Olli said this probably isn't needed.
Created attachment 8543325 [details] [diff] [review] Don't PokeGC in SetNewDocument. This was smaug's idea, so I probably shouldn't have him review. ;) try run: https://treeherder.mozilla.org/#/jobs?repo=try&revision=5d619be17e91
Comment on attachment 8543325 [details] [diff] [review] Don't PokeGC in SetNewDocument. Looks like the GC there is super old code, http://bonsai.mozilla.org/cvsblame.cgi?file=mozilla/dom/src/base/nsGlobalWindow.cpp&rev=1.1&root=/cvsroot Search for ::SetNewDocument (blame/marking doesn't seem to be able to deal with that file properly)
Comment on attachment 8543325 [details] [diff] [review] Don't PokeGC in SetNewDocument. Well, you could then review your own patch if the idea was mine ;)
Yeah, I guess. When I was testing this it looked like any time we were doing a poke for SetNewDocument, we were getting a poke for a bunch of other things so it should be okay in any event.
Looks like this wasn't the cause of the orange (though it certainly could have been): https://treeherder.mozilla.org/ui/#/jobs?repo=try&revision=c40fb1a889b9
Backed out for apparently making Android reftests significantly more failure-prone (generally dying mid-run with no usable error). https://hg.mozilla.org/integration/mozilla-inbound/rev/58767ba4867e https://treeherder.mozilla.org/logviewer.html#?job_id=5172849&repo=mozilla-inbound https://treeherder.mozilla.org/logviewer.html#?job_id=5172775&repo=mozilla-inbound https://treeherder.mozilla.org/logviewer.html#?job_id=5170636&repo=mozilla-inbound etc
I'm just going to close this for now. There's probably not a lot of value in tiny GC heuristic adjustments like this.