810517 - "TypeError: listenerWindow is null" when running test_browser.py twice in a row

Status: RESOLVED FIXED

(Remote Protocol :: Marionette, defect)

Description

[Jonathan Griffin (:jgriffin)]

According to Zac, https://github.com/zacc/gaia-ui-tests/blob/browser_smoke_test/gaiatest/tests/test_browser.py fails when run twice in a row.  Needs investigation.

Comment 1

[Jonathan Griffin (:jgriffin)]

The symptoms are thus:

The test runs ok the first time.  On the second run, it launches the browser and successfully navigates the browser to mozilla.org.  When it tries to switch into the browser's iframe (https://github.com/zacc/gaia-ui-tests/blob/master/gaiatest/tests/test_browser.py#L49), an error occurs and the call to switch_to_frame dies:

E/GeckoConsole(  106): [JavaScript Error: "TypeError: listenerWindow is null" {file: "chrome://marionette/content/marionette-actors.js" line: 1658}]

Comment 2

[Jonathan Griffin (:jgriffin)]

The problem is here:  http://mxr.mozilla.org/mozilla-central/source/testing/marionette/marionette-actors.js#1654

Child processes and the parent process maintain completely independent lists of windows, from which outerWindowID comes.  When a register message is received from an OOP frame, the outerWindowID comes from the child processes' window list...the same outerWindowID in the parent process may not exist, or may point to some random window.

Comment 3

[Jonathan Griffin (:jgriffin)]

Attachment: Check for valid listenerWindow,

The solution is just to add another check that listenerWindow is valid before attempting to access it's href.  Unfortunately, I don't know of a better way to determine whether the frame registration is coming from an OOP process, but it would be nice if there were.

Comment 4

[Jonathan Griffin (:jgriffin)]

Actually, I wonder if we can use window.parent to determine this?

Comment 5

[Jonathan Griffin (:jgriffin)]

(In reply to Jonathan Griffin (:jgriffin) from comment #4)
> Actually, I wonder if we can use window.parent to determine this?

No, we can't; that trick only works *from* the OOP process, not the parent one.

Comment 6

[Malini Das [:mdas] - Away, not checking bugmail]

Comment on attachment 680739 [details] [diff] [review]
Check for valid listenerWindow,

Review of attachment 680739 [details] [diff] [review]:
-----------------------------------------------------------------

Good catch! I'd add a comment about the outerWindowID mismatch if you can, to explain why we may not find the listenerWindow.

Comment 7

[Jonathan Griffin (:jgriffin)]

pushed to try: https://tbpl.mozilla.org/?tree=Try&rev=0eb0be2c7d95

Comment 8

[Jonathan Griffin (:jgriffin)]

Attachment: Check for valid listenerWindow,

Comment 9

[Jonathan Griffin (:jgriffin)]

Comment on attachment 680866 [details] [diff] [review]
Check for valid listenerWindow,

Review comments addressed; carrying r+ forward.

Comment 10

[Jonathan Griffin (:jgriffin)]

Also pushed to ash:  https://tbpl.mozilla.org/?tree=Ash&rev=4edf1ed07d1e

Comment 11

[Jonathan Griffin (:jgriffin)]

re-pushed to try, since my original didn't seem to trigger tests on the emulator:  https://tbpl.mozilla.org/?tree=Try&rev=6066eca91645

Comment 12

[Jonathan Griffin (:jgriffin)]

https://hg.mozilla.org/integration/mozilla-inbound/rev/76509d0847ec

Comment 13

[Jonathan Griffin (:jgriffin)]

ash looks ok, pushing to aurora: https://hg.mozilla.org/releases/mozilla-aurora/rev/484b0596984a

Comment 14

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/76509d0847ec