Remove GAP stones

[fingolfin]

This implements issue #1060.

However, it is not quite ready yet, because I did nothing to adapt testtravis, which previously run a subset of testinstall and teststandard, selecting only those .tst files whoe GAPstones were below a certain threshold.

I think we should change testtravis to not run testinstall (as we run that in a separate travis task already anyway). Next, we should identify which teststandard tasks take particularly long, and then either make them faster, or else consider moving them into a yet another category, say "testslow" or "testextensive" or so.

[codecov-io]

Current coverage is 56.85% (diff: 90.18%)

Merging #1072 into master will increase coverage by 2.31%

@@             master      #1072   diff @@
==========================================
  Files           432        433      +1   
  Lines        224765     225073    +308   
  Methods        3431       3431           
  Messages          0          0           
  Branches          0          0           
==========================================
+ Hits         122572     127955   +5383   
+ Misses       102193      97118   -5075   
  Partials          0          0           

Diff Coverage	File Path
0%	lib/profile.g
••••••••• 90%	new tst/teststandard/arithlst.g
•••••••••• 100%	lib/test.gi

Powered by Codecov. Last update 75245e7...e70c579

[fingolfin]

OK, I now added a bunch of tweaks that go beyond removing stones, but which try to get a better working "testravis" target. Let's see how that works out.

For this, I moved some of the slowest .tst files to a new directory testextra, which is not yet hooked by anything. I could of course add a tst/testextra.g file to run them, but I thought I'd first see whether (a) this works as I hope it will work, and (b) whether anybody hates the idea...

BTW, I think the reason Travis kills our slow tests is not that they take long, but rather that there is no output for more than 10 minutes. So if we split them into multiple quicker files, or enabled the showProgress option, we likely could run them anyway. The latter would be a bit annoying as we'd get lots of unhelpful output, though.

[ChrisJefferson]

I am strongly in favour of splitting up test files into smaller pieces.

Just an advanced warning. I tried this with bugfix.tst and found it broke other tests, as pieces were run earlier, and messed up bits of gap state. Could move bugfix.tst also to its own directory so it can be split up of course.

[fingolfin]

Awesome, those test failures are due to a genuine bug :-)

[fingolfin]

Rebased version, now builds on PR #1100. Also removed the "do not merge" label -- from my point of view, this is ready.

[olexandr-konovalov]

This is thorough walk through everything related to GAP4Stones. It also contains other changes, most notably it deletes some old and not used any more .g files from tst, and moves some text files to the new testextra directory - one should hook them to a test or create a new issue to remember. As an intermediate step, TestDirectory accepts a list of directories too, so one could add testextra to teststandard to have previous behaviour.

Anticipating various surprises coming from scripts relying on old format of output (e.g. to extract runtime of some individual tests) and now having to deal with

testing: /Users/travis/build/gap-system/gap/tst/testinstall/alghom.tst
     238 msec for alghom.tst
testing: /Users/travis/build/gap-system/gap/tst/testinstall/algmat.tst
    3706 msec for algmat.tst
...
testing: /Users/travis/build/gap-system/gap/tst/testinstall/zmodnze.tst
     226 msec for zmodnze.tst
-----------------------------------
total  128973 msec

instead of

testing: /home/gap-jenkins/workspace/GAP-master-test/GAPCOPTS/32build/GAPTARGE\
T/install/label/graupius/GAP-master-snapshot/tst/testinstall/grppcnrm.tst
grppcnrm.tst           10853          12367
-------------------------------------------
total                      0         766649

but that's inevitable and should be done at some point. I should be able to deal with that after the merge.

It may be worth mentioning briefly in the changes overview (https://github.com/gap-system/gap/wiki/Changes-between-GAP-4.8-and-GAP-4.9) that we had revised tests and reorganised test directories.

[olexandr-konovalov]

Any special reasons to remove this?

[fingolfin]

It took several seconds (5-10?) to complete, so does not belong into testinstall.

I contemplated moving it to a new test file in teststandard, but ultimately decided that this is a weird random test, for whch it is unclear what exactly it is meant to exercise. Actually, the whole xgap.tst somewhat falls into that category. From the original commit by Max N., who added it 1999-05-30: "Added tests for fp groups for problems found with XGAP. Max."

So ultimately, this file should go, its contents (or equivalent content) distributed over more appropriate files (e.g. fpgrps.tst or files in opers).

[fingolfin]

@alex-konovalov As to adapting scripts which parse the output of TestDirectory: That sounds quite brittle. Were are those scripts? What are they used for? Perhaps we can find a better solution than parsing output meant for humans, which will always be brittle.

[fingolfin]

@alex-konovalov Also not PR #1100, which is a subset of this PR. We could get that in first (with suitable changes), as that should not break any of your scripts.