1159810 - Kill full gc timer when gc starts

Status: RESOLVED WONTFIX

(Core :: DOM: Core & HTML, defect)

Description

[Olli Pettay [:smaug][[e-mail]]]

Attachment: v1

We have similar Kill*Timer calls in nsJSContext::GarbageCollectNow.

Comment 1

[Olli Pettay [:smaug][[e-mail]]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=cb0f9c7da004

Comment 2

[Andrew McCreight [:mccr8]]

Comment on attachment 8599403 [details] [diff] [review]
v1

Review of attachment 8599403 [details] [diff] [review]:
-----------------------------------------------------------------

Oh, PreciseGCRunnable::Run() starts a GC by directly calling JSAPIs.  Fun.

Comment 3

[Ting-Yu Chou [:ting] (away)]

I guess we want this on FxOS v2.2 as well?

Comment 4

[Olli Pettay [:smaug][[e-mail]]]

well, if this helps with the testing. In practice this shouldn't affect actual behavior too much.

Comment 5

[Olli Pettay [:smaug][[e-mail]]]

tryserver is a bit orange. Most of the failures are known, but not all :/

Comment 6

[Olli Pettay [:smaug][[e-mail]]]

Another tryserver push https://treeherder.mozilla.org/#/jobs?repo=try&revision=5a6d2236762c

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

I'll land this to m-i.  For some reason also that previous try push had quite a few known failures.
Hopefully this isn't making try more flaky (though that would mean we have some bugs in the tests).

Comment 8

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/256f10c0dbdc

Comment 9

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

This apparently made reftest-unaccelerated tests on at least Windows 7 start failing, so I backed it out in https://hg.mozilla.org/integration/mozilla-inbound/rev/d1c2ecd59fc0

https://treeherder.mozilla.org/logviewer.html#?job_id=9558026&repo=mozilla-inbound

Comment 10

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

Pretty sure it's also responsible for android reftest-8 failures like https://treeherder.mozilla.org/logviewer.html#?job_id=9556030&repo=mozilla-inbound

Comment 11

[Phil Ringnalda (:philor)]

Not just Android debug reftest-8, final tally looks more like 1, 2, 3, 4, 7, 8 which probably means "any of them if you happen to hit them right." Impossible to say whether you had any similar effect on Android debug mochitests, since you landed on top of a patch which made them fail to start and infinitely retry, so it'd be a good idea to include them in the try push where you check whether you've stopped randomly killing Android debug reftests.

Comment 12

[Olli Pettay [:smaug][[e-mail]]]

Sorry about that.

Looks like fixing this isn't really worth.

For https://bugzilla.mozilla.org/show_bug.cgi?id=1099195#c2027 we should just make
PreciseGCRunnable to use the normal GC triggering methods.