Re: run.c translator

[Marcus Brinkmann]

On Wed, Mar 20, 2002 at 02:08:46PM -0800, Thomas Bushnell, BSG wrote:
> I'm pretty sure that's a bug; standard pipes shouldn't be
> bidirectional.  Roland, what do you think?

In SVR4, they behave this way (acording to Stevens, who alos calls them
stream pipes).

> > What I was asking about is the interface between A and T.
> 
> Ok, so what exactly *is* between A and T, if it's not a pipe?
> 

It's the canonical file interface.  I mean, T is a translator on a node,
A opens the node with O_RDWR and starts to write to it and read from it.
Now when it has nothing more to write, I would like to signal the
translator about it, so it can close its pipe to the program P and let
A read out the remaining stuff.

It is not a pipe in the sense that it doesn't use the pflocal server or
uses any of the "pipe"ly stuff.  But the translator advertises that it
is a pipe/fifo in the stat flags, and the above usage of the "file" 
is pretty much how you use a bidirectional pipe (one filedescriptor,
two directions).  So, I guess my question is how you would signal the
ther end of a bidrectional pipe (or a streamed pipe) of the shutdown
of one direction, and because of the close resemblance between stream
pipes and socket pairs, shutdown seems to be the correct answer.
But I repeat that I am still talking about the direct communication
between A and T via the active translator in the filesystem.

Thanks,
Marcus