1380025 - GC 'poke' trigger doesn't have its own reason code

Status: RESOLVED FIXED

(Core :: JavaScript: GC, enhancement)

Description

[Jon Coppeard (:jonco)]

As mentioned in bug 1377466, we don't know how often pokes are causing GCs because they don't have their own reason code.

Comment 1

[Jon Coppeard (:jonco)]

Attachment: bug1380025-poke-gc

This patch does a few things:
 - 'poke' is renamed to 'notifyRootsRemoved' to better explain its purpose
 - GCs thus triggered get their own reason code
 - the logic in GCRuntime::collect is updated to ensure we wait until the end of GC to trigger a new one

Comment 2

[Steve Fink [:sfink] [:s:]]

Comment on attachment 8885327 [details] [diff] [review]
bug1380025-poke-gc

Review of attachment 8885327 [details] [diff] [review]:
-----------------------------------------------------------------

Wow, I had no idea what any of this did. Thanks for clearing it up!

::: js/src/jsgc.cpp
@@ +6841,5 @@
> +                /* Need to re-schedule all zones for GC. */
> +                JS::PrepareForFullGC(rt->activeContextFromOwnThread());
> +                repeat = true;
> +                reason = JS::gcreason::ROOTS_REMOVED;
> +            } else if (shouldRepeatForDeadZone(reason) && !wasReset) {

The !wasReset here is redundant. That case was already handled.

::: js/src/vm/NativeObject.cpp
@@ -2864,5 @@
>          // as the result parameter. This always succeeds when there is no hook.
>          return CallJSDeletePropertyOp(cx, obj->getClass()->getDelProperty(), obj, id, result);
>      }
>  
> -    cx->runtime()->gc.poke();

Why is this one no longer necessary?

Comment 3

[Jon Coppeard (:jonco)]

(In reply to Steve Fink [:sfink] [:s:] from comment #2)
> Why is this one no longer necessary?

I found a description of what GC poke was originally for in bug 481922 comment 11.

The second case is still valid, but any check we had to skip a GC if we hadn't been poked must have been removed a long time ago.  This poke was for the first case.

Comment 4

[Jon Coppeard (:jonco)]

(In reply to Steve Fink [:sfink] [:s:] from comment #2)
> The !wasReset here is redundant. That case was already handled.

Cheers, fixed.

Comment 5

[Jon Coppeard (:jonco)]

I accidentally landed this patch with bug 1377466 in the commit message.

Comment 6

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

I'm seeing a spike in test failures that start with `Assertion failure: false (Ran out of memory while building cycle collector graph), at nsCycleCollector.cpp:930` in the last few days (bug 1381177). Any chance I can pin those on this patch?

Comment 7

[Steve Fink [:sfink] [:s:]]

These landings were

http://hg.mozilla.org/integration/mozilla-inbound/rev/ed043698eafa
http://hg.mozilla.org/mozilla-central/rev/ed043698eafa