Re: sparc-sun-solaris2.6 egcs-1.1.2 post increment bug

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

  In message <Pine.GSO.4.10.9907270851480.16877-100000@re.vss.fsi.com>you write
:
  > Ok.  I understand no warning needs to be posted, especially in light of
  > the eminent gcc-2.95 release that fixes the bug.  I think a silent codegen
  > bug for something as simple as pre and post increment *is* pretty big.
  > But, I didn't mean to sound as if I was that *bothered* by it.
Nearly every bug looks simple once you understand it.  I can trigger a
multitude of bugs in every compiler ever used with testcases that are
just a few lines long.

This particular bug only triggers on targets which use the mem* functions
to implement block moves and to clear blocks of memory.  It only happens when
we have one of those block moves (particularly for structure copies) that
proceed a function call and the addresses are auto-inc'd within the call to
the mem* function.

There's a couple other things that need to happen for this bug to trigger, but
those are the key elements.

Is it a bad bug.  Certainly.  Is it the end of the world?  Not really.  If we
were not just about to release gcc-2.95 I would have considered an egcs-1.1.3
release which would have fixed this problem and likely a few others.

  > I think egcs, now gcc, is a very good compiler and I applaud all the
  > volunteer effort that is put in.  I hope, though, that there is a test
  > case in the testsuite for something as basic as this.  It seems like an
  > embarrasing bug to let out the door.  I hope it was not known at release
  > time, or if it was, I hope it was going to be a bear to fix.
Yes, we added a testcase to the regression tests once a proper bugs was
submitted.  No, this bug was not known at release time for egcs-1.1.2.  In 
fact, it was introduced in an attempt to fix a long standing bug in every
version of gcc since about 1991 that was causing a multitude of problems.


jeff