segfault in xmodalg package tests in the master branch

[olexandr-konovalov]

This is a new failure observed in standard tests of the Browse package run in the Docker container for the master branch (built on 24th April): https://travis-ci.org/gap-system/gap-docker-pkg-tests-master/jobs/370829546

########> Diff in /home/gap/inst/gap-master/pkg/Browse/tst/test.tst:311
# Input is:
BrowseGapManuals( "inline/collapsed" );
# Expected output:
# But found:
Error, recursion depth trap (5000)
########

This happens when GAP is started with -A, or all packages are loaded. Tests pass with default packages loaded at GAP startup.

CC @frankluebeck

[olexandr-konovalov]

For the record, in the same build, two other packages have other kinds of failures, may be related:

• @bh11 irredsol https://travis-ci.org/gap-system/gap-docker-pkg-tests-master/jobs/370829585
• @aodabas xmodalg https://travis-ci.org/gap-system/gap-docker-pkg-tests-master/jobs/370829616

[olexandr-konovalov]

@frankluebeck For Browse, I've restarted the test with the new container built this morning, and the problem disappeared.

[olexandr-konovalov]

@bh11: The problem with irredsol also vanished in the new build.

@aodabas: xmodalg still segfaults when all packages are loaded: https://travis-ci.org/gap-system/gap-docker-pkg-tests-master/jobs/370829616

I will rename the issue to reflect the new state.

[cdwensley]

PR gap-packages/xmodalg#20 in XModAlg attempts to deal with this problem.

[fingolfin]

If a package tests starts to segfault for a package that does not include a kernel extension, then in 99% of all cases, that's GAP's problem, not the packages: plain GAP code should not lead to a segfault, period, if it does, even if the package does something unusual, it's our (GAP's) fault. I'd make an exception for cases where they deliberately use a low-level, undocumented SNAKE_CASE function, but in all other cases, I blame GAP.

In this case, I wonder if the recent regression in GAP's master branch (see issue #2415) could be responsible for some of crashes?

[olexandr-konovalov]

@fingolfin I think so - this is no longer happens.