474801 - Checking for MaybeGC conditions when allocating GC things in JS shell

Status: RESOLVED FIXED

(Core :: JavaScript Engine, enhancement)

Description

[Andrei Saprykin]

This is part 1 of the original bug 453432. It adds "MaybeGC" conditions check for allocation of GC thins, but it does not remove the MaybeGC functionality for DOM objects.

Comment 1

[Andrei Saprykin]

Attachment: Initial patch

Comment 2

[Igor Bukanov]

landed to tm - http://hg.mozilla.org/tracemonkey/rev/6657640cbbb2

Comment 3

[Andrei Saprykin]

Attachment: Updated patch

Comment 4

[Igor Bukanov]

(In reply to comment #3)
> Created an attachment (id=358409) [details]
> Updated patch

As discussed with Andrei a possible reason for tinderbox failures with the previous patch was subtle changes in the frequency of cycle collector runs. The patches replaces the explicit tracking of the number JS_GC calls with using JSRuntime.gcNumber exposed through a new API. But that means that JS_GC calls done outside nsJSEnvironment, like JS_GC done from the cycle collector itself, or JS_GC called when destroying JSContext, would also contribute to the counter.

Effectively it may make nsJSRuntime::sGCCount to run faster leading to more frequent cycle collections. It principle it may explain the crash test timeout. It may also explain the mochitest leak as the cycle collector may not run at the proper moment that test expects.

Now, the new version tries to offset that faster nsJSRuntime::sGCCount rate with ignoring at least the GC runs done during the cycle collection. It could be enough to offset that rate effect so I would try to land the updated patch again.

Comment 5

[Andrei Saprykin]

Attachment: Updated patch

Comment 6

[Andrei Saprykin]

Attachment: Updated patch

Previous patch was made in mozilla-central instead of tracemonkey

Comment 7

[Igor Bukanov]

Attachment: patch without nsJSEnvironment.cpp changes

Given the instability of tinderbox tests I prefer to play safe here. So I would try to land the patch without nsJSEnvironment changes related to the sGCCounter.

Comment 8

[Igor Bukanov]

landed to TM - http://hg.mozilla.org/tracemonkey/rev/799649d4e416

Comment 9

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/799649d4e416

Comment 10

[Robert Sayre]

http://hg.mozilla.org/releases/mozilla-1.9.1/rev/4da0fb4dfc49