Re: [boost] Re: generic polyvalent input/output ['pickle']

From: Jens Maurer (Jens.Maurer_at_[hidden])
Date: 2000-11-17 15:55:24

Asger Alstrup Nielsen wrote:
> >  - I think that the "buffer" layer can be removed entirely.
> > I believe the correct way is to write a streambuf for C files
> > and fixed-size memory areas.  (std::stringstream is close (but
> > not fixed-size) to a memory area, btw.)
> 
> Could you elaborate on this idea a bit more?
> 
> I certainly agree that the buffer layer is poorly documented, and it
> should be great to clean this area up, since it's the weakest point of
> the concept as I see it.

I am suggesting to use the C++ Standard iostreams interface
instead of the buffer layer.  It already has all the required
functions: Pushing some data into it, flushing it to disk, advanced
caching is possible, reading enough data from the disk on-demand etc.

And it re-uses a standard component instead of re-inventing I/O.

However, the current iostreams only cover one of the three buffer
alternatives for XTL: iostream.  The C++ iostreams library is
extensible, though, so you can write your own specialized C++ stream
which handles the "cfile_buffer" and "mem_buffer" cases.  The proper
way to do that is to write your own streambuf.  For cfile_buffer,
one could argue that C-style FILE*'s have been superseded by C++
iostreams anyway, so a replacement is not required.  For mem_buffer,
as said before, stringstream goes a long way (including automatic
dynamic size).

Please ask if I'm still not making sense to you :-)

Jens Maurer