Recursion depth trap when repeatedly calling Test

[james-d-mitchell]

Observed behaviour

If the contents of the test file bug.tst is:

gap> START_TEST("bug.tst");
gap> IsAbelian("A");
Error, no method found! For debugging hints type ?Recovery from NoMethodFound
Error, no 1st choice method found for `IsCommutative' on 1 arguments
gap> STOP_TEST("bug.tst");

and I load GAP without any packages gap -A, then do

    i := 0;; repeat i := i + 1; until not Test("bug.tst");

then when i is 1665

########> Diff in bug.tst:2
# Input is:
IsAbelian("A");
# Expected output:
Error, no method found! For debugging hints type ?Recovery from NoMethodFound
Error, no 1st choice method found for `IsCommutative' on 1 arguments
# But found:
Error, no method found! For debugging hints type ?Recovery from NoMethodFound
Error, recursion depth trap (5000)

This is a minimal example of the problem, but this is causing some of the Semigroups package continuous integration test to fail, see the discussion here for more details. This bug is confirm by several others, not only me, so unlikely to be an issue with my setup.

Expected behaviour

The loop to run forever.

Copy and paste GAP banner (to tell us about your setup)

 ┌───────┐   GAP 4.8.7-4353-ga179f09 of today
 │  GAP  │   https://www.gap-system.org
 └───────┘   Architecture: x86_64-apple-darwin16.6.0-clang-default64
 Configuration:  gmp 6.1.1
 Loading the library and packages ...
 Components: trans 1.0, prim 3.0, small* 1.0, id* 1.0
 Packages:   GAPDoc 1.5.1
 Try '??help' for help. See also '?copyright', '?cite' and '?authors'
gap>

[ChrisJefferson]

I suspect the problem is we don't reset the depth of the stack when we longjmp.

[james-d-mitchell]

It seems to be a bit unpredictable, whether it happens or not.

[ChrisJefferson]

It will be (I suspect) connected to the number of tests which exit via an error.

[wilfwilson]

Yes. In our continuous integration, it was always one of the last files to be tested that was having problems.

[ChrisJefferson]

Ah, just checking, is this only on 4.8? I have a feeling I fixed something related to this in master (and the error doesn't seem to trigger in master for me)

[wilfwilson]

We only test with master at the moment, so I don't know anything about whether it's present in stable-4.8.

[ChrisJefferson]

Ah, no, I can reproduce it, it just takes a while.

If I add a way of printing out GAP's RecursionDepth, then I can see it increases by 3 each time we run the test, so eventually we overflow.

[ChrisJefferson]

Fix coming up.

[james-d-mitchell]

Awesome @ChrisJefferson, thanks

[ChrisJefferson]

This is (hopefully) fixed by #1753.

[markuspf]

Fixed by #1753