Ubuntu 24.04 for github actions

[jodavies]

For running the tests on my Flint branch I moved the github actions image to 24.04. Then we have various failures which are nothing to do with Flint. Here is a run of the master branch on 24.04:
https://github.com/jodavies/form/actions/runs/12926713489/job/36050328201

As I understand Sta_Fill_1 is supposed to fail, certainly on my machine (24.04) this is a segfault, but it doesn't take 10s to finish. The options here are to increase the timeout for this test, or to change the test such that it doesn't crash. The point of the test is to check for the warning, not to actually demonstrate the crash I think.

The parform failures all seem to be due to printing from multiple threads messing up the logs. I can't reproduce this locally, but I am using openmpi and not mpich.

[tueda]

I agree that the Sta_Fill_1 test case doesn't need to test the crash;

    Table B(1:1);
    Local dummy = 1;
    .sort
    Fill B(1) = dummy;
    .end
    assert finished?
    assert warning?

should be fine.

Though Valgrind+MPI is expected not to work on >= 22.04 (#429), it is very strange (I can reproduce it on Docker) that the ParFORM build gave warnings like

In member function ‘__ct ’,
    inlined from ‘init’ at grcc.cc:3837:40,
    inlined from ‘__ct_base ’ at grcc.cc:3728:9:
grcc.cc:5133:32: warning: argument 1 value ‘18446744073709551615’ exceeds maximum object size 9223372036854775807 [-Walloc-size-larger-than=]
 5133 |     opics    = new MCOpi[snodes];
      |                                ^
/usr/include/c++/13/new: In member function ‘__ct_base ’:
/usr/include/c++/13/new:128:26: note: in a call to allocation function ‘operator new []’ declared here
  128 | _GLIBCXX_NODISCARD void* operator new[](std::size_t) _GLIBCXX_THROW (std::bad_alloc)
      |                          ^

In function ‘Malloc1’,
    inlined from ‘GCDfunction’ at ratio.c:960:22:
tools.c:2322:15: warning: argument 1 range [18446744004990074880, 18446744073709551584] exceeds maximum object size 9223372036854775807 [-Walloc-size-larger-than=]
 2322 |         mem = (VOID *)M_alloc(size);
      |               ^
/usr/include/stdlib.h: In function ‘GCDfunction’:
/usr/include/stdlib.h:672:14: note: in a call to allocation function ‘malloc’ declared here
  672 | extern void *malloc (size_t __size) __THROW __attribute_malloc__
      |              ^

and the test run prints 0 workers:

ParFORM 4.1 (Dec 10 2024, v4.1-20131025-673-g78b7e9b) 0 workers  Run: Thu Jan 23 10:02:31 2025

[jodavies]

Is this with openmpi or mpich? At least on my 24.04 system with openmpi, I don't see this warning in the build.

[tueda]

This is MPICH because Open MPI is not Valgrind-error-free (the same for recent MPICH, though).

[jodavies]

and the test run prints 0 workers:

ParFORM 4.1 (Dec 10 2024, v4.1-20131025-673-g78b7e9b) 0 workers  Run: Thu Jan 23 10:02:31 2025

This is the reason for the multiple-print of everything presumably. mpirun is just running 4 independent master parform processes.

[jodavies]

https://github.com/jodavies/form/actions/runs/12931507401/job/36067004950

The failures on this run are due to awalsh128/cache-apt-pkgs-action#145 . As I understand we can "just wait" for updated tags and things should run again. FORM's actions config is using v4 already.