https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117803
Tobias Burnus <burnus at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Resolution|--- |INVALID
Status|UNCONFIRMED |RESOLVED
--- Comment #3 from Tobias Burnus <burnus at gcc dot gnu.org> ---
> The base3 case looks valid, but I don't see it rejected.
And it is accepted, sorry. Pilot error.
* * *
> That testcase looks invalid to me.
> Try calling repl1() directly, it will fail the same way, one can only
> call it with explicit template parameters, so repl1<int>();
> or repl1<T>(); etc.
I think part of my problem is that the OpenMP specification does not clearly
spell out how templates are handled.
I think my confusion stemmed from the question
how the replacement happens, i.e. does one replace
#pragma omp dispatch
x = base1<int>();
by
x = repl1(); // this will fail
or
x = repl1<int>(); // this will work
However, the second variant assumes that the variant function
uses the same template arguments (order and/or name).
→ Thus, I concur that this is invalid.
* * *
For 'T' as return value, it is in principle supportable (as Clang
demonstrates),
but one can also argue that the compiler shouldn't.
=> Close as INVALID.
