[Bug c++/113620] New: Specialized Template

[Bug c++/113620] New: Specialized Template

[hans.buchmann.wantuch at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

            Bug ID: 113620
           Summary: Specialized Template
           Product: gcc
           Version: 13.2.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hans.buchmann.wantuch at gmail dot com
  Target Milestone: ---

Created attachment 57225
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57225&action=edit
Bug report generated with: -freport-bug

Compiling the code bug-1.cc

------8<------8<------8<------8<------
class CRC
{
 public:
  template<typename T>
  struct Def
  {
   using TYP=T;
   static const TYP INIT=0;
  };

  enum Method {MSB_FIRST,LSB_FIRST};

  template<Method M,typename DEF>
  class BIT{};

  template<typename DEF>
  struct BIT<MSB_FIRST,DEF>
  {
    DEF::TYP r=DEF::INIT;
  };
};

int main(int argc,char** args)
{
 CRC::BIT<CRC::MSB_FIRST,CRC::Def<unsigned short>> crc;
 return 0;
}
------8<------8<------8<------8<------

compiling with:

g++ -freport-bug -std=c++2a -c -o bug-1.o bug-1.cc

results in the following output:

see attached bug-report.txt

Sincerely

Hans Buchmann

[Bug c++/113620] Specialized Template

[xry111 at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

Xi Ruoyao <xry111 at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |xry111 at gcc dot gnu.org
             Status|UNCONFIRMED                 |NEW
           Keywords|                            |ice-on-invalid-code
     Ever confirmed|0                           |1
   Last reconfirmed|                            |2024-01-26

--- Comment #1 from Xi Ruoyao <xry111 at gcc dot gnu.org> ---
Confirmed.

t1.c:19:21: internal compiler error: tree check: accessed elt 2 of 'tree_vec'
with 1 elts in tsubst, at cp/pt.cc:16216
   19 |     DEF::TYP r=DEF::INIT;
      |                     ^~~~
0x910d44 tree_vec_elt_check_failed(int, int, char const*, int, char const*)
        ../../gcc/gcc/tree.cc:9153
0x7a8e83 tree_vec_elt_check(tree_node*, int, char const*, int, char const*)
        ../../gcc/gcc/tree.h:3865
0x7a8e83 tsubst(tree_node*, tree_node*, int, tree_node*)
        ../../gcc/gcc/cp/pt.cc:16216
0xc3ada0 tsubst_scope
        ../../gcc/gcc/cp/pt.cc:16857
0xc3ada0 tsubst_qualified_id
        ../../gcc/gcc/cp/pt.cc:17133
0xc153cd tsubst_expr(tree_node*, tree_node*, int, tree_node*)
        ../../gcc/gcc/cp/pt.cc:20199
0xb03627 maybe_instantiate_nsdmi_init(tree_node*, int)
        ../../gcc/gcc/cp/init.cc:624
0xb03b0a maybe_instantiate_nsdmi_init(tree_node*, int)
        ../../gcc/gcc/cp/init.cc:679
0xb03b0a get_nsdmi(tree_node*, bool, int)
        ../../gcc/gcc/cp/init.cc:656
0xb34ed1 walk_field_subobs
        ../../gcc/gcc/cp/method.cc:2466
0xb35576 walk_field_subobs
        ../../gcc/gcc/cp/method.cc:2390
0xb35576 synthesized_method_walk
        ../../gcc/gcc/cp/method.cc:2835
0xb35eed get_defaulted_eh_spec(tree_node*, int)
        ../../gcc/gcc/cp/method.cc:2877
0xc1a078 maybe_instantiate_noexcept(tree_node*, int)
        ../../gcc/gcc/cp/pt.cc:26655
0xc19ee6 maybe_instantiate_noexcept(tree_node*, int)
        ../../gcc/gcc/cp/pt.cc:26645
0xae729a mark_used(tree_node*, int)
        ../../gcc/gcc/cp/decl2.cc:5811
0xa1af3e build_over_call
        ../../gcc/gcc/cp/call.cc:10524
0xa183bf build_new_method_call(tree_node*, tree_node*, vec<tree_node*, va_gc,
vl_embed>**, tree_node*, int, tree_node**, int)
        ../../gcc/gcc/cp/call.cc:11796
0xa1962a build_special_member_call(tree_node*, tree_node*, vec<tree_node*,
va_gc, vl_embed>**, tree_node*, int, int)
        ../../gcc/gcc/cp/call.cc:11281
0xb07819 expand_default_init
        ../../gcc/gcc/cp/init.cc:2202

With -std=c++17 it's fine:

t1.c:19:5: error: need 'typename' before 'DEF::TYP' because 'DEF' is a
dependent scope
   19 |     DEF::TYP r=DEF::INIT;
      |     ^~~
      |     typename

[Bug c++/113620] ICE with missing "typename" and -std=c++20

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Keywords|ice-on-invalid-code         |ice-on-valid-code
      Known to fail|                            |5.1.0

--- Comment #2 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Xi Ruoyao from comment #1)
> With -std=c++17 it's fine:
> 
> t1.c:19:5: error: need 'typename' before 'DEF::TYP' because 'DEF' is a
> dependent scope
>    19 |     DEF::TYP r=DEF::INIT;
>       |     ^~~
>       |     typename

Note once you add the typename (which is optional in C++20 by the way), the ICE
is back in C++11 mode even.

This has been ICEing from at least GCC 5.1.0.

[Bug c++/113620] ICE with missing "typename" and -std=c++20

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

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

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.9.0

--- Comment #3 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
In GCC 4.8.5, it was rejected (incorrectly):
```
<source>:19:25: error: 'DEF' has not been declared
     typename DEF::TYP r=DEF::INIT;
                         ^

```

[Bug c++/113620] ICE with inner class template speciailization and depdent NSDMI

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Created attachment 57229
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=57229&action=edit
Reduced testcase

I reduced the testcase slightly to show it is NSDMI related dealing with a
dependent value only with the specialization.

Note a workaround is to add a constructor which does the initialization instead
of using NSDMI.

[Bug c++/113620] ICE with inner class template speciailization and depdent NSDMI

[ppalka at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

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

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

--- Comment #5 from Patrick Palka <ppalka at gcc dot gnu.org> ---
Seems to be a name lookup issue ultimately:

struct A {
  template<class T, class U>
  struct B;

  template<class V>
  struct B<int, V> {
    int x = V::value; // error: 'V' has not been declared
    int y = T::value; // no error!
    int z = U::value; // no error!
  };
};

[Bug c++/113620] ICE with inner class template speciailization and depdent NSDMI

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=113620

--- Comment #6 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
(In reply to Patrick Palka from comment #5)
> Seems to be a name lookup issue ultimately:
> 
> struct A {
>   template<class T, class U>
>   struct B;
> 
>   template<class V>
>   struct B<int, V> {
>     int x = V::value; // error: 'V' has not been declared
>     int y = T::value; // no error!
>     int z = U::value; // no error!
>   };
> };

That would make sense on why it was rejected before GCC 4.9 even.

Now I wonder if this is related to the diagnostic issue mentioned in
https://gcc.gnu.org/pipermail/gcc-patches/2024-January/643811.html even.