From: Christophe R. <cs...@ca...> - 2004-01-29 18:49:55
|
Daniel Barlow <da...@te...> writes: > Given that there's been no release this month I'm not sure if we're in > freeze or slush or full-on patch-u-like, but in any case I don't have > the connectivity necessary to fix this if it turns out broken. My assumption is that we're in patch-u-like mode, and... > 1) The year-old allocation region bug for which I removed the > workaround in 0.8.7.3, and caused much corrupting weirdness throughout > January until Christophe reinstated the workaround. The relevant > changes for this are in gencgc.c: the actual fix is the test for > `allocation & BOXED_PAGE' instead of `allocation == BOXED_PAGE'. When > an open region is sharing a page with other data, newspace scavenging > should not skip over that other data. > I'm reasonably certain this fix is correct. ... on this basis I've merged this one in 0.8.7.21. I looked at it, and your diagnosis makes sense, and explains symptoms, so as a good empirical scientist I think it's reasonable. Of course, users should beware of running this on critical systems in the next few days, given what's already happened this month; on the other hand, if you have systems which aren't critical and a moment spare to test things, that would be good too. Cheers, Christophe -- http://www-jcsu.jesus.cam.ac.uk/~csr21/ +44 1223 510 299/+44 7729 383 757 (set-pprint-dispatch 'number (lambda (s o) (declare (special b)) (format s b))) (defvar b "~&Just another Lisp hacker~%") (pprint #36rJesusCollegeCambridge) |