On 3/26/20 11:49 PM, Patrick Palka wrote:
On Thu, 26 Mar 2020, Jason Merrill wrote:
On 3/26/20 6:36 PM, Patrick Palka wrote:
On Thu, 26 Mar 2020, Jason Merrill wrote:
On 3/26/20 5:28 PM, Patrick Palka wrote:
This adds support to detect and recover from the case where an opening
brace
immediately follows the start of a requires-clause. So rather than
emitting
the
error
error: expected primary-expression before '{' token
followed by a slew of irrevelant errors, we now assume the user had
intended
to
write "requires requires {" and diagnose and recover accordingly.
Tested on x86_64-pc-linux-gnu, does this look OK?
gcc/cp/ChangeLog:
PR c++/94306
* parser.c (cp_parser_requires_clause_opt): Diagnose and recover from
"requires {" when "requires requires {" was probably intended.
gcc/testsuite/ChangeLog:
PR c++/94306
* g++.dg/concepts/diagnostic8.C: New test.
---
gcc/cp/parser.c | 17 ++++++++++++++++-
gcc/testsuite/g++.dg/concepts/diagnostic8.C | 6 ++++++
2 files changed, 22 insertions(+), 1 deletion(-)
create mode 100644 gcc/testsuite/g++.dg/concepts/diagnostic8.C
diff --git a/gcc/cp/parser.c b/gcc/cp/parser.c
index 05363653691..73c2c2cb010 100644
--- a/gcc/cp/parser.c
+++ b/gcc/cp/parser.c
@@ -27639,7 +27639,22 @@ cp_parser_requires_clause_opt (cp_parser
*parser,
bool lambda_p)
}
return NULL_TREE;
}
- cp_lexer_consume_token (parser->lexer);
+
+ cp_token *tok2 = cp_lexer_peek_nth_token (parser->lexer, 2);
+ if (tok2->type == CPP_OPEN_BRACE)
Do we want to handle an open paren the same way?
Hmm, wouldn't the error recovery be significantly different for an open
paren here? Just pretending that there's another "requires" token would
mean we proceed as if we got "requires requires (" but that is still
ill-formed I think.
Not ill-formed; a requires-expression can have a parameter-list.
Ah, of course.. But then again, wouldn't "requires (" be not
necessarily ill-formed, since the open paren could possibly be a part of
the constraint-expression e.g. "requires (a || b)"?
Ah, yes.
We could perhaps look ahead of the open paren and check if we match the
ill-formed pattern "requires (...) {...}" and if so, handle the open
paren the same way. Oh, but even that pattern is not necessarily
ill-formed, at least when parsing a lambda: the following
"[] <typename T> () -> void requires (true) { }"
is apparently a well-formed lambda-expression whose tokens starting from
its requires-clause would match that pattern. So we would probably have
to condition this open-paren handling on !lambda_p.
Yeah, this is too much trouble. Your patch is OK as is.
+ {
+ /* An opening brace following the start of a requires-clause is
+ ill-formed; the user likely forgot the second `requires' that
+ would start a requires-expression. */
+ gcc_rich_location richloc (tok2->location);
+ richloc.add_fixit_insert_before (" requires");
+ error_at (&richloc, "missing additional %<requires%> to start "
+ "a requires-expression");
+ /* Don't consume the `requires', so that it's reused as the start
of
a
+ requires-expression. */
+ }
+ else
+ cp_lexer_consume_token (parser->lexer);
if (!flag_concepts_ts)
return cp_parser_requires_clause_expression (parser, lambda_p);
diff --git a/gcc/testsuite/g++.dg/concepts/diagnostic8.C
b/gcc/testsuite/g++.dg/concepts/diagnostic8.C
new file mode 100644
index 00000000000..70d7e4a9cc1
--- /dev/null
+++ b/gcc/testsuite/g++.dg/concepts/diagnostic8.C
@@ -0,0 +1,6 @@
+// PR c++/94306
+// { dg-do compile { target c++2a } }
+
+template<typename T> struct S { };
+template<typename T> requires { typename T::type; } struct S<T> { };
+// { dg-error "missing additional .requires." "" { target *-*-* } .-1 }
