Akihiro Munemasa noticed a bug (politely denoted by him only as 'strange
behaviour') in the 'CharTable' routine.
It will arise, if there is no 'freedom' for the degrees of the non
linear characters, i.e. if the character degrees are determined exactly
by the number of classes and the number of linear characters.
In this case, due to a forgotten '=', the largest character
degrees are accidently thrown away, leading to problems.
Fixing this bug fortunately is very easy (though I hope, this fix will also
be included in the next patch): In the library file grpctbl.g change
line 272 (which also is identified uniquely by its contents) from
D.degreePool:=Filtered(D.degreePool,i->i>1 and i<z/2);
D.degreePool:=Filtered(D.degreePool,i->i>1 and i<=z/2);
GAP successfully computed the character table of all solvable
nonabelian groups of order at most 100, except those of order 27.
With this fix, GAP will compute the character tables of all these
groups (and also the non solvable ones).
Side remark: 27 is a prime power. For groups of prime power order,
CharTablePGroup normally works somehow faster than CharTable. As we were
planning to use CharTablePGroup in these cases instead of the standard
Dixon/Schneider algorithm, I never checked the two nonabelian groups of
size 27. I better had. Please excuse this stupid bug.