1179782 - test_bug436801.html, test_bug494328.html, test_bug589543.html, test_registerHandler.html leaks for Windows & Mac (debug platforms), with run-by-dir enabled. test_bug589543.html also leaks for linux.

Status: RESOLVED FIXED

(Firefox Graveyard :: RSS Discovery and Preview, defect)

Description

[Kaustabh Datta Choudhury]

User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:36.0) Gecko/20100101 Firefox/36.0
Build ID: 20150321175551



Actual results:

As we can see on try, even after removal of a single test, the leaks persist. This was repeated for all the tests, and unless all the tests were removed, the leaks persisted.

https://treeherder.mozilla.org/#/jobs?repo=try&revision=1ea5b2b8a6cb

Snippet of the log, detailing the leak : 
https://pastebin.mozilla.org/8838364

Comment 1

[Florian Quèze [:florian]]

These tests are all old, right? What has changed for this issue to become visible now?

Comment 2

[Kaustabh Datta Choudhury]

(In reply to Florian Quèze [:florian] [:flo] from comment #1)
> These tests are all old, right? What has changed for this issue to become
> visible now?

We are trying to minimize failures when run-by-dir is enabled. These tests are causing leaks, so we'd like to have them resolved. 
When we run with rin-by-dir enabled, we only run the browser for the tests in that directory & then shut it down. This might cause us to terminate before delayed cleanup takes place.
Any leads on how to solve them would be helpful.

Comment 3

[Kaustabh Datta Choudhury]

Resolving this bug would resolve all leaks for mochitest-plain with run-by-dir enabled. These are causing leaks in Windows, mac, linux and linux-asan.

Comment 4

[Bobby Holley (:bholley)]

Most of the stuff in the leak log looks generic, except for nsAuthURLParser. I'd start there.

Leaking a Backstage pass and a bunch of XPC* objects just means that we're leaking a system compartment (either JS-implemented XPCOM component or JSM) - it doesn't tell us how that's happening.

Comment 5

[Joel Maher ( :jmaher ) (UTC -8)]

:bholley, any tips on how to start on the nsAuthURLParser?  is that something related to https?

Comment 6

[Bobby Holley (:bholley)]

(In reply to Joel Maher (:jmaher) from comment #5)
> :bholley, any tips on how to start on the nsAuthURLParser?  is that
> something related to https?

For general tips on debugging mochitest memory leaks, you probably want Andrew.

Comment 7

[Andrew McCreight [:mccr8]]

I'll take a look. This RSS feed stuff comes up pretty frequently when you try to change how collection happens, because it involves a lot of ancient XPCOM stuff that isn't cycle collected properly.

Comment 8

[Joel Maher ( :jmaher ) (UTC -8)]

with bug 1184400 fixed should some or all of this be fixed?  maybe a try push this weekend could tell us?

Comment 9

[Andrew McCreight [:mccr8]]

(In reply to Joel Maher (:jmaher) from comment #8)
> with bug 1184400 fixed should some or all of this be fixed?  maybe a try
> push this weekend could tell us?

Bug 1184400 fixes the entire leak for me locally when I run
  ./mach mochitest -f plain browser/components/feeds/test/
Let me know if it is still broken in your test runs.

Comment 10

[Kaustabh Datta Choudhury]

This is fixed : https://treeherder.mozilla.org/#/jobs?repo=try&revision=156e71936b5f

Thanks for getting this working.