This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Clean up the gimple type verifier [1/2]
On Fri, 5 Sep 2008, Andrew Haley wrote:
> Richard Guenther wrote:
> > On Fri, 5 Sep 2008, Andrew Haley wrote:
> >
> >> Richard Guenther wrote:
> >>> On Fri, 5 Sep 2008, Andrew Haley wrote:
> >>>
> >>>> Tom Tromey wrote:
> >>>>>>>>>> "Richard" == Richard Guenther <rguenther@suse.de> writes:
> >>>>> Richard> This is a re-post with the changes requested by Diego. Tom - is the
> >>>>> Richard> Java part ok? The cases in the C/C++ frontend it refers to have been
> >>>>> Richard> removed one and a half years ago.
> >>>>>
> >>>>> It seems ok to me, but I'd like it if Andrew would look at this.
> >>>>> Andrew, I quoted the patch in question at the end.
> >>>> Hmm. I have no idea whether this will break binary compatibility.
> >>>> We'd have to test i.
> >>> Why should it? It simply causes a function that returns bool (QImode)
> >>> to have a RESULT_DECL of type bool (QImode) instead of int (SImode).
> >>> Expansion ensures that proper widening is done if the ABI requires it.
> >>>
> >>> The patch ensures that on the tree level if we for example inline
> >>>
> >>> bool foo() { return true; }
> >>>
> >>> we do not get an assignment from int to bool, but from bool to bool.
> >> I understand what you're saying, but I thought the old behaviour was to
> >> widen the return value. And some existing CNI code might depend on that.
> >
> > FYI the similar treatment in the C++ and C frontends have been removed
> > with
> >
> > 2007-07-09 Richard Guenther <rguenther@suse.de>
> >
> > * decl.c (start_preparsed_function): Do not promote return type.
> >
> > 2007-07-09 Richard Guenther <rguenther@suse.de>
> >
> > * c-decl.c (start_function): Do not promote return type.
>
> I remember, and at the time I complained it broke the ABI, but I was
> told "the back ends should do it instead." To the best of my
> knowledge no-one did anything to make the back ends do this where
> required, though.
Right, because the ABI doesn't require it which is at least what
the consensus was on the ABI mailing list. I remember libffi
was "fixed" somehow though.
Ok, if we are not going to "fix" the Java frontend I'll have to disable
this part of the GIMPLE typeckecking.
Richard.