This Bug stems from the setting of a new signal handler in 
GC_SysVGetDataStart() and the SIGSEGV is not getting caught.

../gcc_trunk/boehm-gc/os_dep.c :
...
    GC_setup_temporary_fault_handler();
    if (SETJMP(GC_jmp_buf) == 0) {
    	/* Try writing to the address.	*/
    	*result = *result;                    /* SIGSEGV */
        GC_reset_fault_handler();
    } else {
...

A quick workaround is to delete or rename your 'gjar' before
gcc's configury detects it. The gcc scripts will workaround
the absence of 'gjar' during the build. 

Rob

I am reducing this from a "BLOCKER" only because there is a simple
'workaround' (and no 'fix') that allows one to easily continue the
build without any modification to the source. Lowered to "CRITICAL".

Rob

Same problem on my x86_64-linux-gnu system.

Err, libjava/classpath/tools/gnu/classpath/tools/jar/Creator.java is just
complete bullshit.  Citing:

  protected Manifest createManifest(Main parameters)
    throws IOException
  {
    if (! parameters.wantManifest)
      return null;
    if (parameters.manifestFile != null)
      {
        // User specified a manifest file.
        InputStream contents = new FileInputStream(parameters.manifestFile);
        return new Manifest(contents);
      }
    return new Manifest();
  }

  protected void writeCommandLineEntries(Main parameters, OutputStream os)
    throws IOException
  {
    manifest = createManifest(parameters);
    /* If no version is specified, provide the same manifest version default
     * as Sun's jar tool */
    Attributes attr = manifest.getMainAttributes();

WTF do you expect to happen when createManifest returns NULL?  Well, a crash.

Simpler testcase:

> gjar-4.4 -cfM -
jar: internal error:
java.lang.NullPointerException
   at gnu.classpath.tools.jar.Creator.writeCommandLineEntries(Creator.java:219)
   at gnu.classpath.tools.jar.Creator.run(Creator.java:244)
   at gnu.classpath.tools.jar.Main.run(Main.java:276)
   at gnu.classpath.tools.jar.Main.main(Main.java:284)


Diff between what is in gcc 4.3 and what is in gcc 4.4:

@@ -222,9 +214,15 @@
     throws IOException
   {
     manifest = createManifest(parameters);
+    /* If no version is specified, provide the same manifest version default
+     * as Sun's jar tool */
+    Attributes attr = manifest.getMainAttributes();
+    if (attr.getValue(Attributes.Name.MANIFEST_VERSION) == null)
+      attr.putValue(Attributes.Name.MANIFEST_VERSION.toString(), "1.0");
+    attr.putValue("Created-By", System.getProperty("java.version") +
+                 " (" + System.getProperty("java.vendor") + ")");


oh, really?  Gosh.

So, who owns that brown paperbag?

Same code in GCC trunk.  This is a major regression to classpath 0.96.

I guess a simple

 if (parameters.manifestFile)
   {

   }

around that offending change would fix it.  But the way classpath is handled
inside the gcc tree is very unfriendly to simple changes.  Tom, can you
get this fixed and installed on the 4.4 branch and trunk?

This kind of fix works for me.

(In reply to comment #7)
> This kind of fix works for me.

I'll leave it (java) to the Experts, so if you like it, commit it.


The only useful comment I could make is that I filed this on "2009-02-13".

I described it as a "Blocker" or "Critical" and you describe this as a 
"major regression"; some eights months later we are here ...

Thanks for being the one to fix it,
Rob

*** Bug 41768 has been marked as a duplicate of this bug. ***

Looks like classpath development is dead.  I have a set of files I can drop
into the GCC repositories but am not prepared to do a proper classpath update
(nor doing the upstream fix).  openSUSE packages now have this fixed.

Well, you can at least submit the same patch to classpath-patches, if you haven't done so already.  Maybe just Cc: classpath-patches when you submit the patch to java-patches.

Subject: Re:  trunk revision 144128 - jar: internal error:
 java.lang.NullPointerException

On Wed, 21 Oct 2009, aph at gcc dot gnu dot org wrote:

> ------- Comment #11 from aph at gcc dot gnu dot org  2009-10-21 12:38 -------
> Well, you can at least submit the same patch to classpath-patches, if you
> haven't done so already.  Maybe just Cc: classpath-patches when you submit the
> patch to java-patches.

Done.

Oh, it's moderated.  Well, so much for the fish.

From the discussion it sounds like this isn't going to make it into a released gcc any time soon.  Any chance someone could upload a set of patches that I can apply locally?

Subject: Bug 39177

Author: rguenth
Date: Thu Oct 22 15:41:32 2009
New Revision: 153461

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=153461
Log:
2009-10-22  Richard Guenther  <rguenther@suse.de>

        PR cp-tools/39177 * tools/gnu/classpath/tools/jar/Creator.java
        (writeCommandLineEntries):
        Do not use uninitialized manifest.
        * tools/classes/gnu/classpath/tools/jar/Creator.class: Re-generated.

Modified:
    trunk/libjava/classpath/ChangeLog
    trunk/libjava/classpath/tools/classes/gnu/classpath/tools/jar/Creator.class
    trunk/libjava/classpath/tools/gnu/classpath/tools/jar/Creator.java

Subject: Bug 39177

Author: rguenth
Date: Thu Oct 22 15:50:46 2009
New Revision: 153462

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=153462
Log:
2009-10-22  Richard Guenther  <rguenther@suse.de>

        PR cp-tools/39177
        * tools/gnu/classpath/tools/jar/Creator.java (writeCommandLineEntries):
        Do not use uninitialized manifest.
        * tools/classes/gnu/classpath/tools/jar/Creator.class: Re-generated.

Modified:
    branches/gcc-4_4-branch/libjava/classpath/ChangeLog
    branches/gcc-4_4-branch/libjava/classpath/tools/classes/gnu/classpath/tools/jar/Creator.class
    branches/gcc-4_4-branch/libjava/classpath/tools/gnu/classpath/tools/jar/Creator.java

Fixed in the GCC copy of classpath.

revision 1.9774
date: 2009-10-22 16:51:18 +0000;  author: aph;  state: Exp;  lines: +12 -0;  commitid: 5aOqqFT1hGw5gy8u;
2009-10-22  Andrew Haley  <aph@redhat.com>

        * native/jni/native-lib/cpnet.c (cpnet_addMembership): Fix
        aliasing warning.
        (cpnet_dropMembership): Likewise.

2009-10-21  Richard Guenther  <rguenther@suse.de>

        PR cp-tools/39177
        * tools/gnu/classpath/tools/jar/Creator.java (writeCommandLineEntries):
        Do not use uninitialized manifest.

*** Bug 42849 has been marked as a duplicate of this bug. ***

*** Bug 53400 has been marked as a duplicate of this bug. ***