> Different math libraries return different error strings. There really
> is no way to patch it.
Hmm, if that is the problem, can't we build a little program that
just produces all the error strings and stuffs them somewhere the
regression test can find them?
Something like
float a = MAXFLOAT + 1;
fprintf(somewhere, "OVERFLOW = %s\n", errstr[errno]);
and the regression test pick them up and patch their expected files?
mjl
------------------------------