-7kk9_dof/meson-logs/meson-log.txt
Can you post the latter file, and the top level
config.log ?
Aaron Lauve
lauve@edinburgh src % pwd
/Applications/sage/local/var/lib/sage/venv-python3.13/var/tmp/sage/build/scipy-1.15.2/src
lauve@edinburgh src % ls -al
total 288
drwxr-xr-x 22 lauve admin 704 Jul 31 09:21 .
drwxr-xr-x 24 lauve admin 768 Aug 1 13:41 ..
drwxr-xr-x 4 lauve admin 128 Jul 31 09:21 .devcontainer
drwxr-xr-x 6 lauve admin 192 Jul 31 09:21 benchmarks
-rw-r--r-- 1 lauve admin 1277 Jul 31 09:21 CITATION.bib
-rw-r--r-- 1 lauve admin 1112 Jul 31 09:21 CONTRIBUTING.rst
-rw-r--r-- 1 lauve admin 59699 Jul 31 09:21 dev.py
drwxr-xr-x 2 lauve admin 64 Jul 31 09:21 dist
drwxr-xr-x 6 lauve admin 192 Jul 31 09:21 doc
-rw-r--r-- 1 lauve admin 1536 Jul 31 09:21 LICENSE.txt
-rw-r--r-- 1 lauve admin 12881 Jul 31 09:21 LICENSES_bundled.txt
-rw-r--r-- 1 lauve admin 6168 Jul 31 09:21 meson.build
-rw-r--r-- 1 lauve admin 759 Jul 31 09:21 meson.options
-rw-r--r-- 1 lauve admin 13973 Jul 31 09:21 mypy.ini
-rw-r--r-- 1 lauve admin 8511 Jul 31 09:21 PKG-INFO
-rw-r--r-- 1 lauve admin 6263 Jul 31 09:21 pyproject.toml
-rw-r--r-- 1 lauve admin 1270 Jul 31 09:21 pytest.ini
-rw-r--r-- 1 lauve admin 3576 Jul 31 09:21 README.rst
drwxr-xr-x 9 lauve admin 288 Jul 31 09:21 requirements
drwxr-xr-x 30 lauve admin 960 Jul 31 09:21 scipy
drwxr-xr-x 3 lauve admin 96 Jul 31 09:21 subprojects
drwxr-xr-x 24 lauve admin 768 Jul 31 09:21 tools
Dima Pasechnik
Your gfortran is rejected by the scipy build script. Does "brew doctor" report that everything is OK with your Homebrew install?
Aaron Lauve
lauve@edinburgh pkgs % brew doctor
Please note that these warnings are just used to help the Homebrew maintainers
with debugging if you file an issue. If everything you use Homebrew for is
working fine: please don't worry or file an issue; just ignore this. Thanks!
Warning: Some installed kegs have no formulae!
This means they were either deleted or installed manually.
You should find replacements for the following formulae:
arb
Warning: Some installed formulae are deprecated or disabled.
You should find replacements for the following formulae:
icu4c@76
six
lauve@edinburgh pkgs % brew info gcc
==> gcc: stable 15.1.0, HEAD
GNU compiler collection
Installed
/opt/homebrew/Cellar/gcc/15.1.0 (1,957 files, 492.3MB) *
Poured from bottle using the formulae.brew.sh API on 2025-06-24 at 09:16:18
From: https://github.com/Homebrew/homebrew-core/blob/HEAD/Formula/g/gcc.rb
License: GPL-3.0-or-later WITH GCC-exception-3.1
==> Dependencies
Build: make ✘
Required: gmp ✔, isl ✔, libmpc ✔, mpfr ✔, zstd ✔
==> Options
--HEAD
Install HEAD version
==> Analytics
install: 81,926 (30 days), 289,275 (90 days), 1,317,053 (365 days)
install-on-request: 45,610 (30 days), 163,188 (90 days), 680,062 (365 days)
build-error: 3,245 (30 days)
lauve@edinburgh pkgs % which gfortran
/opt/homebrew/bin/gfortran
Dima Pasechnik
over the latest beta, maybe it helps. If no, and
if upgrading xcode followed by upgrading homebrew didn't work,
I would have tried to build scipy separately,
just to see if it works.
Perhaps upgrading scipy might help (either to 1.15.3, or to the latest, 1.16.1)