540306 - Ubiquity hangs since tracemonkey 2010-01-05-04, no "Unresponsive script" message

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Nickolay_Ponomarev]

Steps to reproduce:
1. hg clone https://ubiquity.mozilla.com/hg/ubiquity-firefox
2. hg tip
changeset:   3447:55f148ad19d6
tag:         tip
user:        satyr <[e-mail]>
date:        Sat Dec 05 09:00:19 2009 +0900
summary:     locale/it-IT: Removed an accidental "f".
3. cd ubiquity-firefox
4. ./manage.py install -p <clean profile>
5. Run firefox on that profile, activate ubiquity (Ctrl+TAB on windows, Alt+TAB on Mac by default).
6. Press "z"

Actual results: Firefox stops responding, no "Unresponsive script" message, despite the dom.max_chrome_script_run_time > 0.

Expected results: no hang.

2010-01-04-04-tracemonkey: works
2010-01-05-04-tracemonkey: fail
2010-01-05-04-mozilla-central: works (so this is not due to the merge from m-c)
2010-01-11-04-mozilla-central: works
2010-01-12-05-mozilla-central: fail (with the merge from tracemonkey)
http://hg.mozilla.org/tracemonkey/pushloghtml?fromchange=c00e5b745651&tochange=6d00d56ab3ed

The hanging process is initialized in __delayedProcessInput in ubiquity.js, which apparently creates a generator that does the actual work.

No minimized testcase, sorry, because I hit bug 531886.

Comment 1

[Nickolay_Ponomarev]

> 5. Run firefox on that profile, activate ubiquity (Ctrl+TAB on windows, Alt+TAB
> on Mac by default).

Sorry, CTRL/ALT+SPACE, not TAB.

Comment 2

[Nickolay_Ponomarev]

Attachment: sample from activity monitor

Comment 3

[David Mandelin [:dmandelin]]

Any chance you could test builds and bisect that TM regression range down to one patch?

Comment 4

[Nickolay_Ponomarev]

If I didn't mess it up, bisect points to http://hg.mozilla.org/tracemonkey/rev/3862a7e48e79 (bug 98409)... If so, ubiquity probably was relying on the old behavior and should be fixed, but it requires someone familiar with that code to figure it out...

The "Unresponsive script" not appearing is probably a separate issue, which I think is a bug.

Comment 5

[Nickolay_Ponomarev]

Posted http://groups.google.com/group/ubiquity-core/browse_frm/thread/7d90a08031203343

Comment 6

[Dietrich Ayala (:dietrich)]

can you clarify what the fix is? remove the re-use of a regexp inside a loop?

Comment 7

[Brendan Eich [:brendan]]

Reusing a regexp literal in a loop now means re-evaluating it to a fresh object every iteration, just as with object, array, and function literals -- so it seems there's a loop that expected the regexp literal to create a singleton, which would eventually mismatch due to its one and only lastIndex marching off the end of the target string -- assuming the g flag was used.

Someone needs to use firebug or poor-man's profiling (interrupt in gdb or msvc) and find the loop.

/be

Comment 8

[satyr]

https://ubiquity.mozilla.com/hg/ubiquity-firefox/rev/e1cd63803d5c

Did this work?

Comment 9

[Nickolay_Ponomarev]

satyr: yes, that appears to fix the hang I experienced. Thanks!

Anyone interested in figuring out why the "Unresponsive script" dialog didn't pop up in this case?