This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: C++ PATCH to fix ICE in strip_typedefs with -Wpadded (PR c++/79900)
- From: Jason Merrill <jason at redhat dot com>
- To: Marek Polacek <polacek at redhat dot com>
- Cc: Jakub Jelinek <jakub at redhat dot com>, GCC Patches <gcc-patches at gcc dot gnu dot org>
- Date: Thu, 9 Mar 2017 17:35:06 -0500
- Subject: Re: C++ PATCH to fix ICE in strip_typedefs with -Wpadded (PR c++/79900)
- Authentication-results: sourceware.org; auth=none
- References: <20170308230048.GA3172@redhat.com> <20170309210653.GK22703@tucnak> <20170309223231.GH3172@redhat.com>
Yep, I just checked in that change.
On Thu, Mar 9, 2017 at 5:32 PM, Marek Polacek <polacek@redhat.com> wrote:
> On Thu, Mar 09, 2017 at 10:06:53PM +0100, Jakub Jelinek wrote:
>> On Thu, Mar 09, 2017 at 12:00:48AM +0100, Marek Polacek wrote:
>> > We crash on an assert in strip_typedefs, because we find ourselves in a
>> > scenario where RESULT, the main variant of a struct, was modified in
>> > finalize_record_size (its TYPE_ALIGN changed), but its variants (T in
>> > strip_typedefs) weren't fixed-up yet; that will happen slightly later in
>> > finalize_type_size. So there's a discrepancy that confuses the code.
>> >
>> > This patch implements what Jason suggested to me on IRC, i.e., skip the
>> > attribute handling if RESULT is complete and T isn't.
>> >
>> > Bootstrapped/regtested on x86_64-linux, ok for trunk?
>> >
>> > 2017-03-08 Marek Polacek <polacek@redhat.com>
>> >
>> > PR c++/79900 - ICE in strip_typedefs
>> > * tree.c (strip_typedefs): Skip the attribute handling if T is
>> > a variant type which hasn't been updated yet.
>> >
>> > * g++.dg/warn/Wpadded-1.C: New test.
>> >
>> > diff --git gcc/testsuite/g++.dg/warn/Wpadded-1.C gcc/testsuite/g++.dg/warn/Wpadded-1.C
>> > index e69de29..b3f0581 100644
>> > --- gcc/testsuite/g++.dg/warn/Wpadded-1.C
>> > +++ gcc/testsuite/g++.dg/warn/Wpadded-1.C
>> > @@ -0,0 +1,22 @@
>> > +// PR c++/79900 - ICE in strip_typedefs
>> > +// { dg-do compile }
>> > +// { dg-options "-Wpadded" }
>> > +
>> > +template <class> struct A;
>> > +template <typename> struct B { // { dg-warning "padding struct size to alignment boundary" }
>> > + long _M_off;
>> > + int _M_state;
>> > +};
>>
>> This fails on i686-linux, there is obviously no padding of struct size to
>> alignment boundary in that case.
>
> Ah, of course...
>
>> Can't _M_state be e.g. char, then it would work at least on all targets
>> where long has alignment of at least 2 (that can be all of them)?
>> Perhaps also make _M_off long long or add a double or long double field.
>
> Like this? Passes with -m32.
>
> 2017-03-09 Marek Polacek <polacek@redhat.com>
>
> * g++.dg/warn/Wpadded-1.C: Change types of struct B fields.
>
> diff --git gcc/testsuite/g++.dg/warn/Wpadded-1.C gcc/testsuite/g++.dg/warn/Wpadded-1.C
> index b3f0581..af375a4 100644
> --- gcc/testsuite/g++.dg/warn/Wpadded-1.C
> +++ gcc/testsuite/g++.dg/warn/Wpadded-1.C
> @@ -4,8 +4,8 @@
>
> template <class> struct A;
> template <typename> struct B { // { dg-warning "padding struct size to alignment boundary" }
> - long _M_off;
> - int _M_state;
> + long long _M_off;
> + char _M_state;
> };
> template <> struct A<char> { typedef B<int> pos_type; };
> enum _Ios_Openmode {};
>
> Marek