Re: Hurd IPC (client side)

[Ludovic Courtès]

Hi,

On Tue, Oct 15, 2002 at 10:45:45AM -0400, Neal H. Walfield wrote:
> On L4, a handle will be unique to a server which means that the client
> must store both the server's communication thread id and the handle
> id.  On Mach, a handle would be task local--it is just a normal port
> managed by Mach.

One more question. You said that a client would need to keep both a server
thread id and a handle id to identify a communication channel. On top of
Mach, we use ports to do that, and ports are the first argument of each RPC
call.

The problem is that on L4, since we would have to ids instead of one, it would
require to add an extra argument (the handle) to each RPC (unless we can make
IDL4 extract the thread id and handle id from a structure and generate the
appropriate IPC calls(s)).  If the stub generator cannot do this
"abstraction", then we would need our own abstraction on the client side: for
instance, io_read () would not be the actual stub but rather a wrapper that
would call the actual stub with an additionnal argument (the handle id).

I hope I was clear enough so that someone can enlight me or whether I missed
something somewhere. ;)

Thanks,
Ludovic.