https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70449
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |jakub at gcc dot gnu.org,
| |jason at gcc dot gnu.org
--- Comment #8 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
The reason for the UNKNOWN_LOCATION is that in the testcase we try to
instantiate_decl the function being currently parsed, and do there
22017 input_location
22018 = DECL_STRUCT_FUNCTION (code_pattern)->function_end_locus;
but because we are in the middle of parsing that template, we haven't parsed
the closing } of the function template yet, so function_end_locus is unset.
That is also the reason for the bogus warning, the so far parsed statements of
the function template don't contain any return statements.
What does C++11/14 say about this?
clang++ 3.8 on this testcase warns:
warning: inline function 'f<0>' is not defined [-Wundefined-inline]
and if I add additional
constexpr int c = f<0> ();
after the template, it errors on it:
error: expression is not an integral constant expression
Both the clang diagnostics look really weird, I hope this really is invalid in
C++, can't see what the compiler could do there, because it needs the function
to be already parsed to use proper enumerators.
If I try
constexpr int f ()
{
enum E { a = f () };
return 0;
}
instead, then clang++ diagnostics is similarly weird, but g++ reports a nice
pr70449-3.C: In function ‘constexpr int f()’:
pr70449-3.C:3:18: error: ‘constexpr int f()’ called in a constant expression
before its definition is complete
enum E { a = f () };
~~^~
pr70449-3.C:3:19: error: enumerator value for ‘a’ is not an integer constant
enum E { a = f () };
^
error. So I think it would be best if we could emit the same diagnostics even
with templates.
