I fixed a bug in some analysis scripts and discovered some more cases where we use autostrings on the heap. Patches forthcoming, separated by module.

Pushed to mozilla-central: http://hg.mozilla.org/mozilla-central/index.cgi/rev/0d58f04be0dd http://hg.mozilla.org/mozilla-central/index.cgi/rev/bc1896fdf73a

Comment on attachment 329887 [details] [diff] [review] layout heap autostrings, rev. 1 I think this is going to cause a substantial increase in allocations, and I don't see why it's bad to use nsAutoString in cases like this.

Comment on attachment 329887 [details] [diff] [review] layout heap autostrings, rev. 1 (And if you request review again, please include startup performance test results showing no regression.)

ok, this marks this particular heap autostring as ok, while keeping the general rule. Looking at the usage of mIdent in nsCSSParser, I'm surprised that an autostring helps, but I can't figure out how to get precise stats on it and try-talos wasn't handing out consistent numbers, so I'll just go with the annotated override.

Pushed the PSM patch to mozilla-central, http://hg.mozilla.org/mozilla-central/index.cgi/rev/13ed1c63513b

Comment on attachment 334296 [details] [diff] [review] Mark CSSToken::mIdent as ok-on-heap, rev. 1 r=dbaron (I'm probably not a valid reviewer for stack.js, but I'm not sure if it requires review)

Layout patch pushed: http://hg.mozilla.org/mozilla-central/index.cgi/rev/6437cdc9973e It turns out there was an analysis error which, once fixed, has found additional situations, but I'm going to spin those in separate bugs to keep this one sane.