gcj: verification bug when exception range reaches method end

[Gregg Townsend <gmt@baskerville.CS.Arizona.EDU>]

Gcj incorrectly reports "bad pc in exception table" if an exception
range extends all the way to the end of the method code:

    p_l$rfib$main.java: In class `p_l$rfib$main':
    p_l$rfib$main.java: In method `Call(jcon.vDescriptor)':
    p_l$rfib$main.java:0: bad pc in exception_table

The problem is that verify_jvm_instructions() is checking the (nonexistent)
instruction beyond the end of the range for validity.  Below is a patch.

   ---------------------------------------------------------------------------
   Gregg Townsend         Staff Scientist      The University of Arizona
   gmt@cs.arizona.edu     Computer Science     Tucson, Arizona, USA



*** verify.old	Mon Aug 23 17:53:47 1999
--- verify.c	Mon Aug 23 17:58:19 1999
*************** verify_jvm_instructions (jcf, byte_ops, 
*** 397,403 ****
  	  || handler_pc < 0 || handler_pc >= length
  	  || (handler_pc >= start_pc && handler_pc < end_pc)
  	  || ! (instruction_bits [start_pc] & BCODE_INSTRUCTION_START)
! 	  || ! (instruction_bits [end_pc] & BCODE_INSTRUCTION_START)
  	  || ! (instruction_bits [handler_pc] & BCODE_INSTRUCTION_START))
  	{
  	  error ("bad pc in exception_table");
--- 397,404 ----
  	  || handler_pc < 0 || handler_pc >= length
  	  || (handler_pc >= start_pc && handler_pc < end_pc)
  	  || ! (instruction_bits [start_pc] & BCODE_INSTRUCTION_START)
! 	  || (end_pc < length &&
! 	     ! (instruction_bits [end_pc] & BCODE_INSTRUCTION_START))
  	  || ! (instruction_bits [handler_pc] & BCODE_INSTRUCTION_START))
  	{
  	  error ("bad pc in exception_table");