build/pkgs/openblas/spkg-configure.m4: Reject version 0.3.22

[mkoeppe]

📚 Description

openblas 0.3.22 is broken, see:

• openblas: Update to 0.3.23 #35371
• BUG: complex determinants broken with openblas 0.3.22 scipy/scipy#18208
• 0.3.22 breaks functions downstream in GNU Octave OpenMathLib/OpenBLAS#3976

We reject it.

📝 Checklist

• The title is concise, informative, and self-explanatory.
• The description explains in detail what this PR is about.
• I have linked a relevant issue or discussion.
• I have created tests covering the changes.
• I have updated the documentation accordingly.

⌛ Dependencies

[tobiasdiez]

Shouldn't the conda version be restricted as well, otherwise this will lead to problems once 0.3.22 is released on conda (which happens with conda-forge/openblas-feedstock#150)

[mkoeppe]

Trusting the packagers to not package a broken version

[jhpalmieri]

With this branch applied to 10.0.beta6, I ran ./bootstrap and ./configure and it rejected the system's version of openblas, which is good. I am seeing doctest failures, though. For example:

File "src/sage/tests/books/computational-mathematics-with-sagemath/float_doctest.py", line 402, in sage.tests.books.computational-mathematics-with-sagemath.float_doctest
Failed example:
    NearlySingularMatrix(RDF,n).det() == 0.0
Expected:
    True
Got:
    False

and

File "src/sage/matrix/matrix_integer_dense.pyx", line 3722, in sage.matrix.matrix_integer_dense.Matrix_integer_dense.determinant
Failed example:
    A.determinant(algorithm='linbox',proof=False) == d
Expected:
    True
Got:
    False

The full list of failures:

sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/plot/plot3d/plot3d.py  # Killed due to segmentation fault
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/geometry/polyhedron/library.py  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/modular/modform_hecketriangle/readme.py  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/modular/modform_hecketriangle/abstract_space.py  # 3 doctests failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/tests/books/computational-mathematics-with-sagemath/float_doctest.py  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/graphs/generators/families.py  # 2 doctests failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/matrix/matrix_integer_dense.pyx  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/modular/modform_hecketriangle/space.py  # 2 doctests failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/graphs/graph_generators.py  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/combinat/posets/posets.py  # Killed due to abort
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/stats/time_series.pyx  # 8 doctests failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/doc/en/thematic_tutorials/numerical_sage/scipy.rst  # Killed due to segmentation fault
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/geometry/polyhedron/base7.py  # 1 doctest failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/doc/it/faq/faq-usage.rst  # Killed due to segmentation fault
sage -t --long --random-seed=244469894400536995489090131539290714853 src/doc/en/faq/faq-usage.rst  # Killed due to segmentation fault
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/matrix/matrix_double_dense.pyx  # 2 doctests failed
sage -t --long --random-seed=244469894400536995489090131539290714853 src/sage/numerical/sdp.pyx  # 2 doctests failed

Some of these are known issues being tracked elsewhere, but some are new.

[dimpase]

could you try uninstalling brew's openblas and see if it helps these tests?

[jhpalmieri]

First a clarification: I haven't run full tests on an Apple M2 machine, but I don't see any failures in src/sage/matrix, for example: I'm only seeing the failures on an Intel machine.

Second, I uninstalled homebrew's openblas and I got the same failures. (I did brew uninstall --ignore-dependencies openblas, for what it's worth.) These must be old failures and not due to this ticket, but it's a problem if Sage's openblas doesn't work well.

[GMS103]

• SageMath 10.0.beta7
• Apple Silicon Macs ("M" series)
• macOS 11.7.5, macOS 12.6.4, macOS 13.3
• Homebrew up to date but without OpenBLAS (still at version 0.3.22 which is buggy).

make installs OpenBLAS SPKG version 0.3.21.
make ptestlong succeeds (except for unrelated issues).

[jhpalmieri]

My earlier problems are probably just dependent on that particular machine. On another Intel mac, all relevant tests pass.

[dimpase]

I gather that 0.3.23, fixing this bug, is out

…

On Sun, 2 Apr 2023, 18:01 John H. Palmieri, ***@***.***> wrote: ***@***.**** approved this pull request. — Reply to this email directly, view it on GitHub <#35377 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJXYHEACPAMX7VFJTC6XRDW7GWHVANCNFSM6AAAAAAWLGSVZQ> . You are receiving this because your review was requested.Message ID: ***@***.***>

[GMS103]

Yes and no Dima, see OpenMathLib/OpenBLAS#3976 Guillermo

…

On Sun, 2 Apr 2023 at 19:14, Dima Pasechnik ***@***.***> wrote: I gather that 0.3.23, fixing this bug, is out Message ID: ***@***.***>

[jhpalmieri]

Even if a later version fixes this bug, we still need to black-list 0.3.22.