Re: [PATCH] c++: NTTP object wrapper substitution fixes [PR103346, ...]

[Patrick Palka <ppalka@redhat.com>]

On Tue, 6 Dec 2022, Patrick Palka wrote:

> On Tue, 6 Dec 2022, Patrick Palka wrote:
> 
> > This patch fixes some issues with substitution into a C++20 template
> > parameter object wrapper:
> > 
> > * The first testcase demonstrates a situation where the same_type_p
> >   assert in relevant case of tsubst_copy doesn't hold, because (partial)
> >   substitution of {int,} into the VIEW_CONVERT_EXPR wrapper yields
> >   A<int> but substitution into the underlying TEMPLATE_PARM_INDEX is a
> >   nop and yields A<T> due to tsubst's level == 1 early exit test.  So
> >   this patch just gets rid of the assert; the type mismatch doesn't
> >   seem to be a problem in practice since the coercion is from one
> >   dependent type to another.
> > 
> > * In the second testcase, dependent substitution into the underlying
> >   TEMPLATE_PARM_INDEX yields a CALL_EXPR with empty TREE_TYPE, which
> >   tsubst_copy doesn't expect.  This patch fixes this by handling empty
> >   TREE_TYPE the same way as a non-const TREE_TYPE.  Moreover, after
> >   this substitution we're left with a VIEW_CONVERT_EXPR wrapping a
> >   CALL_EXPR instead of a TEMPLATE_PARM_INDEX, which during the subsequent
> >   non-dependent substitution tsubst_copy doesn't expect either.  So
> >   this patch also relaxes the tsubst_copy case to accept such
> >   VIEW_CONVERT_EXPR too.
> > 
> > * In the third testcase, we end up never resolving the call to
> >   f.modify() since tsubst_copy doesn't do overload resolution.
> >   This patch fixes this by moving the handling of these
> >   VIEW_CONVERT_EXPR wrappers from tsubst_copy to tsubst_copy_and_build.
> >   And it turns out (at least according to our testsuite) that
> >   tsubst_copy doesn't directly need to handle the other kinds of
> >   NON_LVALUE_EXPR and VIEW_CONVERT_EXPR, so this patch also gets rid
> >   of the location_wrapper_p handling from tsubst_copy and moves the
> >   REF_PARENTHESIZED_P handling to tsubst_copy_and_build.
> 
> On second thought, getting rid of the location_wrapper_p and
> REF_PARENTHESIZED_P handling from tsubst_copy is perhaps too
> risky at this stage.  The following patch instead just moves
> the tparm object wrapper handling from tsubst_copy to
> tsubst_copy_and_build and leaves the rest of tsubst_copy alone.
> Smoke tested so far, full bootstrap+regtest in progress:
> 
> -- >8--
> 
> Subject: [PATCH] c++: NTTP object wrapper substitution fixes [PR103346, ...]
> 
> This patch fixes some issues with substitution into a C++20 template
> parameter object wrapper:
> 
> * The first testcase demonstrates a situation where the same_type_p
>   assert in relevant case of tsubst_copy doesn't hold, because (partial)
>   substitution of {int,} into the VIEW_CONVERT_EXPR wrapper yields
>   A<int> but substitution into the underlying TEMPLATE_PARM_INDEX is a
>   nop and yields A<T> due to tsubst's level == 1 early exit test.  So
>   this patch just gets rid of the assert; the type mismatch doesn't
>   seem to be a problem in practice, I suppose because the coercion is
>   from one dependent type to another.
> 
> * In the second testcase, dependent substitution into the underlying
>   TEMPLATE_PARM_INDEX yields a CALL_EXPR with empty TREE_TYPE, which
>   tsubst_copy doesn't expect.  This patch fixes this by handling empty
>   TREE_TYPE the same way as a non-const TREE_TYPE.  Moreover, after
>   this substitution we're left with a VIEW_CONVERT_EXPR wrapping a
>   CALL_EXPR instead of a TEMPLATE_PARM_INDEX, which during the subsequent
>   non-dependent substitution tsubst_copy doesn't expect either.  So
>   this patch also relaxes tsubst_copy to accept such VIEW_CONVERT_EXPR
>   too.
> 
> * In the third testcase, we end up never resolving the call to
>   f.modify() since tsubst_copy doesn't do overload resolution.
>   This patch fixes this by moving the handling of these
>   VIEW_CONVERT_EXPR wrappers from tsubst_copy to tsubst_copy_and_build.
>   For good measure tsubst_copy_and_build should also handle
>   REF_PARENTHESIZED_P wrappers instead of delegating to tsubst_copy.
> 
> After this patch, VIEW_CONVERT_EXPR substitution is ultimately just
> moved from tsubst_copy to tsubst_copy_and_build and made more
> permissive.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk?

Ping.

> 
> 	PR c++/103346
> 	PR c++/104278
> 	PR c++/102553
> 
> gcc/cp/ChangeLog:
> 
> 	* pt.cc (tsubst_copy) <case VIEW_CONVERT_EXPR>: In the handling
> 	of C++20 template parameter object wrappers: Remove same_type_p
> 	assert.  Accept non-TEMPLATE_PARM_INDEX inner operand.  Handle
> 	empty TREE_TYPE on substituted inner operand.  Move it to ...
> 	(tsubst_copy_and_build): ... here.  Also handle REF_PARENTHESIZED_P
> 	VIEW_CONVERT_EXPRs.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/cpp2a/nontype-class52a.C: New test.
> 	* g++.dg/cpp2a/nontype-class53.C: New test.
> 	* g++.dg/cpp2a/nontype-class54.C: New test.
> 	* g++.dg/cpp2a/nontype-class55.C: New test.
> ---
>  gcc/cp/pt.cc                                  | 73 ++++++++++---------
>  gcc/testsuite/g++.dg/cpp2a/nontype-class52a.C | 15 ++++
>  gcc/testsuite/g++.dg/cpp2a/nontype-class53.C  | 25 +++++++
>  gcc/testsuite/g++.dg/cpp2a/nontype-class54.C  | 23 ++++++
>  gcc/testsuite/g++.dg/cpp2a/nontype-class55.C  | 15 ++++
>  5 files changed, 116 insertions(+), 35 deletions(-)
>  create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class52a.C
>  create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class53.C
>  create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class54.C
>  create mode 100644 gcc/testsuite/g++.dg/cpp2a/nontype-class55.C
> 
> diff --git a/gcc/cp/pt.cc b/gcc/cp/pt.cc
> index 2d8e4fdd4b5..0a196f069ad 100644
> --- a/gcc/cp/pt.cc
> +++ b/gcc/cp/pt.cc
> @@ -17271,42 +17271,16 @@ tsubst_copy (tree t, tree args, tsubst_flags_t complain, tree in_decl)
>  	      return maybe_wrap_with_location (op0, EXPR_LOCATION (t));
>  	    }
>  	  tree op = TREE_OPERAND (t, 0);
> -	  if (code == VIEW_CONVERT_EXPR
> -	      && TREE_CODE (op) == TEMPLATE_PARM_INDEX)
> -	    {
> -	      /* Wrapper to make a C++20 template parameter object const.  */
> -	      op = tsubst_copy (op, args, complain, in_decl);
> -	      if (!CP_TYPE_CONST_P (TREE_TYPE (op)))
> -		{
> -		  /* The template argument is not const, presumably because
> -		     it is still dependent, and so not the const template parm
> -		     object.  */
> -		  tree type = tsubst (TREE_TYPE (t), args, complain, in_decl);
> -		  gcc_checking_assert (same_type_ignoring_top_level_qualifiers_p
> -				       (type, TREE_TYPE (op)));
> -		  if (TREE_CODE (op) == CONSTRUCTOR
> -		      || TREE_CODE (op) == IMPLICIT_CONV_EXPR)
> -		    {
> -		      /* Don't add a wrapper to these.  */
> -		      op = copy_node (op);
> -		      TREE_TYPE (op) = type;
> -		    }
> -		  else
> -		    /* Do add a wrapper otherwise (in particular, if op is
> -		       another TEMPLATE_PARM_INDEX).  */
> -		    op = build1 (code, type, op);
> -		}
> -	      return op;
> -	    }
>  	  /* force_paren_expr can also create a VIEW_CONVERT_EXPR.  */
> -	  else if (code == VIEW_CONVERT_EXPR && REF_PARENTHESIZED_P (t))
> +	  if (code == VIEW_CONVERT_EXPR && REF_PARENTHESIZED_P (t))
>  	    {
>  	      op = tsubst_copy (op, args, complain, in_decl);
>  	      op = build1 (code, TREE_TYPE (op), op);
>  	      REF_PARENTHESIZED_P (op) = true;
>  	      return op;
>  	    }
> -	  /* We shouldn't see any other uses of these in templates.  */
> +	  /* We shouldn't see any other uses of these in templates
> +	     (tsubst_copy_and_build handles C++20 tparm object wrappers).  */
>  	  gcc_unreachable ();
>  	}
>  
> @@ -21569,12 +21543,41 @@ tsubst_copy_and_build (tree t,
>  
>      case NON_LVALUE_EXPR:
>      case VIEW_CONVERT_EXPR:
> -      if (location_wrapper_p (t))
> -	/* We need to do this here as well as in tsubst_copy so we get the
> -	   other tsubst_copy_and_build semantics for a PARM_DECL operand.  */
> -	RETURN (maybe_wrap_with_location (RECUR (TREE_OPERAND (t, 0)),
> -					  EXPR_LOCATION (t)));
> -      /* fallthrough.  */
> +      {
> +	tree op = RECUR (TREE_OPERAND (t, 0));
> +	if (location_wrapper_p (t))
> +	  /* We need to do this here as well as in tsubst_copy so we get the
> +	     other tsubst_copy_and_build semantics for a PARM_DECL operand.  */
> +	  RETURN (maybe_wrap_with_location (op, EXPR_LOCATION (t)));
> +
> +	gcc_checking_assert (TREE_CODE (t) == VIEW_CONVERT_EXPR);
> +	if (REF_PARENTHESIZED_P (t))
> +	  /* force_paren_expr can also create a VIEW_CONVERT_EXPR.  */
> +	  RETURN (finish_parenthesized_expr (op));
> +
> +	/* Otherwise, we're dealing with a wrapper to make a C++20 template
> +	   parameter object const.  */
> +	if (TREE_TYPE (op) == NULL_TREE
> +	    || !CP_TYPE_CONST_P (TREE_TYPE (op)))
> +	  {
> +	    /* The template argument is not const, presumably because
> +	       it is still dependent, and so not the const template parm
> +	       object.  */
> +	    tree type = tsubst (TREE_TYPE (t), args, complain, in_decl);
> +	    if (TREE_CODE (op) == CONSTRUCTOR
> +		|| TREE_CODE (op) == IMPLICIT_CONV_EXPR)
> +	      {
> +		/* Don't add a wrapper to these.  */
> +		op = copy_node (op);
> +		TREE_TYPE (op) = type;
> +	      }
> +	    else
> +	      /* Do add a wrapper otherwise (in particular, if op is
> +		 another TEMPLATE_PARM_INDEX).  */
> +	      op = build1 (VIEW_CONVERT_EXPR, type, op);
> +	  }
> +	RETURN (op);
> +      }
>  
>      default:
>        /* Handle Objective-C++ constructs, if appropriate.  */
> diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class52a.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class52a.C
> new file mode 100644
> index 00000000000..ae5d5df70ac
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class52a.C
> @@ -0,0 +1,15 @@
> +// A version of nontype-class52.C where explicit template arguments are
> +// given in the call to f (which during deduction need to be partially
> +// substituted into the NTTP object V in f's signature).
> +// { dg-do compile { target c++20 } }
> +
> +template<class> struct A { };
> +
> +template<auto> struct B { };
> +
> +template<class T, A<T> V> void f(B<V>);
> +
> +int main() {
> +  constexpr A<int> a;
> +  f<int>(B<a>{});
> +}
> diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class53.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class53.C
> new file mode 100644
> index 00000000000..9a6398c5f57
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class53.C
> @@ -0,0 +1,25 @@
> +// PR c++/103346
> +// { dg-do compile { target c++20 } }
> +
> +struct Item {};
> +
> +template<class T, T... ts>
> +struct Sequence { };
> +
> +template<Item... items>
> +using ItemSequence = Sequence<Item, items...>;
> +
> +template<Item... items>
> +constexpr auto f() {
> +  constexpr auto l = [](Item item) { return item; };
> +  return ItemSequence<l(items)...>{};
> +}
> +
> +using ty0 = decltype(f<>());
> +using ty0 = ItemSequence<>;
> +
> +using ty1 = decltype(f<{}>());
> +using ty1 = ItemSequence<{}>;
> +
> +using ty3 = decltype(f<{}, {}, {}>());
> +using ty3 = ItemSequence<{}, {}, {}>;
> diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class54.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class54.C
> new file mode 100644
> index 00000000000..8127b1f5426
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class54.C
> @@ -0,0 +1,23 @@
> +// PR c++/104278
> +// { dg-do compile { target c++20 } }
> +
> +struct foo {
> +  int value;
> +  constexpr foo modify() const { return { value + 1 }; }
> +};
> +
> +template<foo f, bool Enable = f.value & 1>
> +struct bar {
> +  static void run() { }
> +};
> +
> +template<foo f>
> +struct qux {
> +  static void run() {
> +    bar<f.modify()>::run();
> +  }
> +};
> +
> +int main() {
> +  qux<foo{}>::run();
> +}
> diff --git a/gcc/testsuite/g++.dg/cpp2a/nontype-class55.C b/gcc/testsuite/g++.dg/cpp2a/nontype-class55.C
> new file mode 100644
> index 00000000000..afcb3d64495
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/nontype-class55.C
> @@ -0,0 +1,15 @@
> +// PR c++/102553
> +// { dg-do compile { target c++20 } }
> +
> +struct s1{};
> +template<int> inline constexpr s1 ch{};
> +
> +template<s1 f> struct s2{};
> +template<s1 f> using alias1 = s2<f>;
> +
> +template<class T>
> +void general(int n) {
> +  alias1<ch<1>>{};
> +}
> +
> +template void general<int>(int);
> -- 
> 2.39.0.rc2
> 
>