[PATCH] c++: class-scope qualified constrained auto [PR107188]

[PATCH] c++: class-scope qualified constrained auto [PR107188]

[Patrick Palka]

Here when parsing the class-scope auto constrained by a qualified
concept-id, we first tentatively parse the overall member-declaration as
a deprecated access-declaration, during which we parse C<int> as a
standalone TEMPLATE_ID_EXPR (not part of the auto) and end up emitting
the bogus error

concepts-placeholder11.C:9:6: error: wrong number of template arguments (1, should be 2)
    9 |   N::C<int> auto f() { return 0; }
      |      ^~~~~~
concepts-placeholder11.C:5:34: note: provided for ‘template<class, class> concept N::C’
    5 |   template<class, class> concept C = true;
      |                                  ^

from build_concept_id called from cp_parser_template_id_expr.

We could fix this by adding a complain parameter to build_concept_id and
passing tf_none when parsing tentatively.  However, it seems we can fix
this in a more general way that might benefit non-concepts code: when
tentatively parsing an access-declaration, abort the parse early if the
qualifying scope isn't possibly a class type, so that we avoid parsing
C<int> as a TEMPLATE_ID_EXPR in the first place.  This patch takes this
latter approach.

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

	PR c++/107188

gcc/cp/ChangeLog:

	* parser.cc (cp_parser_using_declaration): Abort the tentative
	parse early if the scope of an access-declaration isn't possibly
	a class type.

gcc/testsuite/ChangeLog:

	* g++.dg/cpp2a/concepts-placeholder11.C: New test.
---
 gcc/cp/parser.cc                                    |  5 +++++
 gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C | 10 ++++++++++
 2 files changed, 15 insertions(+)
 create mode 100644 gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C

diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
index e8a50904243..ccacf6d7dd0 100644
--- a/gcc/cp/parser.cc
+++ b/gcc/cp/parser.cc
@@ -21670,6 +21670,11 @@ cp_parser_using_declaration (cp_parser* parser,
 
   cp_warn_deprecated_use_scopes (qscope);
 
+  if (access_declaration_p && !MAYBE_CLASS_TYPE_P (qscope))
+    /* If the qualifying scope of an access-declaration isn't possibly
+       a class type then it must be invalid.  */
+    cp_parser_simulate_error (parser);
+
   if (access_declaration_p && cp_parser_error_occurred (parser))
     /* Something has already gone wrong; there's no need to parse
        further.  Since an error has occurred, the return value of
diff --git a/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C b/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C
new file mode 100644
index 00000000000..61eef743bae
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C
@@ -0,0 +1,10 @@
+// PR c++/107188
+// { dg-do compile { target c++20 } }
+
+namespace N {
+  template<class, class> concept C = true;
+}
+
+struct X {
+  N::C<int> auto f() { return 0; }
+};
-- 
2.39.0.rc2

Re: [PATCH] c++: class-scope qualified constrained auto [PR107188]

[Jason Merrill]

On 12/8/22 11:42, Patrick Palka wrote:
> Here when parsing the class-scope auto constrained by a qualified
> concept-id, we first tentatively parse the overall member-declaration as
> a deprecated access-declaration, during which we parse C<int> as a
> standalone TEMPLATE_ID_EXPR (not part of the auto) and end up emitting
> the bogus error
> 
> concepts-placeholder11.C:9:6: error: wrong number of template arguments (1, should be 2)
>      9 |   N::C<int> auto f() { return 0; }
>        |      ^~~~~~
> concepts-placeholder11.C:5:34: note: provided for ‘template<class, class> concept N::C’
>      5 |   template<class, class> concept C = true;
>        |                                  ^
> 
> from build_concept_id called from cp_parser_template_id_expr.
> 
> We could fix this by adding a complain parameter to build_concept_id and
> passing tf_none when parsing tentatively.  However, it seems we can fix
> this in a more general way that might benefit non-concepts code: when
> tentatively parsing an access-declaration, abort the parse early if the
> qualifying scope isn't possibly a class type, so that we avoid parsing
> C<int> as a TEMPLATE_ID_EXPR in the first place.  This patch takes this
> latter approach.
> 
> Bootstrapped and regtested on x86_64-pc-linux-gnu, does this look OK for
> trunk?

OK.

> 	PR c++/107188
> 
> gcc/cp/ChangeLog:
> 
> 	* parser.cc (cp_parser_using_declaration): Abort the tentative
> 	parse early if the scope of an access-declaration isn't possibly
> 	a class type.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* g++.dg/cpp2a/concepts-placeholder11.C: New test.
> ---
>   gcc/cp/parser.cc                                    |  5 +++++
>   gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C | 10 ++++++++++
>   2 files changed, 15 insertions(+)
>   create mode 100644 gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C
> 
> diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
> index e8a50904243..ccacf6d7dd0 100644
> --- a/gcc/cp/parser.cc
> +++ b/gcc/cp/parser.cc
> @@ -21670,6 +21670,11 @@ cp_parser_using_declaration (cp_parser* parser,
>   
>     cp_warn_deprecated_use_scopes (qscope);
>   
> +  if (access_declaration_p && !MAYBE_CLASS_TYPE_P (qscope))
> +    /* If the qualifying scope of an access-declaration isn't possibly
> +       a class type then it must be invalid.  */
> +    cp_parser_simulate_error (parser);
> +
>     if (access_declaration_p && cp_parser_error_occurred (parser))
>       /* Something has already gone wrong; there's no need to parse
>          further.  Since an error has occurred, the return value of
> diff --git a/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C b/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C
> new file mode 100644
> index 00000000000..61eef743bae
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/cpp2a/concepts-placeholder11.C
> @@ -0,0 +1,10 @@
> +// PR c++/107188
> +// { dg-do compile { target c++20 } }
> +
> +namespace N {
> +  template<class, class> concept C = true;
> +}
> +
> +struct X {
> +  N::C<int> auto f() { return 0; }
> +};