261510 – [compiler] Deadlock in static initializer of JDT classes

Bug 261510 - [compiler] Deadlock in static initializer of JDT classes

Summary: [compiler] Deadlock in static initializer of JDT classes

Status:	VERIFIED FIXED

Alias:	None

Product:	JDT
Classification:	Eclipse Project
Component:	Core (show other bugs)
Version:	3.4.2
Hardware:	PC Windows XP

Importance:	P3 critical (vote)
Target Milestone:	3.5 M6
Assignee:	Philipe Mulet
QA Contact:

URL:
Whiteboard:
Keywords:

Depends on:
Blocks:

Reported:	2009-01-19 10:14 EST by Xu XIANG
Modified:	2009-03-10 10:21 EDT (History)
CC List:	2 users (show)

See Also:

Attachments
snip of thread dump (4.42 KB, text/plain) 2009-01-19 10:14 EST, Xu XIANG	no flags	Details
Proposed patch (83.19 KB, patch) 2009-02-02 07:12 EST, Philipe Mulet	no flags	Details \| Diff
Better patch (83.39 KB, patch) 2009-02-02 10:12 EST, Philipe Mulet	no flags	Details \| Diff
Proposed patch for 3.4 (8.29 KB, patch) 2009-02-02 10:15 EST, Philipe Mulet	no flags	Details \| Diff
Show Obsolete (1) View All Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.

Description Xu XIANG

2009-01-19 10:14:39 EST

Created attachment 122943 [details]
snip of thread dump

Build ID: I20080617-2000

We faced the deadlock in our JUnit test run, see attached snip of thread dump. We refer to Java spec (http://java.sun.com/docs/books/jls/third_edition/html/execution.html#12.4.2) and are assuming:

- Thread "Java indexing" is going to initialize Constant class (= Constant class is locked) and tries to initialize DoubleConstant (which is locked by Worker-22)

- Worker-22 locks DoubleConstant during initialization of LongLiteral and tries to lock the parent class Constant


BTW: It is anyway questionable to access a child class in the static initializer of Constant

More information:
[2009.01.19 02:56:25] !SESSION 2009-01-19 02:56:04.717 -----------------------------------------------
[2009.01.19 02:56:25] eclipse.buildId=I20080617-2000
[2009.01.19 02:56:25] java.version=1.6.0_05
[2009.01.19 02:56:25] java.vendor=Sun Microsystems Inc.

Comment 1 Philipe Mulet

2009-02-02 05:29:20 EST

This is a long standing weakness.
And, actually, LongLiteral#FORMAT_ERROR could be eliminated entirely.

Comment 2 Philipe Mulet

2009-02-02 07:09:51 EST

This is a rare situation which has been existing since day 1.

Comment 3 Philipe Mulet

2009-02-02 07:10:40 EST

Also IntLiteral is exhibiting the same pattern.

Comment 4 Philipe Mulet

2009-02-02 07:12:25 EST

Created attachment 124410 [details]
Proposed patch

Also cleaned up some dusty code.

Comment 5 Philipe Mulet

2009-02-02 10:12:35 EST

Created attachment 124424 [details]
Better patch

Comment 6 Philipe Mulet

2009-02-02 10:15:28 EST

Created attachment 124425 [details]
Proposed patch for 3.4

Comment 7 Philipe Mulet

2009-02-02 11:21:50 EST

Released for 3.5M6
Fixed

Comment 8 Kent Johnson

2009-03-10 10:21:21 EDT

Verified for 3.5M6 using I20090310-0100