PR java/25535: gcj broken on 64-bit big-endian systems

[Andrew Haley]

Tom Tromey writes:
 > >>>>> "Andrew" == Andrew Haley <aph@redhat.com> writes:
 > 
 > Andrew> BC-compiled code doesn't work on a big-endian 64-bit box
 > Andrew> because of a silly mistake initializing entries in the
 > Andrew> constant pool.  In all cases the initializer must appear as
 > Andrew> the first 32-bit integer of a machine word stored in memory.
 > 
 > Andrew> -	  jword temp = outgoing_cpool->data[i].w;
 > Andrew> +	  unsigned HOST_WIDE_INT temp = outgoing_cpool->data[i].w;
 > Andrew> +
 > Andrew> +	  /* Make sure that on a 64-bit big-endian machine this 32-bit
 > Andrew> +	     jint appears in the first word.  */
 > Andrew> +	  if (BYTES_BIG_ENDIAN && BITS_PER_WORD > 32)
 > Andrew> +	    temp <<= BITS_PER_WORD - 32;
 >  
 > This is related to PR 18148 -- internally gcj has the wrong type for
 > itable and some other fields of Class.  It seems like it would be more
 > robust (though more painful) to have the proper declarations inside
 > gcj and to generate a real union constructor here.

It can't be done: AFAIK a constructor for a union can only initialize
the first element of the union.

Andrew.