Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[tromey]

Synopsis: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

Responsible-Changed-From-To: unassigned->tromey
Responsible-Changed-By: tromey
Responsible-Changed-When: Mon Mar 25 22:27:53 2002
Responsible-Changed-Why:
    I'm handling the PR manipulation.
State-Changed-From-To: open->closed
State-Changed-By: tromey
State-Changed-When: Mon Mar 25 22:27:53 2002
State-Changed-Why:
    Per checked in the fix and asked me to close the PR.

http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&pr=5812

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[Per Bothner]

The following reply was made to PR java/5812; it has been noted by GNATS.

From: Per Bothner <per@bothner.com>
To: tromey@redhat.com
Cc: adam@medovina.org, gcc-gnats@gcc.gnu.org
Subject: Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler
Date: Tue, 05 Mar 2002 11:34:36 -0800

 This is a multi-part message in MIME format.
 --------------090401000906060607030103
 Content-Type: text/plain; charset=us-ascii; format=flowed
 Content-Transfer-Encoding: 7bit
 
 With this patch I get the expected output:
 
 eureka<3517>/home/bothner/GNU/linux-gcc/gcc/jc1 Bug.class -o Bug.s -quiet
 Bug.java: In class `Bug':
 Bug.java: In method `Bug.invoke()':
 Bug.java:4: warning: unreachable bytecode from 3 to before 6
 Bug.java:7: warning: unreachable bytecode from 10 to before 12
 Bug.java:9: warning: unreachable bytecode from 15 to the end of the method
 -- 
 	--Per Bothner
 per@bothner.com   http://www.bothner.com/per/
 
 --------------090401000906060607030103
 Content-Type: text/plain;
  name="expr.patch"
 Content-Transfer-Encoding: 7bit
 Content-Disposition: inline;
  filename="expr.patch"
 
 Index: expr.c
 ===================================================================
 RCS file: /cvs/gcc/gcc/gcc/java/expr.c,v
 retrieving revision 1.134
 diff -u -p -r1.134 expr.c
 --- expr.c	2002/02/28 23:27:09	1.134
 +++ expr.c	2002/03/05 19:32:15
 @@ -48,7 +48,7 @@ static tree pop_value PARAMS ((tree));
  static void java_stack_swap PARAMS ((void));
  static void java_stack_dup PARAMS ((int, int));
  static void build_java_athrow PARAMS ((tree));
 -static void build_java_jsr PARAMS ((tree, tree));
 +static void build_java_jsr PARAMS ((int, int));
  static void build_java_ret PARAMS ((tree));
  static void expand_java_multianewarray PARAMS ((tree, int));
  static void expand_java_arraystore PARAMS ((tree));
 @@ -607,15 +607,18 @@ build_java_athrow (node)
  /* Implementation for jsr/ret */
  
  static void
 -build_java_jsr (where, ret)
 -    tree where;
 -    tree ret;
 +build_java_jsr (target_pc, return_pc)
 +     int target_pc, return_pc;
  {
 +  tree where =  lookup_label (target_pc);
 +  tree ret = lookup_label (return_pc);
    tree ret_label = fold (build1 (ADDR_EXPR, return_address_type_node, ret));
    push_value (ret_label);
    flush_quick_stack ();
    emit_jump (label_rtx (where));
    expand_label (ret);
 +  if (instruction_bits [return_pc] & BCODE_VERIFIED)
 +    load_type_state (ret);
  }
  
  static void
 @@ -2972,13 +2975,8 @@ process_jvm_instruction (PC, byte_ops, l
      build_java_ret (find_local_variable (index, ptr_type_node, oldpc));	\
    }
  
 -#define JSR(OPERAND_TYPE, OPERAND_VALUE)		\
 -  {							\
 -    tree where = lookup_label (oldpc+OPERAND_VALUE);	\
 -    tree ret   = lookup_label (PC);			\
 -    build_java_jsr (where, ret);			\
 -    load_type_state (ret);				\
 -  }
 +#define JSR(OPERAND_TYPE, OPERAND_VALUE) \
 +  build_java_jsr (oldpc+OPERAND_VALUE, PC);
  
  /* Push a constant onto the stack. */
  #define PUSHC(OPERAND_TYPE, OPERAND_VALUE) \
 
 --------------090401000906060607030103--
 

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[Adam Dingle]

The following reply was made to PR java/5812; it has been noted by GNATS.

From: Adam Dingle <adam@medovina.org>
To: tromey@redhat.com
Cc: gcc-gnats@gcc.gnu.org, Per Bothner <per@bothner.com>
Subject: Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3
 compiler
Date: Sat, 02 Mar 2002 12:05:35 -0800

 > However, given that Sun's compiler generated this, perhaps they intend
 > that we only verify reachable code.  In this case, the `return' is not
 > reachable since there is no `ret' corresponding to the `jsr'.
 
 I'm pretty sure that Sun verifies only reachable code and believes that this 
 is what a byte code verifier should do.  The description of the bytecode 
 verification process in section 4.9.2 of the Java Virtual Machine 
 Specification describes a data-flow analyzer which will ignore unreachable 
 code; see
 
 http://java.sun.com/docs/books/vmspec/2nd-edition/html/ClassFile.doc.html#9801
 
 -adam

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[Per Bothner]

The following reply was made to PR java/5812; it has been noted by GNATS.

From: Per Bothner <per@bothner.com>
To: tromey@redhat.com
Cc: adam@medovina.org, gcc-gnats@gcc.gnu.org
Subject: Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler
Date: Sat, 02 Mar 2002 10:56:45 -0800

 Tom Tromey wrote:
 > It isn't clear to me whether this should verify.
 > The `return' at PC=15 is invalid, since the method signature is ()I.
 > Ordinarily this would be a verification error.
 
 I agree.
 
 > However, given that Sun's compiler generated this, perhaps they intend
 > that we only verify reachable code.
 
 Who knows what the by intended - but we have to live with their bugs.
 
 > Per, what do you think about this?
 
 Gcj already emits an 'unreachable bytecode' warning. I think that
 would be enough, once we fix whather is causing gcj to crash. I'll
 take a look.
 -- 
 	--Per Bothner
 per@bothner.com   http://www.bothner.com/per/
 

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[Tom Tromey]

The following reply was made to PR java/5812; it has been noted by GNATS.

From: Tom Tromey <tromey@redhat.com>
To: adam@medovina.org
Cc: gcc-gnats@gcc.gnu.org, Per Bothner <per@bothner.com>
Subject: Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler
Date: 02 Mar 2002 12:10:11 -0700

 >>>>> "Adam" == Adam Dingle <adam@medovina.org> writes:
 
 Adam> When I compile the class definition below with the Sun JDK 1.3
 Adam> compiler and then feed the class file to GCJ 3.0.2, GCJ
 Adam> segfaults.  I've included the .class file as an attachment.
 
 First, I agree that gcj should not crash.  That is a bug.
 
 However, this is unusual bytecode.
 Here it is:
 
   0: jsr 12
   3: goto 15
   6: astore_1
   7: jsr 12
  10: aload_1
  11: athrow
  12: astore_2
  13: iconst_1
  14: ireturn
  15: return
 
 It isn't clear to me whether this should verify.
 The `return' at PC=15 is invalid, since the method signature is ()I.
 Ordinarily this would be a verification error.
 
 However, given that Sun's compiler generated this, perhaps they intend
 that we only verify reachable code.  In this case, the `return' is not
 reachable since there is no `ret' corresponding to the `jsr'.
 
 Per, what do you think about this?
 
 I do think it is bogus that Sun's compiler does this.
 For this source we generate the aesthetically pleasing:
 
   0: iconst_1
   1: ireturn
 
 Tom

Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[Alexandre Petit-Bianco]

The following reply was made to PR java/5812; it has been noted by GNATS.

From: Alexandre Petit-Bianco <apbianco@cygnus.com>
To: adam@medovina.org
Cc: gcc-gnats@gcc.gnu.org
Subject: Re: java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler
Date: Sat, 2 Mar 2002 10:02:05 -0800 (PST)

 adam@medovina.org writes:
 
 > When I compile the class definition below with the Sun JDK 1.3
 > compiler and then feed the class file to GCJ 3.0.2, GCJ segfaults.
 
 It's also the case with the current tree and 3.1
 
 ./A

java/5812: GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler

[adam]

>Number:         5812
>Category:       java
>Synopsis:       GCJ 3.0.2 segfaults on class file from Sun JDK 1.3 compiler
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    unassigned
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Mar 02 08:56:00 PST 2002
>Closed-Date:
>Last-Modified:
>Originator:     Adam Dingle
>Release:        GNU Java version 3.0.2 20010905 (Red Hat Linux 7.1 3.0.1-3) (i386-redhat-linux)
>Organization:
>Environment:
Red Hat Linux 7.2
>Description:
When I compile the class definition below with the Sun JDK 1.3 compiler and then feed the class file to GCJ 3.0.2, GCJ segfaults.  I've included the .class file as an attachment.

public class Bug
{
	public int invoke() {
		try {
		}
		finally {
			return 1;
		}
	}
}
>How-To-Repeat:
% gcj3 Bug.class
Bug.java: In class `Bug':
Bug.java: In method `Bug.invoke()':
Bug.java:4: Internal error: Segmentation fault
...
% 
>Fix:

>Release-Note:
>Audit-Trail:
>Unformatted:
----gnatsweb-attachment----
Content-Type: application/x-java; name="Bug.class"
Content-Transfer-Encoding: base64
Content-Disposition: attachment; filename="Bug.class"

yv66vgADAC0ADwoAAwAMBwANBwAOAQAGPGluaXQ+AQADKClWAQAEQ29kZQEAD0xpbmVOdW1iZXJU
YWJsZQEABmludm9rZQEAAygpSQEAClNvdXJjZUZpbGUBAAhCdWcuamF2YQwABAAFAQADQnVnAQAQ
amF2YS9sYW5nL09iamVjdAAhAAIAAwAAAAAAAgABAAQABQABAAYAAAAdAAEAAQAAAAUqtwABsQAA
AAEABwAAAAYAAQAAAAEAAQAIAAkAAQAGAAAAOAABAAMAAAAQqAAMpwAMTKgABSu/TQSssQABAAAA
BgAGAAAAAQAHAAAADgADAAAABAAGAAcADwAJAAEACgAAAAIACw==