c++: ICE with concepts TS multiple auto deduction [PR101886]

  In extract_autos_r, we need to reset TYPE_CANONICAL for the template
type parameter after adjusting its index, otherwise we end up with a
comptypes ICE for the below testcase.  Note that such in-place type
adjustment isn't generallly safe to do since the type could be the
TYPE_CANONICAL of another (unadjusted) type, but in this case the
canonical auto (of some level and 0 index) is the first auto (of that
level) that's created, and so any auto that we do end up adjusting can't
be the canonical one.

Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
trunk?

	PR c++/101886

gcc/cp/ChangeLog:

	* pt.cc (extract_autos_r): Reset TYPE_CANONICAL after
	adjusting the template type parameter's index.  Simplify
	by using TEMPLATE_TYPE_IDX.  Add some sanity checks.

gcc/testsuite/ChangeLog:

	* g++.dg/concepts/auto5.C: New test.
---
 gcc/cp/pt.cc                          | 12 +++++++++---
 gcc/testsuite/g++.dg/concepts/auto5.C |  9 +++++++++
 2 files changed, 18 insertions(+), 3 deletions(-)
 create mode 100644 gcc/testsuite/g++.dg/concepts/auto5.C
  

On Wed, 7 Dec 2022, Patrick Palka wrote:

> In extract_autos_r, we need to reset TYPE_CANONICAL for the template
> type parameter after adjusting its index, otherwise we end up with a
> comptypes ICE for the below testcase.  Note that such in-place type
> adjustment isn't generallly safe to do since the type could be the
> TYPE_CANONICAL of another (unadjusted) type, but in this case the
> canonical auto (of some level and 0 index) is the first auto (of that
> level) that's created, and so any auto that we do end up adjusting can't
> be the canonical one.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk?

Ping.  FWIW this blocks the modules + std::source_location patch[1]
because the make_auto() call it adds to cxx_init_decl_processing would
otherwise cause us to ICE on the existing test g++.dg/concepts/auto1.C.

[1]: https://gcc.gnu.org/pipermail/gcc-patches/2022-December/608127.html

> 
> 	PR c++/101886
> 
> gcc/cp/ChangeLog:
> 
> 	* pt.cc (extract_autos_r): Reset TYPE_CANONICAL after
> 	adjusting the template type parameter's index.  Simplify
> 	by using TEMPLATE_TYPE_IDX.  Add some sanity checks.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/concepts/auto5.C: New test.
> ---
>  gcc/cp/pt.cc                          | 12 +++++++++---
>  gcc/testsuite/g++.dg/concepts/auto5.C |  9 +++++++++
>  2 files changed, 18 insertions(+), 3 deletions(-)
>  create mode 100644 gcc/testsuite/g++.dg/concepts/auto5.C
> 
> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
> index 24ed718ffbb..d05a49b1c11 100644
> --- a/gcc/cp/pt.cc
> +++ b/gcc/cp/pt.cc
> @@ -29164,18 +29164,24 @@ extract_autos_r (tree t, void *data)
>      {
>        /* All the autos were built with index 0; fix that up now.  */
>        tree *p = hash.find_slot (t, INSERT);
> -      unsigned idx;
> +      int idx;
>        if (*p)
>  	/* If this is a repeated constrained-type-specifier, use the index we
>  	   chose before.  */
> -	idx = TEMPLATE_PARM_IDX (TEMPLATE_TYPE_PARM_INDEX (*p));
> +	idx = TEMPLATE_TYPE_IDX (*p);
>        else
>  	{
>  	  /* Otherwise this is new, so use the current count.  */
>  	  *p = t;
>  	  idx = hash.elements () - 1;
>  	}
> -      TEMPLATE_PARM_IDX (TEMPLATE_TYPE_PARM_INDEX (t)) = idx;
> +      if (idx != TEMPLATE_TYPE_IDX (t))
> +	{
> +	  gcc_checking_assert (TEMPLATE_TYPE_IDX (t) == 0);
> +	  gcc_checking_assert (TYPE_CANONICAL (t) != t);
> +	  TEMPLATE_TYPE_IDX (t) = idx;
> +	  TYPE_CANONICAL (t) = canonical_type_parameter (t);
> +	}
>      }
>  
>    /* Always keep walking.  */
> diff --git a/gcc/testsuite/g++.dg/concepts/auto5.C b/gcc/testsuite/g++.dg/concepts/auto5.C
> new file mode 100644
> index 00000000000..f1d653efd87
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/concepts/auto5.C
> @@ -0,0 +1,9 @@
> +// PR c++/101886
> +// { dg-do compile { target c++17_only } }
> +// { dg-options "-fconcepts-ts" }
> +
> +template<typename...> struct A { };
> +
> +A<int, int> a;
> +A<auto, auto> b1 = a;
> +A<auto, auto> b2 = a;
> -- 
> 2.39.0.rc2
> 
>
  

On 12/7/22 15:18, Patrick Palka wrote:
> In extract_autos_r, we need to reset TYPE_CANONICAL for the template
> type parameter after adjusting its index, otherwise we end up with a
> comptypes ICE for the below testcase.  Note that such in-place type
> adjustment isn't generallly safe to do since the type could be the
> TYPE_CANONICAL of another (unadjusted) type, but in this case the
> canonical auto (of some level and 0 index) is the first auto (of that
> level) that's created, and so any auto that we do end up adjusting can't
> be the canonical one.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk?

OK.

> 	PR c++/101886
> 
> gcc/cp/ChangeLog:
> 
> 	* pt.cc (extract_autos_r): Reset TYPE_CANONICAL after
> 	adjusting the template type parameter's index.  Simplify
> 	by using TEMPLATE_TYPE_IDX.  Add some sanity checks.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/concepts/auto5.C: New test.
> ---
>   gcc/cp/pt.cc                          | 12 +++++++++---
>   gcc/testsuite/g++.dg/concepts/auto5.C |  9 +++++++++
>   2 files changed, 18 insertions(+), 3 deletions(-)
>   create mode 100644 gcc/testsuite/g++.dg/concepts/auto5.C
> 
> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
> index 24ed718ffbb..d05a49b1c11 100644
> --- a/gcc/cp/pt.cc
> +++ b/gcc/cp/pt.cc
> @@ -29164,18 +29164,24 @@ extract_autos_r (tree t, void *data)
>       {
>         /* All the autos were built with index 0; fix that up now.  */
>         tree *p = hash.find_slot (t, INSERT);
> -      unsigned idx;
> +      int idx;
>         if (*p)
>   	/* If this is a repeated constrained-type-specifier, use the index we
>   	   chose before.  */
> -	idx = TEMPLATE_PARM_IDX (TEMPLATE_TYPE_PARM_INDEX (*p));
> +	idx = TEMPLATE_TYPE_IDX (*p);
>         else
>   	{
>   	  /* Otherwise this is new, so use the current count.  */
>   	  *p = t;
>   	  idx = hash.elements () - 1;
>   	}
> -      TEMPLATE_PARM_IDX (TEMPLATE_TYPE_PARM_INDEX (t)) = idx;
> +      if (idx != TEMPLATE_TYPE_IDX (t))
> +	{
> +	  gcc_checking_assert (TEMPLATE_TYPE_IDX (t) == 0);
> +	  gcc_checking_assert (TYPE_CANONICAL (t) != t);
> +	  TEMPLATE_TYPE_IDX (t) = idx;
> +	  TYPE_CANONICAL (t) = canonical_type_parameter (t);
> +	}
>       }
>   
>     /* Always keep walking.  */
> diff --git a/gcc/testsuite/g++.dg/concepts/auto5.C b/gcc/testsuite/g++.dg/concepts/auto5.C
> new file mode 100644
> index 00000000000..f1d653efd87
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/concepts/auto5.C
> @@ -0,0 +1,9 @@
> +// PR c++/101886
> +// { dg-do compile { target c++17_only } }
> +// { dg-options "-fconcepts-ts" }
> +
> +template<typename...> struct A { };
> +
> +A<int, int> a;
> +A<auto, auto> b1 = a;
> +A<auto, auto> b2 = a;