Bug 37621

Summary:

java compiler creates class with internal inconsistency

Product:

[Eclipse Project] JDT

Reporter:

Ken Chupa <chupa>

Component:

Core

Assignee:

Philipe Mulet <philippe_mulet>

Status:

VERIFIED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

eckart.langhuth, jim-aj, vlahoda

Version:

2.1

Target Milestone:

2.1.1

Hardware:

OS:

Windows 2000

Whiteboard:

Attachments:

Description	Flags
Source code for the class that gets compiled incorrectly	none
The "inconsistent" class file	none

Description Ken Chupa

2003-05-14 17:23:55 EDT

The class gets the following error at runtime:
java.lang.VerifyError: (class: org/spec/jappserver/ruleengine/RuleParser, 
method: TypeDeclaration signature: ()V) Illegal target of jump or branch

The test case is SPECjAppServer2002 v1.14. The ear file contains several EJB 
jar files, with interdependencies. RuleParser is in corp.jar. During the build, 
got the following msg from RMIC
[*Warning] ejbModule/org/spec/jappserver/ruleengine/RuleParser.java(1995):
The static method addDimension(int, LinkedList) from the type RuleProcessor
should be accessed in a static way

Comment 1 Ken Chupa

2003-05-14 17:34:34 EDT

Created attachment 4879 [details]
Source code for the class that gets compiled incorrectly

I will also attach the class file created from this source

Comment 2 Ken Chupa

2003-05-14 17:48:23 EDT

Created attachment 4880 [details]
The "inconsistent" class file

This is the class file that gets the VerifyError at runtime

Comment 3 Philipe Mulet

2003-05-14 18:17:51 EDT

Which build are you using ?

Comment 4 Ken Chupa

2003-05-15 12:15:24 EDT

This is occuring with driver
wswb-M20030507-200305071654

Comment 5 Philipe Mulet

2003-05-16 06:20:18 EDT

I was hoping it to be a dup of bug 37040, but the timestamp looks that cannot 
be true. Will investigate.

Comment 6 Philipe Mulet

2003-05-16 06:39:24 EDT

Reproduced.

Comment 7 Philipe Mulet

2003-05-16 07:07:24 EDT

Working on a fix for it for 2.1.1 (not ready yet).

Problem is due to statements {if (true) throw new ParserException(...)}, which 
are fooling us into thinking we can optimize out the trailing #return bytecode, 
which is a branch target (for the break inside the switch statement).

Simpler test case:
public class X {
	
	public static void main(String[] args) {
		if (args != null) {
			if (true) throw new NullPointerException();
		} else {
			switch (args.length) {
				case 1: 
					if (true) throw new NullPointerException
();
					break;
				case 0: 
					break;
				default: 
					throw new NullPointerException();
			}
		}
	}
}

Comment 8 Philipe Mulet

2003-05-16 07:46:23 EDT

Bug isolated. The flow analysis got fooled by a fake reachable 'break' 
statement which tricked it into ignoring subsequent valid break statements 
which can reach the end of the switch. Thus it thought that all subsequent 
statements behind the switch could be optimized out (in particular the implicit 
return bytecode could be thrown away).

In RuleParser, you may workaround it by changing the following lines 183-184 

    {if (true) throw new ParseException("Interface declarations not supported 
in rule!");}
        break;

into the one line below:

{throw new ParseException("Interface declarations not supported in rule!");}

Comment 9 Philipe Mulet

2003-05-17 04:35:20 EDT

Fixed and released in both 2.1 maintenance and 3.0 streams.

Comment 10 David Audel

2003-06-02 09:23:21 EDT

Verified.

Comment 11 Philipe Mulet

2003-06-03 05:17:30 EDT

*** Bug 38345 has been marked as a duplicate of this bug. ***

Comment 12 Philipe Mulet

2003-06-10 09:32:45 EDT

*** Bug 38702 has been marked as a duplicate of this bug. ***

Comment 13 David Audel

2003-06-10 10:25:01 EDT

Verified for 3.0M1.

Comment 14 Philipe Mulet

2003-06-17 15:10:52 EDT

*** Bug 39020 has been marked as a duplicate of this bug. ***