1250058 - docShell leak on e10s debug when using getTestActorWithoutToolbox

Status: RESOLVED INACTIVE

(DevTools :: General, defect, P2)

Description

[Julian Descottes [:jdescottes]]

Follow up to bug 1244120. 

Locally this issue only happens on windows7, with a debug non-optimized build.
Tests impacted : 
- devtools/client/inspector/test/browser_inspector_initialization.js
- devtools/client/inspector/rules/test/browser_rules_content_02.js

STRs :
- build application on Win7 with --enable-debug & --disable-optimize
- run one of the tests mentioned above with ./mach mochitest --e10s

Actual : 1 docShell is leaked
Expected : no leak

Comment 1

[Ryan VanderMeulen [:RyanVM]]

Affects OSX as well.

Comment 3

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/1d1490dc1e74

Comment 4

[Carsten Book [:Tomcat]]

https://hg.mozilla.org/mozilla-central/rev/1d1490dc1e74

Comment 5

[Ryan VanderMeulen [:RyanVM]]

And Linux.

Comment 6

[Pulsebot]

https://hg.mozilla.org/integration/mozilla-inbound/rev/73270556892f

Comment 7

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/73270556892f

Comment 8

[J. Ryan Stinnett [:jryans] (Use needinfo, replies may be slow)]

Filter on TEAPOT-SPLINES.

Comment 9

[Julian Descottes [:jdescottes]]

Attachment: bug1250058.investigation.patch

I am stuck in my investigation and would need help here. Attaching a "reduced" test case here. In the test, you can see the various combinations that leak and those that don't.

Long story short, if at any point, a TestActor connection is live when opening the toolbox or when creating another TestActor, a docshell leaks. If the connection is closed before doing that, no leak.

Not included in this patch, I also tried to implement a better destroy for the test actor, without success.

Alex: I think you know the test-actor well, any idea what could be going wrong here? Also would be happy to take any suggestion regarding tools/methods to investigate docshell leaks.

Comment 10

[Alexandre Poirot [:ochameau]]

Hum. I would have to dig deeper into this to provide really useful comment...
May be that's related to actor-registry actor which is used in test-actor-registry.js?
May that's him keeping stuff alive? I really don't know.

But here is a little trick I'm using to help figuring out what leaks.
Memory panel is meant to provide new console methods to do even better, but I haven't heard it being implemented yet?

let weak = Cu.getWeakReference(myObject);
require("sdk/timers").setInterval(function () { // /!\ do not use arrow function here as it keeps 'this' alive!!
  Cu.forceCC(); Cu.forceGC();
  if (weak.get()) {
    dump("myObject is still alive\n");
  }
}, 2000);

This snippet is handy if you execute a test with --keep-open, or 'yield new Promise(function () {})' just after a known leak. It allows to narrow down to what leaks. It is laborious, but it's better than nothing ;)

Comment 11

[BugBot [:suhaib / :marco/ :calixte]]

The leave-open keyword is there and there is no activity for 6 months.
:pbro, maybe it's time to close this bug?