Ben Wing wrote:
i've gotten to the point in unicode-internal where it compiles,
builds
and dumps. then it proceeds to hang upon pdump load. when compiling
without kkcc and without mc-alloc, it gets stuck in loadup, loading
european.el. in this case it's not wedged so much as just extremely
slow due to the continual syntax-mirror updating. this is something
that's likely to be a classic case of premature optimization; the
mirror table is trying to save a few cycles in the retrieval of a
syntax code from a char table, but it takes a very long time to
compute, and had no smarts for handling lots of updates, one after the
other. i'm just planning on turning on the mirror table code
that should say "turning off".
and not worry about things, only worrying once the C-profiling
outputs
show that there really is a bottleneck here. (we really need a lot
more profiling. that's something any of you (stephen or otherwise)
could do if you have a chance. remember to profile with a complete,
full-on release build -- no error-checking, lots of optimization,
etc. make sure to do C profiling, not just Lisp profiling, since
that'll tell us where the real slowness is. actually, what would be
real cool is for one of you to do this and then *write down* *exactly*
the steps needed to reproduce the profiling session, in great detail
-- not knowing how to do it is what has tended to defer my attempts at
profiling. also, if you use any sort of framework to drive the
profiling and collect data over various performance sets, please
include that framework.
"framework" might be shell scripts, makefile, lisp files, etc. or
pointer to pre-existing framework of similar files.
ben