Hi,
On Tue, 26 Jun 2012, Jason Merrill wrote:
> > (the gimplifier code should be in Frontend code that lowers to GENERIC
> > really and the WITH_CLEANUP_EXPR code should be C++ frontend specific
> > ...).
>
> TARGET_EXPR has been a back-end code since the dawn of GCC version
> control; if it's still only used by the C++ front end I guess it could
> move to cp-tree.def, but we can't really lower it until gimplification
> time because we need to strip away enclosing COMPOUND_EXPRs and such so
> we can see that it's on the RHS of an initialization and optimize away
> the temporary in that case. And now that GIMPLE isn't a subset of
> GENERIC, we can't just use the gimplifier at genericization time. And
> I'd rather not duplicate the entire gimplifier in the front end.
I agree with Jason. TARGET_EXPR and CLEANUP_POINT_EXPR might currently be
used only for C++, but I think they are sensible general constructs to be
supported by the gimplifier.
But I also think that the option to disable stack slot sharing should be
moved to cfgexpand to trigger non-sharing of everything, not just these
cleanup temporaries. After all using the (c++)temporary after expression
end is a source bug that the option is supposed to work around, just like
this is:
char *p; { char str[50]; p = str; } use(p);
So, IMO the option should also work around this source bug. We had at
least one example of that in our own code base.
Ciao,
Michael.
