We still need to be able to mark a generic Cell* in service of the JITs, and we have enough information to do this easily. Marking void* and making the caller deal with figuring out what kind of thing to mark is a pretty terrible interface though. This is, surprisingly, line length neutral, despite the new infrastructure for shared upcasts not being used extensively yet.

Comment on attachment 8589230 [details] [diff] [review] 4.10_remove_mark_internal-v0.diff Review of attachment 8589230 [details] [diff] [review]: ----------------------------------------------------------------- This looks good. I had a go at doing without the functor adaptor and came up with a slightly different approach. I'll post the patch and see what you think.

This makes CallTyped use variadic template arguments to forward everything to the functor so we don't need to put these into a TracerAdaptorFunctor first.

Comment on attachment 8589605 [details] [diff] [review] call-typed-update Review of attachment 8589605 [details] [diff] [review]: ----------------------------------------------------------------- Oooo, template var args! I should have thought of that. This looks much better.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=8ab9624f3c38 https://hg.mozilla.org/integration/mozilla-inbound/rev/dd24fea25988

I landed a trivial mingw fixup (check for compiler instead of assuming that XP_WIN means MSVC): https://hg.mozilla.org/integration/mozilla-inbound/rev/598ccd9bc144