[Bug c/86869] ICE when taking address of array member of __memx struct pointer

[rguenth at gcc dot gnu.org via Gcc-bugs]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=86869

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|target                      |c
   Last reconfirmed|                            |2023-12-04
             Status|UNCONFIRMED                 |ASSIGNED
     Ever confirmed|0                           |1

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
It still ICEs.  The IL from the frontend looks wrong.  We have

  p = &s->y;

but we should have a ADDR_SPACE_CONVERT_EXPR here, but we have

 <decl_expr 0x7ffff6b2d7e0
    type <void_type 0x7ffff6a210a8 void VOID
        align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6a210a8
        pointer_to_this <pointer_type 0x7ffff6a21150>>
    side-effects
    arg:0 <var_decl 0x7ffff6b62000 p
        type <pointer_type 0x7ffff6b45e70 type <array_type 0x7ffff6b45dc8>
            unsigned HI
...

and its DECL_INITIAL is

 <nop_expr 0x7ffff6b2d7c0
    type <pointer_type 0x7ffff6b45e70
        type <array_type 0x7ffff6b45dc8 type <integer_type 0x7ffff6a2a0a8 char>
            HI
            size <integer_cst 0x7ffff6a0ba98 constant 16>
            unit-size <integer_cst 0x7ffff6a0bab0 constant 2>
            align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6b45dc8 domain <integer_type 0x7ffff6b45930>
            pointer_to_this <pointer_type 0x7ffff6b45e70>>
        unsigned HI size <integer_cst 0x7ffff6a0ba98 16> unit-size <integer_cst
0x7ffff6a0bab0 2>
        align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6b45e70>
    readonly
    arg:0 <addr_expr 0x7ffff6b2d7a0
        type <pointer_type 0x7ffff6b45d20 type <array_type 0x7ffff6b45c78>
            unsigned HI size <integer_cst 0x7ffff6a0ba98 16> unit-size
<integer_cst 0x7ffff6a0bab0 2>
            align:8 warn_if_not_align:0 symtab:0 alias-set -1 canonical-type
0x7ffff6b45d20>
        readonly
        arg:0 <component_ref 0x7ffff6b5f120 type <array_type 0x7ffff6b45c78>
            readonly
            arg:0 <indirect_ref 0x7ffff6b2d780 type <record_type 0x7ffff6b45a80
S address-space-7>
                readonly arg:0 <parm_decl 0x7ffff7fc8080 s>
                t.c:6:26 start: t.c:6:26 finish: t.c:6:27> arg:1 <field_decl
0x7ffff6a28e40 y>
            t.c:6:26 start: t.c:6:25 finish: t.c:6:28>
        t.c:6:24 start: t.c:6:24 finish: t.c:6:28>
    t.c:6:24 start: t.c:6:24 finish: t.c:6:28>

and the issue is that while the RECORD_TYPE has the correct address-space
qualification the array-type does not (the array-type component does).

That's the fault of the C frontend c_build_qualified_type which
special-cases ARRAY_TYPE and doesn't behave like build_array_type_1 would.

I have a fix.