Over the pat few days whenever I shutdown minefield after it has been open for multiple hours, or restart for updates I get a hung process consuming 100% CPU. Attached is the report from OSX after I kill it.

Probably a dupe of bug 474699?

Perhaps, I don't have clear private data on shutdown though

I had this problem again a couple of minutes before. I have created a shark profile and will attach it to the bug. As what I can see we spend most of our time in js_ArrayToJSDoubleBuffer while consuming nearly 100% cpu load. + 99.8%, JS_CompareValues, libmozjs.dylib | + 99.0%, js_ArrayToJSDoubleBuffer, libmozjs.dylib | | + 79.7%, js_ArrayToJSDoubleBuffer, libmozjs.dylib | | | 4.4%, js_ArrayToJSDoubleBuffer, libmozjs.dylib | - 0.5%, 0xee9643 [unknown], XUL | - 0.1%, js_CheckUndeclaredVarAssignment, libmozjs.dylib | - 0.0%, js_LookupProperty, libmozjs.dylib | 0.0%, js_FreeStack, libmozjs.dylib | - 0.0%, 0xee961e [unknown], XUL

(In reply to comment #3) > I had this problem again a couple of minutes before. I have created a shark > profile and will attach it to the bug. Can you reproduce this reliably?

Need better symbols, some static function addresses are being truncated to the nearest lower extern neighbor. /be

No, I'm not able to reproduce it reliable. Brendan, means the Shark profile doesn't help you? Do you better need a gdb stack in such a situation?

I have running a debug build of Shiretoko now and saw a hang during shutdown today. But as long as I don't know what I have to do I cannot spend more time in trying to investigate this problem. It would be great to get some steps so I will know what information I have to acquire for you.

A profile from a shark build should work better: http://ftp.mozilla.org/pub/mozilla.org/firefox/nightly/latest-trunk/firefox-3.6a1pre.en-US.mac-shark.dmg

I was already using a shark enabled build. No idea what was going wrong. Anyway, now I have used a shark enabled debug build and got this profile. Shiretoko needed a couple of minutes to shutdown. This profile shows a 30s range. If you need more I can try to get a bigger one.

The patches from bug 475737 and bug 490592 might help here.

Maybe... but why is the root traversal so slow, exactly? How many roots do we have in this case? Is there something Henrik could do to output that information?

There's probably still another problem, but it's certainly going to be amplified by doing too many GCs on shutdown.

I don't think this a JS engine bug. This has happened to me a couple times, and I think it's imgCacheEntry thread safety problems.: http://pastebin.mozilla.org/648509

Joe, could you have a look at this bug?

There's nothing about imagelib in this shark profile.

(In reply to comment #16) > There's nothing about imagelib in this shark profile. See comment 14 and follow the pastebin link. /be

Should we kick this over into someone else's department for trackability? Its still assigned to JS.

I sincerely doubt comment 14 and that pastebin output have anything to do with this bug as filed (and in particular, I doubt they lead to the shark profile posted herein). There should be a seprate bug filed on that assertion, of course.

(In reply to comment #19) > I sincerely doubt comment 14 and that pastebin output have anything to do with > this bug as filed (and in particular, I doubt they lead to the shark profile > posted herein). yeah, my fault, could be two different bugs.

We just approved the patches peterv mentioned in comment 11 - clearing the blocker nom, please re-approve if we can reproduce this on branch and trunk after those land.

Adding both bugs to the dependency list so we don't forget this bug.

$0.02: I've seen this much less often in recent months; perhaps some unexplained hangs were related to Flash 9 (as mentioned in other bugs). I don't use "Always clear private data" (as in bug 474699). IIRC this (or something much like it) also happened in FF2 sometimes.

Does anybody of you have Ubiquity installed? I suspect that this extension causes this problems at least for me. After its deactivation I don't have those problems anymore. A shutdown happens under 10s now. Last time I tested this I made two snapshots of open files during the shutdown process. None of both lists show places.sqlite but a lot of SQLite files from Ubiquity.

This list shows open files during shutdown after the browser was open for about 3 days.

List of open files after I have restarted the browser and closed it immediately after all tabs have been finished loading.

Is this seen in 4.0?