Bugzilla – Full Text Bug Listing
| Summary: | [implementation] Deadlock in SharedASTProvider.getAST() (on breakpoint creation) | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | [Eclipse Project] JDT | Reporter: | Andrey Loskutov <loskutov> | ||||||||
| Component: | Text | Assignee: | Dani Megert <daniel_megert> | ||||||||
| Status: | VERIFIED FIXED | QA Contact: | |||||||||
| Severity: | critical | ||||||||||
| Priority: | P3 | CC: | benno.baumgartner, cameron.bateman, daniel_megert, david_williams, jtk499, markus.kell.r, mikko.tiihonen, raghunathan.srinivasan, remy.suen, victor.zverovich | ||||||||
| Version: | 3.7 | Flags: | markus.kell.r:
review+
|
||||||||
| Target Milestone: | 3.6.1 | ||||||||||
| Hardware: | PC | ||||||||||
| OS: | Linux | ||||||||||
| Whiteboard: | |||||||||||
| Attachments: |
|
||||||||||
|
Description
Andrey Loskutov
Created attachment 173355 [details]
deadlocked VM dump
Dump created with jstack -l.
Created attachment 174962 [details]
deadlock with 3.6
A similar deadlock happened here (with Eclipse 3.6).
>I will attach stack dump. As noticed in bug 302581 comment 3, usual pattern is
>to check a condition BEFORE calling .wait() method on an object.
We check isReconcile(...) inside the synchronize and before the wait.
I tried many times to reproduce it on double-click but no luck. The only thing I can see at this point in the code is the possibility that someone unexpectedly (cancel on type is expected) cancels the progress monitor and that the progress monitor state isn't reset correctly. The safest fix for that is to simply not do that check and always set 'fReconciling' to 'false'. While this can theoretically lead to additional AST creation in rare cases, it at least closes the hole. Also, our Java editor open & activation performance/stress tests show no additional AST creation and no regressions.
It would still be great to have steps to reproduce the bug in the UI. Can one of you reproduce it? Can you remember the steps you did just before the deadlock happened (e.g. type, switch editor, ...)?
(In reply to comment #3) > It would still be great to have steps to reproduce the bug in the UI. Can one > of you reproduce it? Can you remember the steps you did just before the > deadlock happened (e.g. type, switch editor, ...)? Hard to say how to reproduce this. I've got this by debugging Eclipse from Eclipse, and usually I use the same (Java) perspective for the debugging tasks, so that I have debug view, breakpoints etc in my default Java perspective. But except this everything else was "just usual". *** Bug 321483 has been marked as a duplicate of this bug. *** This is occurring around once a day on my machine. It seems to occur simply by opening a Java editor. This is showstopper for us on Helios if it is widely reproducible among our users. When is 3.6.1 scheduled? >When is 3.6.1 scheduled? September but you can already take one of the maintenance builds towards 3.6.1, e.g. http://download.eclipse.org/eclipse/downloads/drops/M20100811-0800/index.php. This would be very useful in order to tell us whether the fix works for you. (In reply to comment #7) > ... but you can already take one of the maintenance builds towards 3.6.1, > e.g. > http://download.eclipse.org/eclipse/downloads/drops/M20100811-0800/index.php. > This would be very useful in order to tell us whether the fix works for you. This fix has not yet been released for 3.6.1, so it's not yet in an M-build. But it would be great if you could test it in a 3.7 build, e.g.: http://download.eclipse.org/eclipse/downloads/drops/S-3.7M1-201008051700/index.php >This fix has not yet been released for 3.6.1, so it's not yet in an M-build.
Ah right. Sorry about that.
Markus, OK for 3.6.1 RC2? +1 for releasing ASTProvider 1.74 for 3.6.1. It's hard to verify without being able to reproduce, but the scenario with the canceled progress monitor sounds plausible, and the fix removes the deadlock opportunity in that case. Fixed in R3_6_maintenance. Available in builds >= M20100825-0800. Verified in code of M20100825-0800. *** Bug 325927 has been marked as a duplicate of this bug. *** *** Bug 325927 has been marked as a duplicate of this bug. *** *** Bug 335656 has been marked as a duplicate of this bug. *** *** Bug 341901 has been marked as a duplicate of this bug. *** Created attachment 198002 [details]
jstack output for a deadlocked eclipse
Just happened to me in
"Eclipse Java EE IDE for Web Developers.
Version: Helios Service Release 2
Build id: 20110218-0911" on a 64-bit Windows 7.
See the attached jstack.txt
So this should be reopened?
(In reply to comment #18) > Created attachment 198002 [details] > jstack output for a deadlocked eclipse > > Just happened to me in > > "Eclipse Java EE IDE for Web Developers. > > Version: Helios Service Release 2 > Build id: 20110218-0911" on a 64-bit Windows 7. > > See the attached jstack.txt > > So this should be reopened? No. There was another bug which we already fixed in 3.7. |