Re: libstdc++/9580: basic_filebuf<> with custom traits_type failsto link

[Paolo Carlini <pcarlini at unitus dot it>]

Gabriel Dos Reis wrote:

> If I recall correctly, this change was made because of some
> differences in the way a narrow character stream based on
> char_traits<char> behaves and the way the general Traits may behave.
> At that point, I seem to recall that the goal was to have a general
> implementation  of that function plus the special case.  I'll have to
> go back and check my logs, unless Benjamin has a better recollection. 
Thanks Gaby. Well, I'd like to ask a little bit more help. Perhaps we are
on the right track.

The complete error message at link time is the following:

/tmp/ccVSVF9A.o: In function `std::basic_filebuf<char, 
MyTraits>::underflow()':
/tmp/ccVSVF9A.o(.gnu.linkonce.t._ZNSt13basic_filebufIc8MyTraitsE9underflowEv+0x15): 
undefined reference to `std::basic_filebuf<char, 
MyTraits>::_M_underflow_common(bool)'
/tmp/ccVSVF9A.o: In function `std::basic_filebuf<char, MyTraits>::uflow()':
/tmp/ccVSVF9A.o(.gnu.linkonce.t._ZNSt13basic_filebufIc8MyTraitsE5uflowEv+0x15): 
undefined reference to `std::basic_filebuf<char, 
MyTraits>::_M_underflow_common(bool)'

As you can see, underflow and uflow are missing the generic 
_M_underflow_common.

Now, in the current std_fstream.h there is:

 // Explicit specialization declarations, defined in src/fstream.cc.
 template<>
   basic_filebuf<char>::int_type
   basic_filebuf<char>::_M_underflow_common(bool __bump);

#ifdef _GLIBCPP_USE_WCHAR_T
 template<>
   basic_filebuf<wchar_t>::int_type
   basic_filebuf<wchar_t>::_M_underflow_common(bool __bump);
#endif

 // Generic definitions.
 template <typename _CharT, typename _Traits>
   typename basic_filebuf<_CharT, _Traits>::int_type
   basic_filebuf<_CharT, _Traits>::underflow()
   { return _M_underflow_common(false); }

 template <typename _CharT, typename _Traits>
   typename basic_filebuf<_CharT, _Traits>::int_type
   basic_filebuf<_CharT, _Traits>::uflow()
   { return _M_underflow_common(true); }

Naively, something seems suspect here: the *generic* underflow and uflow 
are still there!!

Shouldn't they be disposed too, for the time being, at least?
Paolo.