Doctest failures with alarm on OS X

[jhpalmieri]

Steps To Reproduce

With OS X 13.6.3 (Intel) and OS X 14.2 (Apple M2), Sage 10.3.beta2 and later, make ptest and make ptestlong produce failures that were not present with 10.3.beta1.

Expected Behavior

Tests which used to pass should pass.

Actual Behavior

As noted, tests do not pass. The end of make ptestlong reports:

sage -t --random-seed=226393960629615338750434927109379046589 src/sage/libs/flint/nmod_poly_linkage.pxi  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/libs/gap/element.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/interfaces/gap.py  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/geometry/integral_points.pxi  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/libs/libecm.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/coding/linear_code.py  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/rings/factorint_pari.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/rings/complex_arb.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/rings/integer.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/rings/polynomial/polynomial_zmod_flint.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/rings/polynomial/polynomial_element.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/parallel/decorate.py  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/sets/recursively_enumerated_set.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/structure/coerce_actions.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/schemes/elliptic_curves/descent_two_isogeny.pyx  # Killed due to alarm
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/libs/giac/__init__.py  # Timed out
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/doctest/forker.py  # 5 doctests failed
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/tests/books/computational-mathematics-with-sagemath/mpoly_doctest.py  # 1 doctest failed
sage -t --random-seed=226393960629615338750434927109379046589 src/sage/doctest/control.py  # 8 doctests failed

The first failure, for example:

sage -t --long --random-seed=16301058625430036409274889310772014116 src/sage/libs/flint/nmod_poly_linkage.pxi
    Killed due to alarm
**********************************************************************
Tests run before process (pid=50812) failed:
sage: P.<x> = GF(32003)[] ## line 40 ##
sage: Q.<x> = GF(7)[] ## line 42 ##
sage: sig_on_count() # check sig_on/off pairings (virtual doctest) ## line 43 ##
0
sage: P.<x> = GF(32003)[] ## line 50 ##
sage: del x ## line 51 ##
sage: Q.<x> = GF(7)[] ## line 53 ##
sage: del x ## line 54 ##


  [snip]


sage: f^2 % g ## line 521 ##
15328*x + 6968
sage: pow(f, -2, g) ## line 524 ##
16346/(x + 251)
sage: (f^2 % g)^-1 ## line 526 ##
16346/(x + 251)
sage: pow(f, 5, g) ## line 529 ##
7231*x + 17274
sage: f^5 % g ## line 531 ##
7231*x + 17274
sage: n = 2^23 ## line 536 ##
sage: alarm(0.2); x^n; cancel_alarm() ## line 537 ##

Additional Information

• I believe the failures are due to Use GNU make tokenpool protocol to manage parallelism of doctesting #36640. git bisect led me to that, and then independently 10.3.beta1 + that PR led to the failures.
• I have been unable to reproduce these failures using sage -tp.
• Tests fail with make test, make ptest, make ptestlong.

Environment

- **OS**: OS X 13.6.3 and OS X 14.2, latest Xcode command-line tools
- **Sage Version**: 10.3.beta2
- as many homebrew packages installed as Sage requests

Checklist

• I have searched the existing issues for a bug report that matches the one I want to file, without success.
• I have read the documentation and troubleshoot guide

[mkoeppe]

Thanks for opening this.
I can confirm that this was introduced in 10.3.beta2:

• I see these failures in the CI MacOS for 10.3.beta2 (in stage-2 / local-macos (latest, homebrew-macos-usrlocal-minimal), https://github.com/sagemath/sage/actions/runs/7202678228/job/19631749158#step:9:20392)
• not present in 10.3.beta1 runs (e.g. https://github.com/sagemath/sage/actions/runs/7157804484/job/19492023218)

[mkoeppe]

The tokenpool code does play with SIGALRM. https://github.com/milahu/gnumake-tokenpool/blob/main/py/src/gnumake_tokenpool/tokenpool.py#L178
I'll try to prepare a fix

[mkoeppe]

#36948 has a tentative solution. I haven't been able to reproduce the failure yet locally, so if you have a chance, please try

[mkoeppe]

No, it doesn't fix it.

[jhpalmieri]

No, it doesn't fix it.

Confirmed, doesn't fix it for me either.

[mkoeppe]

Looks like I may have a solution that uses cysignals in combination with tokenpool.

[mkoeppe]

OK, the new version of #36948 does the trick according to my local tests