515287 - Don't trigger GC on every XHR completion

Status: RESOLVED FIXED

(Core :: DOM: Core & HTML, defect, P3)

Description

[Boris Zbarsky [:bzbarsky]]

We trigger a GC after every XHR completion.  We should consider not doing that.  Either only trigger it for completion of parts of a multipart request, or only trigger it every N XHR completions or something...

See bug

Comment 1

[Robert O'Callahan (:roc) (email my personal email if necessary)]

See bug 515215.

Comment 2

[Boris Zbarsky [:bzbarsky]]

er, yes.  ;)  I managed to paste it into the "blocks" field!

Comment 3

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

I think we can remove the whole GC call, since we call MaybeCC(PR_FALSE), 
and that call tries to check if there is lots possible new garbage to collect before calling CC, and CC isn't usually called if user is active.
But I'll do some testing.

Comment 4

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

Attachment: remove the explicit GC call

In my testing removing this doesn't cause any harm, though doesn't speed
up things too much either.
One problem is that nsJSContext::ScriptEvaluated is called very often, and that
calls MaybeGC, which seems to trigger GC pretty often.
But the nsJSContext::ScriptEvaluated handling is a different bug.

Comment 5

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

FYI, the GC call was added in Bug 237319.
And I did test multipart XHR.

Comment 6

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

Based on testing in a debug build (so that I get ifdef DEBUG_smaug log about
when to run CC), the patch does reduce calling CC, but yet when there are enough
new suspected object, it is called.
With the patch and 10 gmail tabs open CC is explicitly called about every 35s
when the browser is in background.

Comment 7

[Boris Zbarsky [:bzbarsky]]

How often does CC get called without the patch in that situation?  I bet more often!

Comment 8

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

Without the patch I get 
"Running CC was delayed because of NS_MIN_CC_INTERVAL" message quite often.
That means that someone tried to call CC more often than every 10s.

Comment 9

[Johnny Stenback (:jst)]

This is borderline blocker material IMO, but it'd be really nice to fix this...

Comment 10

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

http://hg.mozilla.org/mozilla-central/rev/c6bcf4347492

Comment 11

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

http://hg.mozilla.org/releases/mozilla-1.9.2/rev/7f3807ea3e95