Re: ppc glibc 2.3 failure

[Richard Henderson <rth at redhat dot com>]

On Mon, Oct 07, 2002 at 06:09:31PM -0700, Dale Johannesen wrote:
> Well, it's "right" in the sense that it works.

It is not "right", however, in the sense that a prologue instruction
gets marked REG_MAYBE_DEAD, when the instruction is not in fact dead.
Thus bugs elsewhere in post-reload optimizers may cause flow2 to
delete the instruction instead of aborting as it ought.

A correct solution that retains the current structure would use
prologue_epilogue_contains instead of REG_MAYBE_DEAD.  Of course,
that's pointless since we can just avoid recomputing this information
in the first place.  :-)


r~