408927 - test_compute_data_with_start_struct.html leaks documents

Status: VERIFIED WORKSFORME

(Core :: DOM: Core & HTML, defect, P2)

Description

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Attachment: Leak log

running the mochitest layout/style/test/test_compute_data_with_start_struct.html leaks a whole pile of objects, including one nsDocument.

Looks like test_inherit_storage.html is leaking a similar object graph

Comment 1

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

I think this was previously reported in bug 393993 comment 2, but I can't find anywhere it was filed as its own bug.

Comment 2

[Mike Schroepfer]

recommended owner and priority?

Comment 3

[Jeff Walden [:Waldo]]

Actually, this is a regression from the last month or two, I think.  With the XBL fix that Eli wrote I don't remember us leaking any windows/documents/docshells, and I guess I just haven't had time recently to run the tests to notice this.

Comment 4

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Peter, would you have the cycles to look at this one?

If it is in fact a regression (dbaron didn't think it was, but he might have been thinking of the XBL bug Eli fixed) this might be a regression from the document::destroy patch. Though the only indication I'd have of that is that i'm worried about leak regressions from that patch.

Comment 5

[Peter Van der Beken [:peterv]]

I can't reproduce this on an OS X trunk build from today. Can anyone else reproduce this on Windows?

Comment 6

[Jeff Walden [:Waldo]]

I can't reproduce on Linux or Windows, running all the tests in layout/style/test.  Are you absolutely sure you're testing an up-to-date build?

Comment 7

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

I was definitely using an up-to-date build, but I think I might have had flashgot and/or add block plus installed.

Comment 8

[Robert Sayre]

Jonas, still seeing this?

Comment 9

[Jeff Walden [:Waldo]]

I can't reproduce this on OS X, either -- try the following command to quickly verify:

python runtests.py --test-path=layout/style/test --autorun --close-when-done --leak-threshold=0

Comment 10

[Johnny Stenback (:jst)]

Not blocking on this leak bug.

Comment 11

[Johnny Stenback (:jst)]

Can anyone still reproduce this? Renominate appropriately if so...

Comment 12

[Peter Van der Beken [:peterv]]

Going to mark this WFM. Jonas: please reopen if you can reproduce.

Comment 13

[Serge Gautherie (:sgautherie)]

[Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.9.1b1pre) Gecko/20080915194703 Minefield/3.1b1pre] (home, optim default) (W2Ksp4)

No leaks with either

python runtests.py --test-path=layout/style/test/test_compute_data_with_start_struct.html --leak-threshold=0

python runtests.py --test-path=layout/style/test --autorun --close-when-done --leak-threshold=0

V.WorksForMe