Closed Bug 942931 Opened 11 years ago Closed 11 years ago

Can't close browser when debugger is paused

Categories

(DevTools :: Debugger, defect, P2)

defect

Tracking

(firefox25 unaffected, firefox26 affected, firefox27 affected, firefox28 affected)

RESOLVED FIXED
Firefox 28
Tracking Status
firefox25 --- unaffected
firefox26 --- affected
firefox27 --- affected
firefox28 --- affected

People

(Reporter: vporof, Assigned: past)

References

Details

(Keywords: regression)

Attachments

(1 file)

This is exactly like bug 916458, but I'm reincarnating it here. Same str: 1. Open http://htmlpad.org/debugger/ 2. Open Debugger 3. Click me 4. Cmd+Q Happens on fx-team tip.
I'm running into this as well. It causes a hang and I end up needing to kill the process.
When hunting down leaks we disabled the test from bug 916458, which would have caught this regression. *facepalm*
Keywords: regression
This is regression from bug 941787. I have a fix that I'm currently testing.
The patch in bug 941787 expected unrolling the nested event loops in the abrupt shutdown case to proceed normally, but unfortunately that is not the case. When the tab is already gone, we can no longer resume events and timeouts and just need to keep exiting any nested event loops.
Attachment #8342504 - Flags: review?(nfitzgerald)
Assignee: nobody → past
Status: NEW → ASSIGNED
Priority: -- → P2
Comment on attachment 8342504 [details] [diff] [review] Properly exit nested event loops when a tab is abruptly closed Review of attachment 8342504 [details] [diff] [review]: ----------------------------------------------------------------- Can we re-enable the test for this?
Attachment #8342504 - Flags: review?(nfitzgerald) → review+
I also tried to re-enable the browser_dbg_clean-exit-window.js test, but it is still leaking (bug 933950).
If that assert is tripping, that means we didn't pop all the nested event loops before closing the tab, and that we are leaking. The fact that the tab is gone and you can't resume events is precisely how the windows leak. Delayed events that were queued while the events were suppressed get held in nsFocusManager, holding the window alive. I recommend figuring out why the nested event loop depth isn't 0.
Landed this interim fix as discussed on IRC, while we keep trying to fix the leak on window close in bug 933950: https://hg.mozilla.org/integration/fx-team/rev/9473d7b659e9
Whiteboard: [fixed-in-fx-team]
Status: ASSIGNED → RESOLVED
Closed: 11 years ago
Resolution: --- → FIXED
Whiteboard: [fixed-in-fx-team]
Target Milestone: --- → Firefox 28
Keywords: verifyme
I tried to reproduce this issue with the 11/25 Nightly (Fx28) on Mac OS X 10.8.5 and Ubuntu 13.04, but with no results. I tried the steps from comment 0 plus setting multiple breakpoints and trying to close the browser with Ctrl/Cmd + Q when the debugger pauses at a point. Victor, did you do anything else to reproduce this?
Flags: needinfo?(vporof)
Keywords: verifyme
Steps in comment 0 should be exact and sufficient. Make sure you're trying to close the process completely, not hide the window. Alternatively, this might have been uplifted on Firefox 28 without the tracking flags being updated for some reason.
Flags: needinfo?(vporof)
(In reply to Victor Porof [:vporof][:vp] from comment #12) > Steps in comment 0 should be exact and sufficient. Make sure you're trying > to close the process completely, not hide the window. This is exactly what I'm doing. I checked each time and the process was closed.
Product: Firefox → DevTools
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: