Re: your mail

[Davide Libenzi]

On Mon, 25 Feb 2002, Rusty Russell wrote:

> In message <[e-mail]> you writ
> e:
> >
> >
> > On Mon, 25 Feb 2002, Rusty Russell wrote:
> > > First, fd passing sucks: you can't leave an fd somewhere and wait for
> > > someone to pick it up, and they vanish when you exit.  Secondly, you
> >
> > Yes, you can.  Please, RTFS - what is passed is not a descriptor, it's
> > struct file *.  As soon as datagram is sent, descriptors are resolved and
> > after that point descriptor table of sender (or, for that matter, survival
> > of sender) doesn't matter.
>
> Please explain how I leave a fd somewhere for other processes to grab
> it.
>
> And then please explain how they get the fd after I've exited.
>
> Al, you are one of the most unpleasant people to deal with on this
> list.  This is *not* an honor, and I beg you to consider a different
> approach in future correspondence.

Actually, this is one of Al's nicest posts :-)
You obviously can't share fd# but you can share file*
I don't know how you're going to have these semaphores 'externally visible',
if with numbers like IPC sems or if with pathnames like unix sockets ( or
something else ). But you can have internally a number/path/else -> file*
mapping and when a task attaches the sem you map the file* onto an fd# in
the task's file table. If you keep this mapping persistent ( until
explicit deletion ) the file* remain alive event with zero attached
processes. I think it's this what Al was trying to say.

- Davide