Update to GAP.jl 0.11.3 resp. GAP 4.13.1

[fingolfin]

Draft as there is a failing test I had to disable for now. We already have a fix for that in GAP, but I still need to apply it to the GAP_lib_jll . We might also just wait for GAP 4.13.1 which is due soon. But I would like to get full CI tests run now to make sure there aren't other regressions hiding.

Resolves #2341

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 84.64%. Comparing base (224cb53) to head (0d25050).
Report is 14 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff            @@
##           master    #3688    +/-   ##
========================================
  Coverage   84.64%   84.64%            
========================================
  Files         612      614     +2     
  Lines       83240    83343   +103     
========================================
+ Hits        70460    70548    +88     
- Misses      12780    12795    +15     

Files with missing lines	Coverage Δ
src/Groups/GAPGroups.jl	94.13% <ø> (ø)
src/Groups/cosets.jl	89.59% <100.00%> (ø)
src/Groups/directproducts.jl	92.85% <ø> (ø)
src/Groups/homomorphisms.jl	90.76% <ø> (-1.35%)	⬇️

... and 18 files with indirect coverage changes

[fingolfin]

@ThomasBreuer overall this looks pretty good; we could cherry-pick the (simple) fix in GAP that fixes the describe test (which this PR currently commented out), to make a new GAP_lib_jll.

But in our discussion earlier today, it sounded as if you were aware of another issue with the new GAP / GAP.jl which is perhaps not reflected by the tests here (something about a book example) ?

[ThomasBreuer]

I meant this example:

julia> U = dihedral_group(8)
Pc group of order 8

julia> optimal_perm_rep(U)
Group homomorphism
  from pc group of order 8
  to permutation group of degree 4 and order 8
 
julia> isomorphism(PermGroup, U)
Group homomorphism
  from pc group of order 8
  to permutation group of degree 8 and order 8

where the idea was to show that optimal_perm_rep(U) yields something better than the general isomorphism(PermGroup, U) but apparently the new GAP's IsomorphismPermGroup is clever enough to compute the representation on 4 points.

[lgoettgens]

just FYI: the increased number of invalidations is reported to come from

Oscar.jl/src/Rings/localization_interface.jl

Lines 449 to 451 in 7d7bb9b

	function AbstractAlgebra.promote_rule(::Type{S}, ::Type{T}) where {RT, RET, MST, S<:AbsLocalizedRingElem{RT, RET, MST}, T<:RingElement}
	AbstractAlgebra.promote_rule(RET, T) == RET ? S : Union{}
	end

[fingolfin]

sigh two different crashes...

GC error (probable corruption)
Allocations: 592882531 (Pool: 592501723; Big: 380808); GC: 131

!!! ERROR in jl_ -- ABORTING !!!

thread 0 ptr queue:
~~~~~~~~~~ ptr queue top ~~~~~~~~~~
GAP.SmallBag()
==========
GAP.SmallBag()
==========
GAP.GapObj()
==========
...

and also

character fields of Brauer characters |    4      4  0.1s

[3691] signal 11 (1): Segmentation fault
in expression starting at /home/runner/work/Oscar.jl/Oscar.jl/test/Groups/group_characters.jl:1111
MarkBag at /workspace/srcdir/gap-4.13.1/src/julia_gc.c:1070 [inlined]
MarkArrayOfBags at /workspace/srcdir/gap-4.13.1/src/bags.inc:21
BagMarkFunc at /workspace/srcdir/gap-4.13.1/src/julia_gc.c:710
gc_mark_outrefs at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:2952 [inlined]
gc_mark_loop_serial_ at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:2968
gc_mark_loop_serial at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:2991
gc_mark_loop at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:3168 [inlined]
_jl_gc_collect at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:3557
ijl_gc_collect at /cache/build/tester-amdci4-9/julialang/julia-master/src/gc.c:3936

[ThomasBreuer]

Once the upgrade of Oscar to GAP 4.13 is done, we can adjust those parts of the Oscar code that become nicer with GAP 4.13 (at least those with a comment line involving gap-system); and issue #2341 can be closed.

[ThomasBreuer]

This example was originally intended to demonstrate that isomorphism is in general not good enough to get such a small degree for the image. Now this part of the book chapter becomes less suggestive.
(Since the final proofs of the Oscar book are still not available, do we have a chance to replace this example in the book by a better one? Of course the next release of GAP may then destroy the effect again ...)

[fingolfin]

We are getting this error now, any idea what it might be about and how to fix it, @ThomasBreuer

G-sets by right transversals: Error During Test at /home/runner/work/Oscar.jl/Oscar.jl/test/Groups/gsets.jl:328
  Test threw exception
  Expression: Omega[end] == Omega[length(Omega)]
  StackOverflowError:
G-sets by right transversals: Error During Test at /home/runner/work/Oscar.jl/Oscar.jl/test/Groups/gsets.jl:338
  Test threw exception
  Expression: findfirst(is_equal(Omega[i]), Omega) == i
  StackOverflowError:
G-sets by right transversals: Error During Test at /home/runner/work/Oscar.jl/Oscar.jl/test/Groups/gsets.jl:338
  Test threw exception
  Expression: findfirst(is_equal(Omega[i]), Omega) == i
  StackOverflowError:

[ThomasBreuer]

Yes, this is one instance of the problems with julia_to_gap we want to get rid of. I will provide a preliminary fix in Oscar.

Edit: cherry-picked into this branch

[lgoettgens]

CI complains that recog is not available. After digging into it for some time with @ThomasBreuer, we noticed that this is due to no version of io being available (which is a transitive dependency of recog). Just trying to GAP.Packages.load("io", "4.8.2", quiet=false) fails because of oscar-system/GAP.jl#1037.

[lgoettgens]

With GAP.jl v0.11.3 this now look good 🎉🎉
(apart from the macOS-RPTU server not picking up any jobs)

[fingolfin]

We'll restart the RPTU CI runners soon, but in the meantime I've run the test on my mac and they were fine.