does not work with Python 3.11

[dimpase]

On Linux:

  src/pyscipopt/scip.c:463:62: error: invalid use of incomplete typedef 'PyFrameObject' {aka 'struct _frame'}
    463 |   #define __Pyx_PyFrame_SetLineNumber(frame, lineno)  (frame)->f_lineno = (lineno)

this is due to PyFrameObject structure members have been removed from the public C API, cf
https://docs.python.org/3.11/whatsnew/3.11.html

No surprise, as PyFrameObject was always subject to change.

See also https://groups.google.com/d/msgid/sage-devel/42199586-647d-4dc0-9b88-9ff5c7ff8737n%40googlegroups.com.

[dimpase]

pyscipopt-4.2.0.log

[mattmilten]

This may be fixed in a future version of cython. It's not something we can fix in PySCIPOpt.

[mkoeppe]

Actually @mattmilten (sagemath/sage#35299 (comment)): the PyPI sdist of PySCIPOpt contains a scip.c that was prebuilt with an old Cython. Depending on timestamps when unpacking the sdist, this will not be recythonized.

[mkoeppe]

The solution: Merge #652, tag a new release....

[dimpase]

Actually @mattmilten (sagemath/sage#35299 (comment)): the PyPI sdist of PySCIPOpt contains a scip.c that was prebuilt with an old Cython. Depending on timestamps when unpacking the sdist, this will not be recythoniz

these 🥩🍄🔪☢️☠️⚰️ 🥩🍄🔪☢️☠️⚰️ 🥩🍄🔪☢️☠️⚰️ 🥩🍄🔪☢️☠️⚰️
at Cython who still tell ppl to distribute outdated *.c files - @robertwb

[dimpase]

This may be fixed in a future version of cython. It's not something we can fix in PySCIPOpt.

it is fixed in our patched Cython 0.29.33, see sagemath/sage#35084 or https://github.com/dimpase/cython/tree/0.29.33.py311patch

In order to build the latest pyscipopt with Python 3.11, I had to manually remove scip.c from the tarball,
because your setup prefers using it over running Cython.

Please make a new release, without scip.c.

Or you can just modify the current release tarball on PyPI and put it to Assets in
https://github.com/scipopt/PySCIPOpt/releases/tag/v4.2.0 (to do this, click on "Edit release" and upload the file)
This is something we would be able to use in Sage

[dimpase]

Even better, after removing scip.c from the tarball, I can build with unpatched Cython (0.29.33), or with Sage's
patched Cython 0.29.32.

So there is nothing to fix in Cython, the ball is in your court.

[mattmilten]

Should be fixed now. Thanks a lot for pushing for this - this release was long overdue!

[dimpase]

Your new release tarball on PyPI still contains the same generated C file. Even though it might be out of the woods for the Python 3.11, it probably means trouble down the road, just as well what we saw with 4.2.0.

Besides, it takes over 90% of the tarball, it's whooping 7.6MB (uncompressed).

[dimpase]

@mattmilten - would you accept a PR streamlining the setup.py, so that the C file is only generated when the extension is built
(and so it would not go to the tarball) ?

[mattmilten]

Hmm, no, we did it this way so Cython is no dependency when installing the package.

[dimpase]

But this is a hugely outdated approach, dating to times when installing Python packages was a dark art. Nowadays getting Cython is trivial via pip. I can point you to a Cython issue where there is an agreement that these Cython recommendations should be replaced with an instruction on how to set up a package to always run Cython. Of course it all can be done automatically, no need to tell your users to install anything extra.

…

On Sat, 18 Mar 2023, 12:10 Matthias Miltenberger, ***@***.***> wrote: Hmm, no, we did it this way so Cython is no dependency when installing the package. — Reply to this email directly, view it on GitHub <#660 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAJXYHAJLYGEXL6S4DL73EDW4WQZVANCNFSM6AAAAAAV5HQW3I> . You are receiving this because you authored the thread.Message ID: ***@***.***>