> The inequality testing software is elaborate but not really very
> powerful.
Yes, the current code has a very nice lattice system, which allows you
to efficiently test v[a]<v[a] in the presence of a large number of
assertions of the form v[i]<v[j]. I doubt the efficiency is really
called for in the Maxima context....
As for calculation of constants, Barton Willis points out (bug 798571)
that even sign(1/sqrt(2)-sqrt(2)/2) is wrong (for that matter, it ought
to be simplified in the general simplifier, but that's another issue),
so yes, it would be worthwhile to fix that. Bfloat intervals seem like
the obvious way, but maybe there is a better way.
Of course, numeric calculation couldn't tell you that
sign(1/sqrt(2)-sqrt(2)/2) is ZERO, either. I suppose sign can never
return ZERO if you have to calculate approximately.
-s