A missing error in Variety for non-supported base rings has been added.
Procedure SetGBGlobalModular has been added to set whether the global modular algorithm should be used when computing a Groebner basis over the rational field or a number field (while function GetGBGlobalModular gives the current value).
A bug with the SEA point-counting algorithm over finite fields defined by relative extensions was fixed.
Extra printing in Galois group computations has been cleaned up.
A precision check in the computation of Galois groups of reducible polynomials over Q(t) has been fixed.
The automorphism group of a graph is now properly recomputed after labels are changed. This bug only affected V2.23-1.
Traces is now the default algorithm for operations involving the canonical form or equivalency testing.
The Hadamard database routines now explicitly set the canonical form algorithm to nauty for compatibility; this will be changed in a future patch release.
The HypergeometricTraceK intrinsic, applicable for non-Galois data and/or parameter specializations at non-rationals, has been improved and documented. Instigated by J. Voight.
The base algorithm for matrix echelonisation has been greatly improved in general for matrices of small to medium size with entries in GF(p) for p=2,3,5,7. For example, matrix inversion over GF(2) in dimension up to 1000 is roughly 3 to 4 times faster on a typical Intel Core CPU.
A crash in multivariate resultant computation with a very large number of variables has been fixed.
Using a very long undefined identifier no longer crashes.
A bug with coercion of rational zeros into a p-adic field has been fixed. Such an p-adic number could be created, but would not be correctly be recognized as zero in comparisons.
The PowerRelation intrinsic now ensures its output is monic, and there is some attempt to remove superfluous factors in it too.