James Carlson wrote:
> Jeremy Harris writes:
> > That's not a good argument against implementing it (and changing
> > the documentation to match) purely for orthogonality.  It's not
> > impossible to imagine a UDP service where the data, while small,
> > is most conveniently held within a file.  For example, an I18N
> > message catalogue.
> >
> > A better reason for not having done it yet is lack of resource.
> 
> A better reason still is a lack of a clear problem statement.  I'm not
> completely convinced that sendfile() is the right sort of interface to
> solve the application.  It might "help," but there may well be other
> solutions that are _far_ better.

Well, it would still be _nice_ to have either |sendfile()| support for
all types of sockets and pipes in both directions to be consistent. The
current behaviour of Solaris 11/B37 was the reason that we dropped the
idea of using |sendfile()| for the /usr/bin/cat builtin in ksh93's
libcmd (the same works Ok on Linux and most other platforms which have
IPv6+|sendfile()|-support).

Maybe a solution is to implement |sendfile()| in the kernel using a
generic kernel copysendfiledata() loop if special |sendfile()| is
missing for that type of socket/pipe... 

----

Bye,
Roland

-- 
  __ .  . __
 (o.\ \/ /.o) [EMAIL PROTECTED]
  \__\/\/__/  MPEG specialist, C&&JAVA&&Sun&&Unix programmer
  /O /==\ O\  TEL +49 641 7950090
 (;O/ \/ \O;)
_______________________________________________
networking-discuss mailing list
[email protected]

Reply via email to