Re: GCC Status Report (2004-03-09)

[Jakub Jelinek <jakub at redhat dot com>]

On Wed, Mar 31, 2004 at 09:54:48AM +0200, Eric Botcazou wrote:
> > Yes.  I believe I already approved this patch for 3.4, but Eric hasn't
> > applied it yet.  I'm not sure why not; perhaps I lost a message in the
> > thread somewhere.
> 
> Because I'm not convinced it is safe: it may potentially bring back problems 
> that have been hidden by not putting the /u flag on non-static aggregates any 
> more.  See my last message in the thread on the main mailing-list.

But current status quo is not safe either and the hack to kill /u for
non-static aggregates was incomplete (did so only if they were const as
whole, not non-const with all const fields or only some const fields).
So I think this needs some solution for 3.4.

One could be to only write into /u MEMs for non-aggregates, for aggregates
write into them using non-/u MEMs, then put a barrier and then use /u
for the rest of the function (and of course read with /u everywhere in
functions which don't initialize it).

	Jakub