-----maxima-bounces at math.utexas.edu wrote: -----
>I believe Barton has a simplus simplifier and he's run into similar
>issues.
Yes, I've played with an alternative simplus that tries to be infinity
correct (inf - inf --> und, inf + 42 --> inf, and ...). It also uses
sorting to speed up the sum of n distinct summands (O(n log(n) vs
O(n^2)). And presumably, the code is a bit easier to modify than the
old. For example, I could append interval addition (non-tellsimp) or
array - array addition to my code without much trouble, I think.
The last I tried, this simplus makes it all the way through the test
suite except for a bug or two (that I think are somewhere else; for
example 0.0 + [0] --> [0] but my simplus does 0.0 + [0] --> [0.0], I
think.)
For the most part, I did this for fun and practice; I'm undecided if
it's a serious proposal -- it is a 100% rewrite. It may have bugs that
may take a decade to uncover.
Currently, the limit code does some inf - inf --> 0 miscalculations
but manages to get correct values. I think I sent this list a note
about that. I'm guessing that the real trouble would be making
simptimes and simpexpt infinity correct. Both simptimes and simepext
are much messier than simplus, by the way. Expand distributes its work
over simptimes and simpexpt, (too) many switches control branch cut
rules, and ...
Barton