This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH] Enable type verification throughout the compiler.
On Tue, 28 Apr 2009, Dave Korn wrote:
> Richard Guenther wrote:
> > On Tue, 28 Apr 2009, Dave Korn wrote:
> >
> >> H.J. Lu wrote:
> >>> On Mon, Apr 27, 2009 at 10:24 AM, H.J. Lu <hjl.tools@gmail.com> wrote:
> >>>> On Mon, Apr 27, 2009 at 4:15 AM, Richard Guenther <rguenther@suse.de> wrote:
> >>>>> Tromey agreed to my Java fix on IRC after some discussion, Diego acked
> >>>>> the patch in principle.
> >>>>>
> >>>>> Thus, re-bootstrapped and tested on x86_64-unknown-linux-gnu with all
> >>>>> languages and -m32/-m64 testing, installed to trunk.
> >>>>>
> >>>>> Richard.
> >>>>>
> >>>>> 2009-04-27 Richard Guenther <rguenther@suse.de>
> >>>>>
> >>>>> * tree-cfg.c (remove_useless_stmts): Verify stmts afterwards.
> >>>>> (verify_stmts): Dispatch to gimple/type verification code.
> >>>>> * tree-inline.c (remap_gimple_op_r): Work around C++ FE
> >>>>> issue with call argument types.
> >>>>>
> >>>>> java/
> >>>>> PR java/38374
> >>>>> * constants.c (build_constants_constructor): Retain the old
> >>>>> pointer type as valid TYPE_POINTER_TO after patching the
> >>>>> type of the constant pool decl.
> >>>>>
> >>>> This patch caused:
> >>>>
> >>>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39932
> >>>>
> >>> This also caused:
> >>>
> >>> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39937
> >>>
> >> And a java bootstrap failure for the same reason ("type mismatch"). I
> >> posted the details on the tail of the original bug:
> >>
> >> http://gcc.gnu.org/bugzilla/show_bug.cgi?id=38374
> >>
> >> and I think it's almost certainly the same problem underlying. Verified that
> >> reverting r146831 locally fixed it.
> >
> > Same underlying problem as in "wrong-code" in the intermediate language.
> >
> > r146831 simply uncovered all these cases by verifying internal
> > consistency of GIMPLE at more places.
>
> Ok, the java FE is emitting invalid GIMPLE during the libjava build. Argh.
> I'm not in a great position to investigate this immediately because I'm
> struggling with a couple of years' bit-rot in the win32 ports as it is. Once
> I can get it up and running reasonably well with the patch reverted, I'll come
> back to this and get some tree dumps and file a PR with more info, if it's not
> already resolved by then.
Note that you can configure with --disable-checking or
--enable-checking=assert,gc,gimple,rtlflag,runtime,tree (yes w/o types)
Richard.