Re: Incorrect Java methods called from CNI

[Andrew Haley <aph at redhat dot com>]

Bryce McKinlay writes:
 > Dave Menendez wrote:
 > 
 > >	movl	(%eax), %edx
 > >	addl	$56, %edx
 > >	movl	-12(%ebp), %eax
 > >	movl	%eax, (%esp)
 > >	movl	(%edx), %eax
 > >	call	*%eax
 > >
 > >  
 > >
 > Looks like this is the call in question. Its offseting the vtable 
 > pointer by 56 bytes.
 > 
 > Heres the vtable, run through c++filt, cleaned up a bit, and with byte 
 > offsets added:
 > 
 > -8    .long    0
 > -4    .long    0
 >  0    .long    Classification.class$
 >  4    .long    -37748735
 >  8    .long    java.lang.Object.finalize()
 >  12    .long    hashCode()
 >  16    .long    equals(java.lang.Object)
 >  20    .long    toString()
 >  24    .long    clone()
 >  28    .long    addClassificationListener(ClassificationListener)
 >  32    .long    fireClassificationChange()
 >  36    .long    getNumberOfListeners()
 >  40    .long    isImmutable()
 >  44    .long    removeAllListeners()
 >  48    .long    removeClassificationListener(ClassificationListener)
 >  52    .long    getClassification()
 >  56    .long    add(Marking)
 >  60    .long    clear()
 >  64    .long    clearMarkings()
 >  68    .long    combine(Classification)
 >  72    .long    compareTo(Classification)
 >  76    .long    contains(java.lang.Object)
 > 
 > So, its obviously going to call add() - the offset to call clear() 
 > should be 60. For some reason, the C++ compiler's idea of where the 
 > method is in the vtable doesn't match the Java compiler's. Can you post 
 > the relavent portions of the .h file for this class, and the method 
 > declarations from the Java source code (not the code itself, just the 
 > name & access modifiers etc)?

Another possibility is that the superclass .h file for
AbstractClassificationModel is not correct.  g++ -save-temps will tell
us, one way or another.

Andrew.