Martin Buchholz writes:
>>>>> "Soren" == Soren Dayton
<csdayton(a)cs.uchicago.edu> writes:
Soren> See above. Does this _really_ matter so much? The fact is that it _is_
Soren> all shared, and it _is_ demand paged, so this is sort of a possibly
Soren> shaky win. Would it _really_ be worth the work?
I'll add one more comment to this discussion. Although I don't
believe keeping the size of the CODE in the executable small should
really matter on a good OS, the size of the lisp data does matter a
lot. This is because every garbage collection touches ALL the lisp
data, and so ALL the lisp data is part of the minimal working set of
the XEmacs process.
Note that dumped LIsp object pages are read but not written during
GC. So they can be shared across XEmacs instances, if the dumper
puts those pages in the text segment of the binary. This happens
under BSD/OS, but apparently the switch to ELF-style binaries has
hosed this for everyone else.