Re: Bogus warning is perhaps from new alias analysis

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

  In message <199809060714.AAA03132@smtp.earthlink.net>you write:
  > >>>>> "Iain" == Iain McClatchie <iain@mcclatchie.com> writes:
  > 
  >     Iain> Mark,
  > 
  >     Iain> I think I've found a possible problem in the new alias
  >     Iain> analysis code.  The following routine emits an incorrect
  >     Iain> warning, which I think stems from the new alias analysis.
  > 
  > Actually, that's quite unlikely.  At the moment, EGCS is spitting out
  > these `uinitialized variable' warnings much, much too often;
  > apparently the new GCSE pass has something to do with all this.
Agreed.  It is highly unlikely that the new alias code is the cause
of these problems.

The code to detect potentially uninitialized variables is somewhat dumb
and gets confused by certain control and data flow constructs.

In the process of optimizing code the new global cse code tends to
create those construts regularly.

The code to detect uninitialized variables needs some serious work.

jeff