Hi,
I've compiled Maxima (from CVS) with ABCL (from svn) and run_testsuite() shows just a few errors due to different floating point results. Hurray!
The key change to Maxima seems to have been changing EQ to EQUAL in a pattern-matching function ... In ABCL it appears that numbers are not necessarily EQ so a lot of patterns failed. Since the pattern-matching stuff is widely used in Maxima (especially integrals) there were a lot of failed tests. Just that one change seems to have fixed dozens of test failures!
I found it necessary to run the autoconf machinery (sh bootstrap && ./configure at least; I ran make as well, although I don't know if that's actually necessary, and then I ran configure.lisp as described in INSTALL.lisp) before compiling with ABCL via: (require 'asdf) (asdf:operate 'asdf:load-op :maxima) Without the autoconf stuff, src/autoconf-variables.lisp has incorrect values, it seems.
Maxima + ABCL is extremely slow -- something like 7 times slower than Maxima + Clisp, which is the slowest of the established Lisp implementations I've tried (GCL, ECL, SBCL, CMUCL, and Clisp). I'm working on Linux, maybe it helps to run on Windows -- I'm thinking the Java implementation might be faster. Just a guess.
I'm pretty excited about the possibilities here. A long-term project is to organize Maxima into modules in hopes of making it possible to load different modules a la carte -- I think that might make it attractive for bolting it onto different projects.
Thanks to the ABCL developers for all their work.
best
Robert Dodier Maxima developer