Workaround for error #3055 and fix for GQuotient

[hulpke]

This is a bug that arises very rarely, maybe based on random
selections. In this case restart calculations.

(The bug was reported in #3055)

Also fix for GQuotient running slow in certain cases of groups with infinite quotients.

[olexandr-konovalov]

Thanks @hulpke! I've edited the description to add #3055 to the text - otherwise there is no visible link to click to navigate to it (one should either do it manually or click trough the commit message).

[codecov]

Codecov Report

Merging #3110 into master will decrease coverage by <.01%.
The diff coverage is 50%.

@@            Coverage Diff             @@
##           master    #3110      +/-   ##
==========================================
- Coverage   83.57%   83.57%   -0.01%     
==========================================
  Files         686      685       -1     
  Lines      336779   336512     -267     
==========================================
- Hits       281461   281227     -234     
+ Misses      55318    55285      -33

Impacted Files	Coverage Δ
lib/clashom.gi	78.87% <50%> (-0.07%)	⬇️
src/blister.h	88.28% <0%> (-0.91%)	⬇️
lib/init.g	80.03% <0%> (-0.55%)	⬇️
src/system.c	69.63% <0%> (-0.1%)	⬇️
src/gap.c	83.2% <0%> (-0.01%)	⬇️
src/hpc/thread.c	56.91% <0%> (ø)	⬆️
src/sysfiles.c	40.59% <0%> (ø)	⬆️
src/libgap-api.c	20.16% <0%> (ø)	⬆️
src/julia_gc.c
lib/streams.gi	70.45% <0%> (+0.21%)	⬆️
... and 2 more

[fingolfin]

I'd approve the changes, but the test in its current form does not seem suitable for merging.

[fingolfin]

This tests was running for several minutes on my machine, then run out of memory.

That's not acceptable for tst/testbugfix, whch we run as part of our CI on every pull request. It could be added to tst/testextra, though (perhaps in a new dir tst/testextra/bugfix)

[hulpke]

Let me check. This example should run in under a second. Probably it is the assertions again...

[hulpke]

Dear @fingolfin
That is really strange. I just rebased the branch, recompiled and ran the file with Test, and it took indeed under a second. Also, in the travis log I get

testing: /home/travis/build/gap-system/gap/tst/testbugfix/2018-12-17-gquotient.tst
 923 ms (24 ms GC) and 28.7MB allocated for 2018-12-17-gquotient.tst

which is the same magnitude.
Is it possible that you are loading something else that interferes? Could it be the Julia issue Thomas was observing?

[fingolfin]

I am not using Julia builds of GAP for my regular work, so no, that can't be it.

However, I cannot reproduce this anymore (with your branch). Not sure what went wrong...