1157467 - test_request_context.html leaks when run with e10s

Status: RESOLVED FIXED

(Core :: DOM: Workers, defect)

Description

[Andrew McCreight [:mccr8]]

This isn't visible right now because the content process leak threshold is fairly high (20kb) due to some GMP issues that have since been fixed.  If you can't fix this quickly, I'd like to disable this test for e10s.

Comment 1

[Andrew McCreight [:mccr8]]

There are still e10s leaks with this test disabled, just about 5kb of them.

Comment 2

[(no longer active)]

Please don't disable the test.  Instead, please tell me something about the leak that would help me diagnose it?  :-)

Comment 3

[Andrew McCreight [:mccr8]]

To reproduce the leak, disable test_fetch_event.html (which also leaks) then run this:
 ./mach mochitest-plain --e10s dom/workers/test/serviceworkers/

Comment 4

[Andrew McCreight [:mccr8]]

Attachment: information about leaked objects

This file contains information about the objects that are leaked.  Note the top set of leaked objects are things we always leak, so you can ignore those in the second list.  I also included the leaked URLs and some weird assertions I see.

The most interesting thing in the list of leaked objects is probably the doc shell, though there are a few networking and plugin related things in there too which seems a little odd.

Comment 5

[Andrew McCreight [:mccr8]]

I'm not sure how much this helps, but that's what I know.  I'll see if a CC log produces anything useful.

Comment 6

[(no longer active)]

Attachment: Work around leaks in test_request_context.html in e10s

Do not question what you see in this patch.  It's black arts.

Comment 7

[Nikhil Marathe [:nsm] (No longer reading bugmail, please needinfo?)]

Comment on attachment 8597434 [details] [diff] [review]
Work around leaks in test_request_context.html in e10s

Review of attachment 8597434 [details] [diff] [review]:
-----------------------------------------------------------------

This seems like a band-aid. Any idea why we really leak? Does this have something to do with plugins running in a separate process?

Anyway, please remove the 'black arts' description when landing ;)

Comment 8

[Andrew McCreight [:mccr8]]

Yeah, I can file a bug about fixing the actual leak the current case reveals.

Comment 9

[(no longer active)]

I really have no idea.  There are two issues here, one having something to do with XSLT holding the document alive until shutdown, and one with code that handles the network loads made from within a plugin.  Neither is really something that I'm familiar with, and I'd rather not spend time debugging these leaks honestly.

Comment 10

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/13cd89027fa9

Comment 11

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/13cd89027fa9

Comment 12

[Andrew McCreight [:mccr8]]

(In reply to Andrew McCreight [:mccr8] from comment #8)
> Yeah, I can file a bug about fixing the actual leak the current case reveals.

Filed bug 1162229.