On Sat, Dec 13, 2008 at 12:07:51AM +0100, Roland Mainz wrote:
> > > Is there anything special needed ?
> >
> > the system having /dev/fd/N is one part of it
> > the other part is, given fd==3 already open for reading, does
> > fd = open("/dev/fd/3", O_RDONLY);
> > simply do
> > fd = dup(3);
>
> Somehow I suspect this is the case for Solaris... does the POSIX
> standard say anything about this (I suspect it leaves this item
> "undefined") and how do other platforms, e.g. FreeBSD, Darwin, Linux
> behave ?
>
> > if so then 3 and fd will have the same seek offsets
> > the the open() does a real open then 3 and fd will have
> > independent file struct with independent seek offsets
> >
> > or if you really believe the system provides open() semantics
> > then the test could be flawed
>
> I am not sure... that's why I've CC:'ed the *...@opensolaris.org lists for
> feedback...
I've tested it using ksh93 itself:
% integer i=1000
% while ((i<1500))
> do
> ((i+=1))
> print $i
> done > /tmp/foo
% redirect 9<>/tmp/foo
% (read line <#((200)) ; print $line ; read line
</dev/fd/9 ; print $line) <&9
1041
1042
%
So, yes, open(2) of /dev/fd/N is as a dup(2) of N.
Nico
--
