hash-table changes

Gary Byers <[email protected]> Mon, 5 Sep 2005 00:10:01 -0600 (MDT)
Newsgroups gmane.lisp.openmcl.bugs
Message-ID <[email protected]>
I have some changes that're intended to improve hash-table performance in
at least some cases.  So far, I've only tested them on a 32-bit Darwin
platform, and I'd like to check them in (both so that I don't lose them
and so that I can try them out on 64-bit and Linux platforms.

It's possible that these changes will break something.  I did tag the
code that was in CVS a little while ago, to try to make it easier to
back out of this if that happens and if recovery looks difficult.

The changes involve the kernel (because of the way that the GC
interacts with hash tables) as well as lisp code, and once they're
bootstrapped new images will depend on a new kernel (and vice versa.)

Moral: this'll be somewhat disruptive, and might turn out to be a bad
idea.  (I think that improving hashing performance is generally a good
idea; the bad idea might be trying to change things before a release.)

The changes are likely to be a little difficult to bootstrap; I'll
upload new binaries if/when I get things working, and will try to
back out of things if that looks to be difficult.