CI broken due to recent merges plus changes on GAP master and/or package updates

[fingolfin]

CI first started failing when PR #402 by @TWiedemann was merged but only in the GAP devel job (so I suspect the root cause is not in that PR, but just exposed by it; instead I suspect a change in GAp master and/or one of the recent package updates is responsible). The error:

########> Diff in /tmp/gaproot/pkg/recog/tst/working/slow/verification.tst:37
# Input is:
for G in matGroupList do
 testGroup(G);
od;
# Expected output:
# But found:
Error, ConvertToMatrixRep( <mat>, <q> ): not all entries of <mat> wr\
itten over <q>
########

Then when PR #467 was merged (which only happened after all its CI jobs passed!) all CI jobs started to fail. This looks like a bad interaction with PR #402 -- e.g. there may have been a hidden bug in PR #467 that was exposed; or some of the changes in PR #402 had unexpected and undesirable side effects. Dunno.

Afterwards also PR #469 was merged (it was on automerge, so it passed all its CI in isolation). For now I'll assume this PR is otherwise innocent, we'll see.

[fingolfin]

OK, the error involving ConvertToMatrixRep is due to GAP PR gap-system/gap#6232 by @ThomasBreuer. We can observe its cause in this code snippet from GAP, modified in the mentioned PR:

InstallMethod( PreImagesRepresentative,"IsLinearActionHomomorphism",
  FamRangeEqFamElm, [ IsLinearActionHomomorphism, IsPerm ], 0,
function( hom, elm )
  local   V, G, Grep, xset,lab,f;

  # is this method applicable? Test whether the domain contains a vector
  # space basis (respectively just get this basis).
  lab:=LinearActionBasis(hom);
  if lab=fail then
    TryNextMethod();
  fi;

  # PreImagesRepresentative does not test membership
  #if not elm in Image( hom )  then return fail; fi;
  xset:=UnderlyingExternalSet(hom);
  V := HomeEnumerator(xset);
  G:= Source( hom );
  Grep:= Representative( G );
  f:=DefaultFieldOfMatrixGroup(G);

  if not IsBound(hom!.linActBasisPositions) then
    hom!.linActBasisPositions:=List(lab,i->PositionCanonical(V,i));
  fi;
  if not IsBound(hom!.linActInverse) then
    lab:=ImmutableMatrix(f, Matrix( lab, Grep ));
    hom!.linActInverse:=Inverse(lab);
  fi;

  elm:=OnTuples(hom!.linActBasisPositions,elm); # image points
  elm:=V{elm}; # the corresponding vectors
  elm:=ImmutableMatrix(f, Matrix( elm, Grep ));

  return hom!.linActInverse*elm;
end );

Note the Matrix( elm, Grep ) call -- there, Grep is a compressed matrix over GF(3), while elm is a plain lis containing a single element: a compressed and immutable vector over GF(9). The Matrix invocation ends up invoking ConvertToMatrixRep(elm, 3), which fails, because elm contains an immutable compressed vector over the wrong field (if it was mutable, it would succeed, because that vector is really over GF(3)).

The vector comes from V := HomeEnumerator(xset);. We find:

brk> V;
[ [ 0*Z(3) ], [ Z(3)^0 ], [ Z(3) ] ]
brk> List(V, Is8BitVectorRep);
[ true, true, true ]
brk> List(V, Q_VEC8BIT);
[ 9, 9, 9 ]

So this is vectors over GF(9) which all could be over GF(3). So this looks fishy; but I suspect it may be an older issue and we just never noticed? Alas, I did not investigate further and e.g. do not know where and how exactly V := HomeEnumerator(xset); is created.

[fingolfin]

For now, I have locally reverted that GAP PR, so that I can debug the second issue.

[fingolfin]

Workaround is in gap-system/gap#6395

Discussion on the GAP side: gap-system/gap#6389

[fingolfin]

Fixed again thanks to Thomas' gap-system/gap#6396