Re: jni fixlet

[Marco Trudel <mtrudel at gmx dot ch>]

David Daney wrote:
> Marco Trudel wrote:
> > David Daney wrote:
> > > Marco Trudel wrote:
> > > > Attached a jni fixlet.
> > > >
> > > >
> > > > 2007-02-01 Marco Trudel <mtrudel@gmx.ch>
> > > >
> > > > * jni.cc (_Jv_JNI_DeleteWeakGlobalRef): Check for NULL objects.
> > > >
> > > >
> > > > Question? Comments?
> > >
> > > What was the undesirable behavior you observed that this patch fixes?
> >
> > A compiled application crashed with "Aborted". A native lib called 
> > "_Jv_JNI_DeleteWeakGlobalRef" with a null obj what in the end causes a 
> > segmentation fault in unmark_for_gc:
> >     jint val = refcount->intValue() - 1; (refcount null)
> >
> >
> > > How did you test it?
> >
> > I played a little with gdb, saw the segfault, added a check, compiled 
> > my app again and was happy that it worked.
> > Ah, and I then ran the library I use which does this null call in a 
> > Sun JVM to ensure that there _Jv_JNI_DeleteWeakGlobalRef is also 
> > called with NULL. It is...
>
> Usually you whould run the libgcj testsuite before and after the change. 
>  That would show that the patch does not affect any existing test 
> cases.  Optionally you could add a test case for this.
>
> It looks safe, but you can never tell until you run the tests.

Yes, you're right. I'll do that and repost once I have time...


Marco