Re: "[PATCH] take 2: cc1/f771 array overrun - x86 Floating Point"

[Jeffrey A Law <law at cygnus dot com>]

  In message <309F4FC4705DC844987051A517E9E39BF5BCB5@red-pt-02.redmond.corp.mic
rosoft.com>you write:
  > This is take 2, differing from the prior version by the use of
  > ENABLE_CHECKING and an abort call, per Jeff's suggestion.
  > The code continues to fail soft when checking is disabled, since 
  > there appears to be no downside to it, and the possible SIGSEGV 
  > is avoided.
I don't like the code that fails soft -- that just encourages people to
ignore the problem by hacking around the real bug and I can envision
ways to make your fail-soft fail in other strange and unusual ways.

I'd rather put the entire check in an ENABLE_CHECKING block, which also has
the benefit of formatting correctly :-)  Also note that ENABLE_CHECKING
is on by default in the development trees, so it's likely people will
start triggering this failure immediately -- I doubt the bug will survive
very long.

Again, I think the way to fix this is to have init_insn_lengths be called
again after reg-stack has finished.

You always need a space before the open paren for an argument list, you
missed it in a couple places in your patch.

We don't refer to function names in comments with "()" -- just use the
name of the function.  Actually, since we're putting in a generic sanity
test, I think the comment is better worded as

/* This can be caused by bugs elsewhere in the compiler if new insns are 
   created after init_insn_lengths is called.  */

Here's the patch I'm checking in.

Index: final.c
===================================================================
RCS file: /cvs/gcc/egcs/gcc/final.c,v
retrieving revision 1.134
diff -c -3 -p -r1.134 final.c
*** final.c	2000/06/13 16:06:26	1.134
--- final.c	2000/07/07 13:31:59
*************** final (first, file, optimize, prescan)
*** 2014,2019 ****
--- 2014,2025 ----
    for (insn = NEXT_INSN (first); insn;)
      {
  #ifdef HAVE_ATTR_length
+ #ifdef ENABLE_CHECKING
+       /* This can be triggered by bugs elsewhere in the compiler if
+ 	 new insns are created after init_insn_lengths is called.  */
+       if (INSN_UID (insn) >= insn_lengths_max_uid)
+ 	abort ();
+ #endif
        insn_current_address = insn_addresses[INSN_UID (insn)];
  #endif
        insn = final_scan_insn (insn, file, optimize, prescan, 0);