On Mon, Aug 05, 2002 at 12:17:00PM -0700, Geoff Keating wrote:
> The following testcase is miscompiled (in all GCCs since end of 1999).
> The problem is that associate_trees performs an optimization which
> can be done with PLUS_EXPR only (turning x + (-y) into x - y or
> -x + y into y - x) - changing say x | -y into x - y is definitely
> wrong. Bootstrapped and regression tested on i386-redhat-linux
> (together with the bitfield3.C fix where I had to add target keyword
> to the dg-do line to pass), ok to commit?
> 3.2 branch after 3.2 is out too?
This is OK for the mainline. I think it doesn't qualify for the
branch, since it's not a regression, but I'll let Mark decide.
It is a regression from 2.95.x and earlier.