public inbox for gcc-cvs@sourceware.org
help / color / mirror / Atom feed
* [gcc r12-2859] c++: most vexing parse and braced CTAD [PR89062]
@ 2021-08-11 19:59 Patrick Palka
  0 siblings, 0 replies; only message in thread
From: Patrick Palka @ 2021-08-11 19:59 UTC (permalink / raw)
  To: gcc-cvs

https://gcc.gnu.org/g:6186708312780bb2139da01946abdde39667e985

commit r12-2859-g6186708312780bb2139da01946abdde39667e985
Author: Patrick Palka <ppalka@redhat.com>
Date:   Wed Aug 11 15:58:30 2021 -0400

    c++: most vexing parse and braced CTAD [PR89062]
    
    Here grokdeclarator is emitting the error
    
      error: class template placeholder ‘Foo’ not permitted in this context
    
    during the tentative (and ultimately futile) parse of 'x' as a function
    declaration.  This happens because when parsing 'Foo{1}',
    cp_parser_parameter_declaration yields a parameter declaration with no
    declarator and whose type is a CTAD placeholder, and stops short of
    consuming the '{'.  The caller cp_parser_parameter_declaration_list then
    calls grokdeclarator on this declarator, hence the error, and soon
    thereafter we abort this tentative parse since the next token '{'
    doesn't make sense in the context of a parameter list.
    
    Note that we don't have this issue with parenthesized CTAD
    
      Foo<int> x(Foo(1));
    
    because in this case cp_parser_direct_declarator (called indirectly from
    c_p_p_d) consumes the '(' and returns cp_error_declarator instead of a
    NULL declarator (and also simulates a parse error), and grokdeclarator
    exits early for this declarator without emitting any error.
    
    Since grokdeclarator doesn't take a 'complain' parameter, to fix this we
    need to avoid calling grokdeclarator in this situation.  To that end
    this patch makes c_p_p_d simulate an error when a construct is a CTAD
    expression and definitely not a parameter declaration, so that c_p_p_d_l
    can avoid calling grokdeclarator by checking for this simulated error.
    Alternatively we could keep all this logic inside c_p_p_d_l and not
    touch c_p_p_d at all, but this approach seems slightly less adhoc.
    
            PR c++/89062
    
    gcc/cp/ChangeLog:
    
            * parser.c (cp_parser_parameter_declaration_list): Don't call
            grokdeclarator if cp_parser_error_occurred.
            (cp_parser_parameter_declaration): Simulate an error if we see
            the beginning of a CTAD form, i.e. if we see an opening brace
            after the decl-specifier-seq and the type is a CTAD placeholder.
    
    gcc/testsuite/ChangeLog:
    
            * g++.dg/cpp1z/class-deduction97.C: New test.

Diff:
---
 gcc/cp/parser.c                                | 17 +++++++++++++----
 gcc/testsuite/g++.dg/cpp1z/class-deduction97.C |  6 ++++++
 2 files changed, 19 insertions(+), 4 deletions(-)

diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c
index 1be42a1a776..87e8d37d473 100644
--- a/gcc/cp/parser.c
+++ b/gcc/cp/parser.c
@@ -24284,7 +24284,7 @@ cp_parser_parameter_declaration_list (cp_parser* parser, cp_parser_flags flags)
 	 and warn in grokparms if appropriate.  */
       deprecated_state = DEPRECATED_SUPPRESS;
 
-      if (parameter)
+      if (parameter && !cp_parser_error_occurred (parser))
 	{
 	  decl = grokdeclarator (parameter->declarator,
 				 &parameter->decl_specifiers,
@@ -24499,7 +24499,7 @@ cp_parser_parameter_declaration (cp_parser *parser,
       parser->default_arg_ok_p = false;
 
       /* After seeing a decl-specifier-seq, if the next token is not a
-	 "(", there is no possibility that the code is a valid
+	 "(" or "{", there is no possibility that the code is a valid
 	 expression.  Therefore, if parsing tentatively, we commit at
 	 this point.  */
       if (!parser->in_template_argument_list_p
@@ -24512,9 +24512,18 @@ cp_parser_parameter_declaration (cp_parser *parser,
 	     of some object of type "char" to "int".  */
 	  && !parser->in_type_id_in_expr_p
 	  && cp_parser_uncommitted_to_tentative_parse_p (parser)
-	  && cp_lexer_next_token_is_not (parser->lexer, CPP_OPEN_BRACE)
 	  && cp_lexer_next_token_is_not (parser->lexer, CPP_OPEN_PAREN))
-	cp_parser_commit_to_tentative_parse (parser);
+	{
+	  if (cp_lexer_next_token_is (parser->lexer, CPP_OPEN_BRACE))
+	    {
+	      if (decl_specifiers.type
+		  && template_placeholder_p (decl_specifiers.type))
+		/* This is a CTAD expression, not a parameter declaration.  */
+		cp_parser_simulate_error (parser);
+	    }
+	  else
+	    cp_parser_commit_to_tentative_parse (parser);
+	}
       /* Parse the declarator.  */
       declarator_token_start = token;
       declarator = cp_parser_declarator (parser,
diff --git a/gcc/testsuite/g++.dg/cpp1z/class-deduction97.C b/gcc/testsuite/g++.dg/cpp1z/class-deduction97.C
new file mode 100644
index 00000000000..32818681d8f
--- /dev/null
+++ b/gcc/testsuite/g++.dg/cpp1z/class-deduction97.C
@@ -0,0 +1,6 @@
+// PR c++/89062
+// { dg-do compile { target c++17 } }
+
+template<class T> struct Foo { Foo(T); };
+
+Foo<int> x(Foo{1});


^ permalink raw reply	[flat|nested] only message in thread

only message in thread, other threads:[~2021-08-11 19:59 UTC | newest]

Thread overview: (only message) (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2021-08-11 19:59 [gcc r12-2859] c++: most vexing parse and braced CTAD [PR89062] Patrick Palka

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).