[Bug c++/45873] New: Parameter packs not expanding consistently in function return types

[Bug c++/45873] New: Parameter packs not expanding consistently in function return types

[sausage at tehsausage dot com]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

           Summary: Parameter packs not expanding consistently in function
                    return types
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
        AssignedTo: unassigned@gcc.gnu.org
        ReportedBy: sausage@tehsausage.com


Non-variadic templates passed to a function as "template <class ...> class"
have inconsistent behavior when provided with a parameter pack as an argument
and used as a return type than when used in the body of the function in the
same way. I am unsure on the expected behavior or whether passing templates in
such a way is even allowed.


template <template <class ...> class T, class ...Args>
auto foo(Args... args) -> T<Args...>
{ return T<Args...>(args...); }

template <template <class ...> class T, class ...Args>
auto bar(Args... args) -> T<int, double>
{ return T<Args...>(args...); }

int main()
{
   foo<pair>(1, 2.0); // error: no matching function for call to 'foo(int,
double)'
   bar<pair>(1, 2.0); // OK, returns pair<int, double>
}


The following work-around seems to work:

template <template <class ...> class T, class ...Args> struct expand
{ typedef T<Args...> type; };

template <template <class ...> class T, class ...Args>
auto baz(Args... args) -> typename expand<T, Args...>::type
{ return T<Args...>(args...); }

int main()
{
   baz(1, 2.0); // OK, returns pair<int, double>
}

[Bug c++/45873] Parameter packs not expanding consistently in function return types

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
      Known to fail|                            |4.4.3, 4.5.1, 4.6.0

--- Comment #1 from Jonathan Wakely <redi at gcc dot gnu.org> 2010-10-04 08:11:07 UTC ---
(You didn't say which version this PR is for, although it doesn't work with any
I tried)

Here's a self-contained testcase, with a fix so the workaround actually works
(it was missing the template argument on baz<pair>)

template<class, class>
struct pair
{
    template<class T1, class U1>
        pair(T1&& t, U1&& u) { }
};

template <template <class ...> class T, class ...Args>
auto foo(Args... args) -> T<Args...>
{ return T<Args...>(args...); }

template <template <class ...> class T, class ...Args>
auto bar(Args... args) -> T<int, double>
{ return T<Args...>(args...); }


// The following work-around seems to work:
template <template <class ...> class T, class ...Args> struct expand
{ typedef T<Args...> type; };

template <template <class ...> class T, class ...Args>
auto baz(Args... args) -> typename expand<T, Args...>::type
{ return T<Args...>(args...); }


int main()
{
   // error: no matching function for call to 'foo(int, double)'
   foo<pair>(1, 2.0);
   bar<pair>(1, 2.0); // OK, returns pair<int, double>
   baz<pair>(1, 2.0); // OK, returns pair<int, double>
}


I think deduction fails for foo<pair>(1, 2.0), though I'm not sure if it
should.

For baz<pair> deduction succeeds, the template parameters appear in a
nested-name-specifier which is a non-deduced context.

[Bug c++/45873] Parameter packs not expanding consistently in function return types

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

Jonathan Wakely <redi at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jason at gcc dot gnu.org

--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> 2011-05-16 12:01:31 UTC ---
Jason, should deduction succeed here?

template<class, class>
struct pair
{
    template<class T1, class U1>
        pair(T1&& t, U1&& u) { }
};

template <template <class...> class T, class ...Args>
auto foo(Args... args) -> T<Args...>
{ return T<Args...>(args...); }

int main()
{
   // error: no matching function for call to 'foo(int, double)'
   foo<pair>(1, 2.0);
}


It looks as though substituting T=pair fails because T's single parameter pack
fails to match pair's two parameters.

If pair is a variadic template with exactly one parameter pack , like p0 below,
then it matches T:

template<class...> struct p0 { };

template<class, class...> struct p1 { };

template<class, class, class...> struct p2 { };

template <template <class...> class T, class ...Args>
auto foo(Args... args) -> T<Args...>
{ return T<Args...>(); }

int main()
{
   foo<p0>(1, 2.0);  // ok
   foo<p1>(1, 2.0);  // error
   foo<p2>(1, 2.0);  // error
}

Is this behaviour correct?

Clang accepts all three calls

[Bug c++/45873] Parameter packs not expanding consistently in function return types

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

--- Comment #3 from Jason Merrill <jason at gcc dot gnu.org> 2011-05-16 15:42:36 UTC ---
This seems like a dup of 35722.

[Bug c++/45873] Parameter packs not expanding consistently in function return types

[redi at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

--- Comment #4 from Jonathan Wakely <redi at gcc dot gnu.org> 2011-05-16 17:04:12 UTC ---
(In reply to comment #3)
> This seems like a dup of 35722.

Ah yes, it definitely is for the call to foo<p1> in my second example in
comment 2: modifying pt.c to always complain at the FIXME in
coerce_template_parms produces the "sorry" for the call to foo<p1>

45873.c2.cc: In function 'int main()':
45873.c2.cc:14:18: sorry, unimplemented: cannot expand 'Args ...' into a
fixed-length argument list
45873.c2.cc:14:18: error: no matching function for call to 'foo(int, double)'
45873.c2.cc:14:18: note: candidate is:
45873.c2.cc:8:6: note: template<template<class ...> class T, class ... Args>
T<Args ...> foo(Args ...)
45873.c2.cc:15:18: error: no matching function for call to 'foo(int, double)'
45873.c2.cc:15:18: note: candidate is:
45873.c2.cc:8:6: note: template<template<class ...> class T, class ... Args>
T<Args ...> foo(Args ...)

But that "sorry" isn't reached for foo<p2> or for any foo<pair> in the other
testcases.

I don't know if that means something else is going on there, or if it's just an
undiagnosed case of the same situation.

[Bug c++/45873] [C++0x] Parameter packs not expanding consistently in function return types

[jason at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=45873

Jason Merrill <jason at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |DUPLICATE
   Target Milestone|---                         |4.7.0

--- Comment #5 from Jason Merrill <jason at gcc dot gnu.org> 2012-04-16 02:10:37 UTC ---
Works in 4.7.

*** This bug has been marked as a duplicate of bug 35722 ***