[Bug c++/60917] New: sub-optimal diagnostic when instantiating template

[tromey at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=60917

            Bug ID: 60917
           Summary: sub-optimal diagnostic when instantiating template
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: tromey at gcc dot gnu.org

Consider this source code:

typedef int callback ();

int f(char *);

const char *get_something ();

template<typename T, int (*F) (const T *)>
int wrapper ()
{
  return F (get_something ());
}

template<typename T1, typename T2, int (*F) (const T1 *, const T2 *)>
int wrapper ()
{
  return F (get_something ());
}

void add_callback (callback *);

void f ()
{
  add_callback (wrapper<char, f>);
}


(The bug here is that wrapper is wrapping a function that
takes a "const char *", but f takes a "char *".)

Compiling:

barimba. g++ --syntax-only t.cc
t.cc: In function ‘void f()’:
t.cc:23:33: error: no matches converting function ‘wrapper’ to type ‘int (*)()’
   add_callback (wrapper<char, f>);
                                 ^
t.cc:14:5: note: candidates are: template<class T1, class T2, int (* F)(const
T1*, const T2*)> int wrapper()
 int wrapper ()
     ^
t.cc:8:5: note:                 template<class T, int (* F)(const T*)> int
wrapper()
 int wrapper ()
     ^


I found this message pretty confusing.  For instance, it claims
to be an error during conversion, but it would have been much more clear
to me if it had been a failure to instantiate instead.  I don't know
whether that would be correct or not -- but some kind of clue leading
me to consider the argument types of "f" would have helped a lot.

In my real program, "wrapper" is much more complicated, using template
class in its body, etc, leading me to think perhaps g++ had chosen some
unusual instantiation somehow, or that some other bug had prevented
proper instantiation.  I found myself wishing there were some verbose
mode, or "#pragma tell-me-all-the-template-decisions-starting-now" or
a template instantiation debugger...