Segfault testing src/sage/libs/gap/element.pyx on python 3.12

[tornaria]

Concerning the error in src/sage/libs/gap/element.pyx, it happens in line 2489:

for i in range(100):
        rnd = [ randint(-10,10) for i in range(randint(0,7)) ]
        # compute the sum in GAP
        _ = libgap.Sum(rnd)
        try:
            libgap.Sum(*rnd)
            print('This should have triggered a ValueError')
            print('because Sum needs a list as argument')
        except ValueError:
            pass

getting Killed due to segmentation fault. Running this code in the terminal works, the error appears only in tests and with python 3.12.

Originally posted by @enriqueartal in #37011 (comment)

[dimpase]

my eyes bleed from this line:

        rnd = [ randint(-10,10) for i in range(randint(0,7)) ]

How about

        rnd = random.choices(range(-10,10), k=randint(0,7))

?

See https://docs.python.org/3/library/random.html#random.choices

[dimpase]

maybe it segfaults on libgap.Sum(*rnd) when rnd==[] ?

[dimpase]

This also pops up on Fedora 39 with system Python 3.12, see e.g. https://groups.google.com/g/sage-release/c/oTnTrEnB3YU/m/tZiLm66ZAwAJ

[dimpase]

But I can't reproduce this with Python 3.12.2

[enriqueartal]

Is it possible that both @jaapspies and me are missing some python3 system package?

[enriqueartal]

At some point this segfault occured only in tests, now it happens also in the command line

[kiwifb]

my eyes bleed from this line:

        rnd = [ randint(-10,10) for i in range(randint(0,7)) ]

How about

        rnd = random.choices(range(-10,10), k=randint(0,7))

?

See https://docs.python.org/3/library/random.html#random.choices

Tried that but it does not have an effect

sage: import random ## line 2490 ##
sage: for i in range(100):
    rnd = random.choices(range(-10,10), k=randint(0,7))
    # compute the sum in GAP
    _ = libgap.Sum(rnd)
    try:
        libgap.Sum(*rnd)
        print('This should have triggered a ValueError')
        print('because Sum needs a list as argument')
    except ValueError:
        pass ## line 2491 ##
------------------------------------------------------------------------
/usr/lib/python3.12/site-packages/cysignals/signals.cpython-312-x86_64-linux-gnu.so(+0xa0d4)[0x7fbe2865c0d4]
/usr/lib/python3.12/site-packages/cysignals/signals.cpython-312-x86_64-linux-gnu.so(+0xa196)[0x7fbe2865c196]
/usr/lib/python3.12/site-packages/cysignals/signals.cpython-312-x86_64-linux-gnu.so(+0xcbf6)[0x7fbe2865ebf6]
/usr/lib64/libc.so.6(+0x3a210)[0x7fbe2945f210]

[tornaria]

There's a more detailed trace in #36407 (comment)

[antonio-rojas]

Caused by GCC commit https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=049ec9b981d1f4f97736061d5cf7d0ae990b57d7

Upstream report: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114872

Can be reproduced by calling any libgap function in exactly 3 parameters, eg. libgap.AbelianGroup(0,0,0)

[dimpase]

@sheerluck has kindly disassembled broken element....so, resulting from compilation of element.c. I report the result here:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114872#c12
(that is, code generation is broken, the call to GAP_CallFunc3Args() is formed with one parameter less than it should be)

Fixing gcc is above my payrate, though :-)