Linux debug Z7/Z8 tests busted after bug 1542711
Categories
(Thunderbird :: General, defect)
Tracking
(Not tracked)
People
(Reporter: darktrojan, Assigned: darktrojan)
References
(Regression)
Details
(Keywords: regression)
Attachments
(2 files)
(deleted),
patch
|
mkmelin
:
review-
|
Details | Diff | Splinter Review |
(deleted),
patch
|
mkmelin
:
review+
|
Details | Diff | Splinter Review |
I'm not sure why this isn't a failure on Mac and Windows, but Linux debug tests have been burning since bug 1542711 landed. The failing tests are
- testAlarmDefaultValue.js
- pref-window/test-attachments-pane.js
- pref-window/test-font-chooser.js
- … and possibly others
All these tests involve the preferences tab.
I've tracked the fault down to an address book listener that doesn't get removed when the document it is in unloads. It looks like a CE's disconnectedCallback
doesn't fire at the point like an XBL binding's destructor
.
I'm doing a Try run to check that my fix doesn't cause further problems, then all going well, I'll land it.
Assignee | ||
Comment 1•6 years ago
|
||
Pushed by geoff@darktrojan.net:
https://hg.mozilla.org/comm-central/rev/230148914b48
Remove address book listener at document unload; rs=bustage-fix
Assignee | ||
Updated•6 years ago
|
Comment 3•6 years ago
|
||
Updated•6 years ago
|
Comment 4•6 years ago
|
||
So what do I do with this? A landed patch with failed the so-called "post landing review"? Back it out?
Comment 5•6 years ago
|
||
Depends on what the actual underlying problem is/was here. Why was linux debug crashing?
It's expected that disconnectedCallback doesn't run when the document unloads.
Comment 6•6 years ago
|
||
Well, you guys sort it out and let me know.
Assignee | ||
Comment 7•6 years ago
|
||
window.removeEventListener("unload", this._onUnload);
Normally there is no need to remove event listeners (they go away when the
element goes away).
This is also wrong here since connectedCallback() can run multiple times.
There is protection from having it run more than once in this code. For the
case that it's connected again, it would now not have a listener set up
anymore.
So (only) that line is not wanted?
Depends on what the actual underlying problem is/was here. Why was linux debug crashing?
Because the address book code was holding a reference to the menulist that was never released.
Comment 8•6 years ago
|
||
(In reply to Geoff Lankow (:darktrojan) from comment #7)
So (only) that line is not wanted?
There was a problem before the patch too that the listener was being removed in disconnectedCallback but only added the first time in connectedCallback (=> problems when disconnect/connect multiple times)
Comment 9•6 years ago
|
||
So maybe .addEventListener with the "once" option? (And no need to set the listener it as a property then.)
Assignee | ||
Comment 10•6 years ago
|
||
Stops removing the unload listener. I haven't changed the address book listener, I'm just here to fix the "crash".
Comment 11•6 years ago
|
||
Comment 12•6 years ago
|
||
Pushed by mozilla@jorgk.com:
https://hg.mozilla.org/comm-central/rev/e3be3b2f7a06
Follow-up: Address post-landing review comments. r=mkmelin DONTBUILD
Comment 13•6 years ago
|
||
Description
•