OG> I'd like to see --use-minimal-tagbits being default, and the other OG> alternative actually removed. OG> Rationale: OG> - per Kyle's study, performance is equivalent OG> - we have to settle on one of the options for code-freezing time in OG> order to ensure proper testing on all architectures I thought we already have settled on doing things the way they've been for years, i.e. default to gung-ho=no for 21.0. OG> - not munging pointers is infinitely better because: OG> * it removes the need for DATA_SEG_BITS which, basically, sucks OG> * it makes purify-like memory leak detection possible This is currently possible by simply running purify and friends on temacs. Here are some targets in src/Makefile.in.in: puremacs: $(temacs_deps) $(PURIFY_PROG) $(PURIFY_FLAGS) $(LD) $(temacs_link_args) -lpthread run-puremacs: puremacs ./puremacs -batch -l ${srcdir}/../lisp/loadup.el run-temacs Also, VERY few folks will want to go to the trouble of purifying XEmacs. Those that do can use --gung-ho. In a 64-bit world, we might want to EXPAND the tagbits and keep enough type bits to represent all the lisp types. This would allow say, 1 mark bit 7 type bits 56 val bits which would use less memory. Knowing which bits to use is going to be a DATA_SEG_BITS portability problem though. Martin

After thinking about it, I still say leave it the way it is, i.e. gung stuff off. Not enough added value at present. We have enough change in this release already.