-g -O debug-info problem (on x86)

[Jeffrey A Law]

  In message < 19990517045720.21479.qmail@deer >you write:
  > I've been noticing a tendency that, when debugging, the program being
  > debugged can "slide" through the first executable statement after
  > hitting a breakpoint supposedly put on the *entry* to a routine, such
  > as main().
  > 
  > This is apparently due to some internal reorganizing of RTL (?) within
  > egcs (using the 19990502 snapshot) when optimizing.
  > 
  > Here's a sample test program:
[ ... ]
Thanks.  I believe the problem is either the placement of the prologue
or something in flow2 reordering the location of the prologue relative
to the first line note.

The former is more likely, though I don't see from a quick look how it
could be happening.  Maybe there's something different about the CFG
we produce when not optimizing that is causing us to place the insns
for the prologue in the wrong location.  rth might be the best person
to look at this.

jeff