On 8/7/24 9:00 PM, Arsen Arsenović wrote:
Jason Merrill <ja...@redhat.com> writes:
On 8/7/24 7:31 PM, Arsen Arsenović wrote:
Enlargening the function-specific data block is not great.
Indeed, I think it would be better to search DECL_SAVED_TREE for a RETURN_STMT
once we've decided to give an error.
The trouble with that is that finish_return_stmt currently uses
input_location as the location for the entire return expr, so the
location ends up being after the entire return value.
I've hacked in a way to provide a different location to
finih_return_stmt, when applying it like below, the produced result is
the first result in the original email:
diff --git a/gcc/cp/parser.cc b/gcc/cp/parser.cc
index 6cfe42f3bdd6..44b45f16b026 100644
--- a/gcc/cp/parser.cc
+++ b/gcc/cp/parser.cc
@@ -14965,12 +14965,15 @@ cp_parser_jump_statement (cp_parser* parser, tree
&std_attrs)
/* Build the return-statement, check co-return first, since type
deduction is not valid there. */
+ auto l = make_location (token->location,
+ token->location,
+ input_location);
if (keyword == RID_CO_RETURN)
statement = finish_co_return_stmt (token->location, expr);
else if (FNDECL_USED_AUTO (current_function_decl) && in_discarded_stmt)
/* Don't deduce from a discarded return statement. */;
else
- statement = finish_return_stmt (expr);
+ statement = finish_return_stmt (expr, l);
/* Look for the final `;'. */
cp_parser_require (parser, CPP_SEMICOLON, RT_SEMICOLON);
}
... without this change (so, using input_location), the result is:
test.cc:38:11: error: a ‘return’ statement is not allowed in coroutine; did you
mean ‘co_return’?
38 | return {};
| ^
... which is not the best. That's the change I'm referring to in the
original post that I haven't ran the testsuite on. Changing that
location allows for simply searching DECL_SAVED_TREE (fndecl), though,
and getting a good location out of it.
Sounds good.
Jason
