Re: PATCH: Avoid excessive flushing on lower-layer handle

[Jason Merrill <jason at redhat dot com>]

>>>>> "Loren" == Loren James Rittle <rittle@latour.rsch.comm.mot.com> writes:

>   	// Make sure that the internal buffer resyncs its idea of
>   	// the file position with the external file.
> ! 	if (__testput)
>   	  {
>   	    // Need to restore current position after the write.
>   	    off_type __off = _M_out_cur - _M_out_end;
> ! 	    _M_really_overflow(); // _M_file.sync() will be called within
>   	    if (__off)
>   	      _M_file.seekoff(__off, ios_base::cur);
>   	  }
> + 	else
> + 	  _M_file.sync();

Shouldn't this be unconditional?  i.e. drop the "else"?

>   	  // Convert pending sequence to external representation, output.
> + 	  // If eof, then just attempt sync.
>   	  if (!traits_type::eq_int_type(__c, traits_type::eof()))
>   	    {
>   	      char_type __pending = traits_type::to_char_type(__c);
>   	      _M_convert_to_external(&__pending, 1, __elen, __plen);
  
> ! 	      // User code must flush when switching modes (thus don't sync).
> ! 	      if (__elen == __plen)
> ! 		{
> ! 		  _M_set_indeterminate();
> ! 		  __ret = traits_type::not_eof(__c);
> ! 		}
> ! 	    }
> ! 	  else if (!_M_file.sync())
>   	    {
>   	      _M_set_indeterminate();
>   	      __ret = traits_type::not_eof(__c);

Why do we need an fflush here at all?

> If the user must now flush to switch modes, this should always be a safe
> optimization.

My patch didn't change that; that will be a later, trunk-only patch.  But I
think your change is still safe.

Jason