[Bug tree-optimization/94775] New: [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734

[Bug tree-optimization/94775] New: [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734

[zhroma at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

            Bug ID: 94775
           Summary: [8/9/10 Regression] ICE in strip_typedefs, at
                    cp/tree.c:1734
           Product: gcc
           Version: 10.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: tree-optimization
          Assignee: unassigned at gcc dot gnu.org
          Reporter: zhroma at gcc dot gnu.org
  Target Milestone: ---

At least ubuntu v8.3.0 and trunk v10.0.1 (20200402) ICE on Linux x86-64 with
the following reduced testcase while emitting diagnostic:

$ cat bug.ii
template <typename> using a = int;
template <bool, typename, typename> using b = int;
typedef char d;
template <long> using e = int;
template <int f, int q> struct h { using i = b<q, a<e<f>>, e<f>>; };
template <long f, bool g> using j = typename h<f, g>::i;
long ab, k, aj;
const d l[]{};
class m {
public:
  m(int);
};
class n {
  m ad() const;
  template <class ae> void o(long) const {
    using c __attribute__((aligned(1))) = const ae;
  }
  long p;
  template <class, class>
  auto s(unsigned long, unsigned long, unsigned long, unsigned long) const;
  template <bool = false> auto q(unsigned long, unsigned long) const;
};
template <class, class>
auto n::s(unsigned long, unsigned long, unsigned long, unsigned long t) const {
  o<d>(p);
  return t;
}
template <bool g> auto n::q(unsigned long p1, unsigned long p2) const {
  using r = j<4, false>;
  using ai = j<4, g>;
  return s<ai, r>(ab, k, p1, p2);
}
m n::ad() const {
  long f(l[aj]);
  m(q(8, f));
}

$ g++ -std=c++17 -Wall -O2 -c -fsanitize=address bug.ii
bug.ii: In member function 'void n::o(long int) const':
bug.ii:16:11: warning: typedef 'using c = const ae' locally defined but not
used [-Wunused-local-typedefs]
   16 |     using c __attribute__((aligned(1))) = const ae;
      |           ^
bug.ii: In member function 'm n::ad() const':
bug.ii:36:1: warning: no return statement in function returning non-void
[-Wreturn-type]
   36 | }
      | ^
'
during GIMPLE pass: vrp
in strip_typedefs, at cp/tree.c:1734
   33 | m n::ad() const {
      |   ^
0x9b4eeb strip_typedefs(tree_node*, bool*, unsigned int)
...

7.5.0 seems to work fine.

[Bug tree-optimization/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734

[zhroma at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #1 from Roman Zhuykov <zhroma at gcc dot gnu.org> ---
Additionally checked that vanilla gcc-9.3.0 version is also affected.

[Bug tree-optimization/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

Martin Liška <marxin at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
           Keywords|                            |ice-on-valid-code
   Last reconfirmed|                            |2020-04-27
                 CC|                            |marxin at gcc dot gnu.org,
                   |                            |msebor at gcc dot gnu.org
      Known to work|                            |7.4.0
   Target Milestone|---                         |8.5
             Status|UNCONFIRMED                 |NEW
      Known to fail|                            |10.0, 8.3.0, 9.2.0
            Summary|[8/9/10 Regression] ICE in  |[8/9/10 Regression] ICE in
                   |strip_typedefs, at          |strip_typedefs, at
                   |cp/tree.c:1734              |cp/tree.c:1734 since
                   |                            |r8-4668-g8a5ee94a082b3d48

--- Comment #2 from Martin Liška <marxin at gcc dot gnu.org> ---
Confirmed, started with r8-4668-g8a5ee94a082b3d48.

[Bug tree-optimization/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Priority|P3                          |P2

[Bug tree-optimization/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #3 from Martin Sebor <msebor at gcc dot gnu.org> ---
The full stack trace shows the ICE occurs while formatting the text of a
warning.  strip_typedefs is being called while formatting the type of an array
of d and when assertion below fires:

      if (TYPE_USER_ALIGN (t) != TYPE_USER_ALIGN (result)
          || TYPE_ALIGN (t) != TYPE_ALIGN (result))
        {
          gcc_assert (TYPE_USER_ALIGN (t));
          if (remove_attributes)

during GIMPLE pass: vrp
in strip_typedefs, at cp/tree.c:1737
   41 | m n::ad() const {
      |   ^
0xc6cab1 strip_typedefs(tree_node*, bool*, unsigned int)
        /src/gcc/master/gcc/cp/tree.c:1737
0xa823d8 type_to_string
        /src/gcc/master/gcc/cp/error.c:3214
0xa861d6 cp_printer
        /src/gcc/master/gcc/cp/error.c:4289
0x268d13a pp_format(pretty_printer*, text_info*)
        /src/gcc/master/gcc/pretty-print.c:1458
0x2666327 diagnostic_report_diagnostic(diagnostic_context*, diagnostic_info*)
        /src/gcc/master/gcc/diagnostic.c:1159
0x266694b diagnostic_impl
        /src/gcc/master/gcc/diagnostic.c:1309
0x26671c7 warning_at(unsigned int, int, char const*, ...)
        /src/gcc/master/gcc/diagnostic.c:1446
0x19b82ae vrp_prop::check_array_ref(unsigned int, tree_node*, bool)
        /src/gcc/master/gcc/tree-vrp.c:3578
0x19ba715 check_array_bounds
        /src/gcc/master/gcc/tree-vrp.c:4062
0x19f1508 walk_tree_1(tree_node**, tree_node* (*)(tree_node**, int*, void*),
void*, hash_set<tree_node*, false, default_hash_traits<tree_node*> >*,
tree_node* (*)(tree_node**, int*, tree_node* (*)(tree_node**, int*, void*),
void*, hash_set<tree_node*, false, default_hash_traits<tree_node*> >*))
        /src/gcc/master/gcc/tree.c:12000
0x10ed4ac walk_gimple_op(gimple*, tree_node* (*)(tree_node**, int*, void*),
walk_stmt_info*)
        /src/gcc/master/gcc/gimple-walk.c:202
0x19ba871 check_array_bounds_dom_walker::before_dom_children(basic_block_def*)
        /src/gcc/master/gcc/tree-vrp.c:4120
0x23f5aed dom_walker::walk(basic_block_def*)
        /src/gcc/master/gcc/domwalk.c:309
0x19ba8d5 vrp_prop::check_all_array_refs()
        /src/gcc/master/gcc/tree-vrp.c:4137
0x19bdb47 vrp_prop::vrp_finalize(bool)
        /src/gcc/master/gcc/tree-vrp.c:5209
0x19bdbdc execute_vrp
        /src/gcc/master/gcc/tree-vrp.c:5277
0x19bdddb execute
        /src/gcc/master/gcc/tree-vrp.c:5359
Please submit a full bug report,
with preprocessed source if appropriate.
Please include the complete backtrace with any bug report.
See <https://gcc.gnu.org/bugs/> for instructions.

[Bug tree-optimization/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #4 from Martin Sebor <msebor at gcc dot gnu.org> ---
The assertion was introduced in a fix for PR79900 by the commit below:

g:423aec8b5756 1734)       if (TYPE_USER_ALIGN (t) != TYPE_USER_ALIGN (result)
g:423aec8b5756 1735)        || TYPE_ALIGN (t) != TYPE_ALIGN (result))
g:b71983a5baee 1736)      {
g:423aec8b5756 1737)        gcc_assert (TYPE_USER_ALIGN (t));
g:423aec8b5756 1738)        if (remove_attributes)

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[msebor at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

Martin Sebor <msebor at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |mpolacek at gcc dot gnu.org
          Component|tree-optimization           |c++

--- Comment #5 from Martin Sebor <msebor at gcc dot gnu.org> ---
I don't think the warning code is doing anything wrong here.  It seems like a
C++ front end problem.  CC'ng Marek for his input.

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #6 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
The TYPE_USER_ALIGN (t) assert was there even before my change, but I agree
this is a latent problem not directly caused by r8-4668-g8a5ee94a082b3d48.

Here strip_typedefs gets "const d[0]" with TYPE_USER_ALIGN=0.  But the result
built by
result = build_cplus_array_type (type, t0);
is "const char[0]" with TYPE_USER_ALIGN=1.

When we strip_typedefs the element of the array "const d", we see it's a
typedef_variant_p, so we look at its DECL_ORIGINAL_TYPE, which is char, but we
need to add the const qualifier, so we call
cp_build_qualified_type -> build_qualified_type
where get_qualified_type checks to see if we already have such a type by
walking the variants list, which in this case is:

char -> c -> const char -> const char -> d -> const d

Because check_base_type only checks TYPE_ALIGN and not TYPE_USER_ALIGN, we
choose the first const char, which has TYPE_USER_ALIGN set.  If the element
type of an array has TYPE_USER_ALIGN, the array type gets it too.  Thus the
discrepancy.

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #7 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
So a fix could be this, but maybe it would make us create a lot more variants
(?):

--- a/gcc/tree.c
+++ b/gcc/tree.c
@@ -6493,7 +6493,8 @@ check_base_type (const_tree cand, const_tree base)
                    TYPE_ATTRIBUTES (base)))
     return false;
   /* Check alignment.  */
-  if (TYPE_ALIGN (cand) == TYPE_ALIGN (base))
+  if (TYPE_ALIGN (cand) == TYPE_ALIGN (base)
+      && TYPE_USER_ALIGN (cand) == TYPE_USER_ALIGN (base))
     return true;
   /* Atomic types increase minimal alignment.  We must to do so as well
      or we get duplicated canonical types. See PR88686.  */
@@ -6528,6 +6529,7 @@ check_aligned_type (const_tree cand, const_tree base,
unsigned int align)
      && TYPE_CONTEXT (cand) == TYPE_CONTEXT (base)
      /* Check alignment.  */
      && TYPE_ALIGN (cand) == align
+     && TYPE_USER_ALIGN (cand) == TYPE_USER_ALIGN (base)
      && attribute_list_equal (TYPE_ATTRIBUTES (cand),
                   TYPE_ATTRIBUTES (base))
      && check_lang_type (cand, base));

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[jakub at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

Jakub Jelinek <jakub at gcc dot gnu.org> changed:

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

--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
(In reply to Marek Polacek from comment #7)
> So a fix could be this, but maybe it would make us create a lot more
> variants (?):
> 
> --- a/gcc/tree.c
> +++ b/gcc/tree.c
> @@ -6493,7 +6493,8 @@ check_base_type (const_tree cand, const_tree base)
>                     TYPE_ATTRIBUTES (base)))
>      return false;
>    /* Check alignment.  */
> -  if (TYPE_ALIGN (cand) == TYPE_ALIGN (base))
> +  if (TYPE_ALIGN (cand) == TYPE_ALIGN (base)
> +      && TYPE_USER_ALIGN (cand) == TYPE_USER_ALIGN (base))
>      return true;
>    /* Atomic types increase minimal alignment.  We must to do so as well
>       or we get duplicated canonical types. See PR88686.  */
> @@ -6528,6 +6529,7 @@ check_aligned_type (const_tree cand, const_tree base,
> unsigned int align)
>       && TYPE_CONTEXT (cand) == TYPE_CONTEXT (base)
>       /* Check alignment.  */
>       && TYPE_ALIGN (cand) == align
> +     && TYPE_USER_ALIGN (cand) == TYPE_USER_ALIGN (base)
>       && attribute_list_equal (TYPE_ATTRIBUTES (cand),
>                    TYPE_ATTRIBUTES (base))
>       && check_lang_type (cand, base));

It looks like the right thing to me.  I guess if we really wanted, we could
instrument the compiler to see how common this was (though best in a separate
build from just the above patch), i.e. when we would create a new type only
with this patch and not without.

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

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

--- Comment #9 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Thanks, I'll try to obtain some more data.  Mine for now.

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #10 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Marek Polacek <mpolacek@gcc.gnu.org>:

https://gcc.gnu.org/g:6318fe77395fc0dd59419bc4d69cd06ac0158e54

commit r10-8071-g6318fe77395fc0dd59419bc4d69cd06ac0158e54
Author: Marek Polacek <polacek@redhat.com>
Date:   Wed Apr 29 15:36:35 2020 -0400

    tree: Don't reuse types if TYPE_USER_ALIGN differ [PR94775]

    Here we trip on the TYPE_USER_ALIGN (t) assert in strip_typedefs: it
    gets "const d[0]" with TYPE_USER_ALIGN=0 but the result built by
    build_cplus_array_type is "const char[0]" with TYPE_USER_ALIGN=1.

    When we strip_typedefs the element of the array "const d", we see it's
    a typedef_variant_p, so we look at its DECL_ORIGINAL_TYPE, which is
    char, but we need to add the const qualifier, so we call
    cp_build_qualified_type -> build_qualified_type
    where get_qualified_type checks to see if we already have such a type
    by walking the variants list, which in this case is:

      char -> c -> const char -> const char -> d -> const d

    Because check_base_type only checks TYPE_ALIGN and not TYPE_USER_ALIGN,
    we choose the first const char, which has TYPE_USER_ALIGN set.  If the
    element type of an array has TYPE_USER_ALIGN, the array type gets it too.

    So we can make check_base_type stricter.  I was afraid that it might make
    us reuse types less often, but measuring showed that we build the same
    amount of types with and without the patch, while bootstrapping.

            PR c++/94775
            * tree.c (check_base_type): Return true only if TYPE_USER_ALIGN
match.
            (check_aligned_type): Check if TYPE_USER_ALIGN match.

            * g++.dg/warn/Warray-bounds-10.C: New test.

[Bug c++/94775] [8/9 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[8/9/10 Regression] ICE in  |[8/9 Regression] ICE in
                   |strip_typedefs, at          |strip_typedefs, at
                   |cp/tree.c:1734 since        |cp/tree.c:1734 since
                   |r8-4668-g8a5ee94a082b3d48   |r8-4668-g8a5ee94a082b3d48

--- Comment #11 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Fixed on trunk so far.

[Bug c++/94775] [8/9 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

https://gcc.gnu.org/g:05be85b649173b10d0bf10255eb15275c2dcf509

commit r11-42-g05be85b649173b10d0bf10255eb15275c2dcf509
Author: Marek Polacek <polacek@redhat.com>
Date:   Mon May 4 10:26:43 2020 -0400

    Revert "tree: Don't reuse types if TYPE_USER_ALIGN differ [PR94775]"

    This reverts commit 6318fe77395fc0dd59419bc4d69cd06ac0158e54.

            Revert:
            2020-04-30  Marek Polacek  <polacek@redhat.com>

            PR c++/94775
            * tree.c (check_base_type): Return true only if TYPE_USER_ALIGN
match.
            (check_aligned_type): Check if TYPE_USER_ALIGN match.

[Bug c++/94775] [8/9 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #13 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-10 branch has been updated by Marek Polacek
<mpolacek@gcc.gnu.org>:

https://gcc.gnu.org/g:251c85372e088017e79894f50156901d112affee

commit r10-8088-g251c85372e088017e79894f50156901d112affee
Author: Marek Polacek <polacek@redhat.com>
Date:   Mon May 4 10:41:24 2020 -0400

    Revert "tree: Don't reuse types if TYPE_USER_ALIGN differ [PR94775]"

            Revert:
            2020-04-30  Marek Polacek  <polacek@redhat.com>

            PR c++/94775
            * tree.c (check_base_type): Return true only if TYPE_USER_ALIGN
match.
            (check_aligned_type): Check if TYPE_USER_ALIGN match.

[Bug c++/94775] [8/9/10/11 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[8/9 Regression] ICE in     |[8/9/10/11 Regression] ICE
                   |strip_typedefs, at          |in strip_typedefs, at
                   |cp/tree.c:1734 since        |cp/tree.c:1734 since
                   |r8-4668-g8a5ee94a082b3d48   |r8-4668-g8a5ee94a082b3d48

--- Comment #14 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Patch reverted from 10/11.

[Bug c++/94775] [8/9/10/11 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[zhroma at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #15 from Roman Zhuykov <zhroma at gcc dot gnu.org> ---
(In reply to Marek Polacek from comment #14)
> Patch reverted from 10/11.
Maybe I've missed something in gcc-patches, but why it was reverted?  I can
imagine reasons to revert in 10 branch, but in master...

[Bug c++/94775] [8/9/10/11 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #16 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
It caused testsuite failures on non-x86-64 platforms, see bug 94896.

[Bug c++/94775] [8/9/10/11 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #17 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Another attempt:
https://gcc.gnu.org/pipermail/gcc-patches/2021-January/564461.html

[Bug c++/94775] [8/9/10/11 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

--- Comment #18 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Marek Polacek <mpolacek@gcc.gnu.org>:

https://gcc.gnu.org/g:513ee7d2cd9a60339a50dc9c4cba39a8f1c707f0

commit r11-6963-g513ee7d2cd9a60339a50dc9c4cba39a8f1c707f0
Author: Marek Polacek <polacek@redhat.com>
Date:   Thu Jan 28 16:21:50 2021 -0500

    tree: Don't reuse types if TYPE_USER_ALIGN differ [PR94775]

    A year ago I submitted this patch:

    ~~
    Here we trip on the TYPE_USER_ALIGN (t) assert in strip_typedefs: it
    gets "const d[0]" with TYPE_USER_ALIGN=0 but the result built by
    build_cplus_array_type is "const char[0]" with TYPE_USER_ALIGN=1.

    When we strip_typedefs the element of the array "const d", we see it's
    a typedef_variant_p, so we look at its DECL_ORIGINAL_TYPE, which is
    char, but we need to add the const qualifier, so we call
    cp_build_qualified_type -> build_qualified_type
    where get_qualified_type checks to see if we already have such a type
    by walking the variants list, which in this case is:

      char -> c -> const char -> const char -> d -> const d

    Because check_base_type only checks TYPE_ALIGN and not TYPE_USER_ALIGN,
    we choose the first const char, which has TYPE_USER_ALIGN set.  If the
    element type of an array has TYPE_USER_ALIGN, the array type gets it too.

    So we can make check_base_type stricter.  I was afraid that it might make
    us reuse types less often, but measuring showed that we build the same
    amount of types with and without the patch, while bootstrapping.
    ~~

    However, the patch broke a few tests on STRICT_ALIGNMENT platforms and
    had to be reverted.  This is another try.  The original patch is kept
    unchanged, but I added the finalize_type_size hunk that ought to fix the
    STRICT_ALIGNMENT issues.

    The problem is that finalize_type_size can clear TYPE_USER_ALIGN on the
    main variant of a type, but doesn't clear it on any of the variants.
    Then we end up with types which share the same TYPE_MAIN_VARIANT, but
    their TYPE_CANONICAL differs and then the usual "canonical types differ
    for identical types" follows.

    I've created alignas19.C to exercise this scenario.  What happens is:
    - when parsing the class S we create a type S in xref_tag,
    - we see alignas(8) so common_handle_aligned_attribute sets T_U_A in S,
    - we parse the member function fn and build_memfn_type creates a copy
      of S to add const; this variant has T_U_A set,
    - we finish_struct S which calls layout_class_type -> finish_record_type
      -> finalize_size_type where we reset T_U_A in S (but const S keeps it),
    - finish_non_static_data_member for arr calls maybe_dummy_object with
      type = S,
    - maybe_dummy_object calls same_type_ignoring_top_level_qualifiers_p
      to check if S and TREE_TYPE (current_class_ref), which is const S,
      are the same,
    - same_type_ignoring_top_level_qualifiers_p creates cv-unqualified
      versions of the passed types.  Previously we'd use our main variant
      S when stripping "const S" of const, but since the T_U_A flags don't
      match (check_base_type), we create a new variant S'.  Then we crash in
      comptypes because S and S' have the same TYPE_MAIN_VARIANT but
      different TYPE_CANONICALs.

    With my patch we'll clear T_U_A for S's variants too, and then instead
    of S' we'll just use S.

    gcc/ChangeLog:

            PR c++/94775
            * stor-layout.c (finalize_type_size): If we reset TYPE_USER_ALIGN
in
            the main variant, maybe reset it in its variants too.
            * tree.c (check_base_type): Return true only if TYPE_USER_ALIGN
match.
            (check_aligned_type): Check if TYPE_USER_ALIGN match.

    gcc/testsuite/ChangeLog:

            PR c++/94775
            * g++.dg/cpp0x/alignas19.C: New test.
            * g++.dg/warn/Warray-bounds15.C: New test.

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|[8/9/10/11 Regression] ICE  |[8/9/10 Regression] ICE in
                   |in strip_typedefs, at       |strip_typedefs, at
                   |cp/tree.c:1734 since        |cp/tree.c:1734 since
                   |r8-4668-g8a5ee94a082b3d48   |r8-4668-g8a5ee94a082b3d48

--- Comment #19 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
Fixed on trunk.  Not sure if I want to backport to 10...

[Bug c++/94775] [8/9/10 Regression] ICE in strip_typedefs, at cp/tree.c:1734 since r8-4668-g8a5ee94a082b3d48

[mpolacek at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94775

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #20 from Marek Polacek <mpolacek at gcc dot gnu.org> ---
I think I won't backport as the fix caused bug 99510.  Perhaps I could backport
both if someone really wants me to.

For now, fixed in GCC 11.