[Bug c++/106369] New: ICE in output_constructor_regular_field, at varasm.cc:5515

[Bug c++/106369] New: ICE in output_constructor_regular_field, at varasm.cc:5515

[h2+bugs at fsfe dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

            Bug ID: 106369
           Summary: ICE in output_constructor_regular_field, at
                    varasm.cc:5515
           Product: gcc
           Version: 12.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: h2+bugs at fsfe dot org
  Target Milestone: ---

Created attachment 53322
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53322&action=edit
backtrace

Also reproducible with GCC-10.3 (varasm.c:5254).

I don't know if this is a duplicate of
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103285 since that is in the
fortran component.

[Bug c++/106369] ICE in output_constructor_regular_field, at varasm.cc:5515

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2022-07-20
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |needs-reduction

--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.

[Bug c++/106369] ICE in output_constructor_regular_field, at varasm.cc:5515

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Marek Polacek <mpolacek at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jakub at gcc dot gnu.org,
                   |                            |mpolacek at gcc dot gnu.org

--- Comment #2 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
I think this started with r12-2975-g32c3a75390623a.

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|ICE in                      |[12/13 Regression] ICE in
                   |output_constructor_regular_ |output_constructor_regular_
                   |field, at varasm.cc:5515    |field, at varasm.cc:5515
   Target Milestone|---                         |12.2

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
reducing ...

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 53327
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=53327&action=edit
reduced testcase

Note aminoacid_empty_base is important to hit the bug.

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515

[h2+bugs at fsfe dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #5 from Hannes Hauswedell <h2+bugs at fsfe dot org> ---
I have found a workaround for my code:

template <typename derived_type, auto size>
-class aminoacid_base : public alphabet_base<derived_type, size, char>, public
aminoacid_empty_base
+class aminoacid_base : public aminoacid_empty_base, public
alphabet_base<derived_type, size, char>

This change prevents the bug from triggering. Maybe that helps in pinning it
down?

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Hannes Hauswedell from comment #5)

> This change prevents the bug from triggering. Maybe that helps in pinning it
> down?

I suspect it was caused by this part of the patch:
* class.c (remove_zero_width_bit_fields): Remove.
(layout_class_type): Don't call it.

With the empty class being there, there is an "zero" width field.

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |10.4.0, 11.3.0
           Keywords|                            |ice-on-valid-code
                 CC|                            |rguenth at gcc dot gnu.org

--- Comment #7 from Richard Biener <rguenth at gcc dot gnu.org> ---
Hmm, I can reproduce with the reduced testcase and GCC 10+ while GCC 9 dind't
implement enough of C++20 to accept the code.

So is this really a regression?  What's a working compiler version?

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Patrick Palka <ppalka at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |ppalka at gcc dot gnu.org

--- Comment #8 from Patrick Palka <ppalka at gcc dot gnu.org> ---
FWIW I noticed that if we remove the 'constexpr' from aa27::aa27() in the
reduced testcase then the ICE starts with r9-2103-g686dca7612a8ab before which
we accepted the code with -std=c++2a.

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[rguenther at suse dot de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #9 from rguenther at suse dot de <rguenther at suse dot de> ---
On Mon, 25 Jul 2022, ppalka at gcc dot gnu.org wrote:

> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369
> 
> Patrick Palka <ppalka at gcc dot gnu.org> changed:
> 
>            What    |Removed                     |Added
> ----------------------------------------------------------------------------
>                  CC|                            |ppalka at gcc dot gnu.org
> 
> --- Comment #8 from Patrick Palka <ppalka at gcc dot gnu.org> ---
> FWIW I noticed that if we remove the 'constexpr' from aa27::aa27() in the
> reduced testcase then the ICE starts with r9-2103-g686dca7612a8ab before which
> we accepted the code with -std=c++2a.

Is that maybe a different bug then?

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Slightly further reduced:

struct A {
  int a[256];
  constexpr int &operator[] (int n) noexcept { return a[n]; }
  constexpr const int &operator[] (int n) const noexcept { return a[n]; }
};
struct B {};
template <typename T>
struct C {
  constexpr T &foo (const char x) noexcept { c = T::d[x]; return static_cast<T
&>(*this); }
  int c;
};
struct D : public C<D>, public B
{
  constexpr D () noexcept = default;
  static constexpr char e[9] { 'A', 'B', 'C', 'D', 'E', 'F', 'G', 'H', 'I' };
  static constexpr A d = [] () constexpr {
    A f {};
    for (int i = 0; i < 9; ++i)
      f[e[i]] = 1;
    return f;
  } ();
};
constexpr auto g = D{}.foo ('E');

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #11 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
And that testcase has been rejected before r10-5194-g7906797ebec6881d and ICEs
since that revision.

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[jason at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |jason at gcc dot gnu.org
                 CC|                            |jason at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

[Bug c++/106369] [12/13 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #12 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Jason Merrill <jason@gcc.gnu.org>:

https://gcc.gnu.org/g:9efe4e153d994974afcbba09c3c683f5f4a19c63

commit r13-1901-g9efe4e153d994974afcbba09c3c683f5f4a19c63
Author: Jason Merrill <jason@redhat.com>
Date:   Tue Jul 26 11:02:21 2022 -0400

    c++: constexpr, empty base after non-empty [PR106369]

    Here the CONSTRUCTOR we were providing for D{} had an entry for the B base
    subobject at offset 0 following the entry for the C base, causing
    output_constructor_regular_field to ICE due to going backwards.  It might
be
    nice for that function to be more tolerant of empty fields, but it also
    seems reasonable for the front end to prune the useless entry.

            PR c++/106369

    gcc/cp/ChangeLog:

            * constexpr.cc (reduced_constant_expression_p): Return false
            if a CONSTRUCTOR initializes an empty field.

    gcc/testsuite/ChangeLog:

            * g++.dg/cpp1z/constexpr-lambda27.C: New test.

[Bug c++/106369] [12 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2
      Known to work|                            |13.0
      Known to fail|                            |12.1.0
            Summary|[12/13 Regression] ICE in   |[12 Regression] ICE in
                   |output_constructor_regular_ |output_constructor_regular_
                   |field, at varasm.cc:5515    |field, at varasm.cc:5515
                   |since                       |since
                   |r12-2975-g32c3a75390623a    |r12-2975-g32c3a75390623a

--- Comment #13 from Richard Biener <rguenth at gcc dot gnu.org> ---
Good for 12.2?

[Bug c++/106369] [12 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

--- Comment #14 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Richard Biener
<rguenth@gcc.gnu.org>:

https://gcc.gnu.org/g:6b7d570a5001bb79e34c0d1626a8c7f55386dac7

commit r12-8684-g6b7d570a5001bb79e34c0d1626a8c7f55386dac7
Author: Jason Merrill <jason@redhat.com>
Date:   Tue Jul 26 11:02:21 2022 -0400

    c++: constexpr, empty base after non-empty [PR106369]

    Here the CONSTRUCTOR we were providing for D{} had an entry for the B base
    subobject at offset 0 following the entry for the C base, causing
    output_constructor_regular_field to ICE due to going backwards.  It might
be
    nice for that function to be more tolerant of empty fields, but it also
    seems reasonable for the front end to prune the useless entry.

            PR c++/106369

    gcc/cp/ChangeLog:

            * constexpr.cc (reduced_constant_expression_p): Return false
            if a CONSTRUCTOR initializes an empty field.

    gcc/testsuite/ChangeLog:

            * g++.dg/cpp1z/constexpr-lambda27.C: New test.

    (cherry picked from commit 9efe4e153d994974afcbba09c3c683f5f4a19c63)

[Bug c++/106369] [12 Regression] ICE in output_constructor_regular_field, at varasm.cc:5515 since r12-2975-g32c3a75390623a

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=106369

Richard Biener <rguenth at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED
      Known to work|                            |12.1.1

--- Comment #15 from Richard Biener <rguenth at gcc dot gnu.org> ---
Fixed.