To: stik@ON-Luebeck.DE
Subject: Re: STIK: Any old Pipes?                                                        


On Tue, 15 Oct 96 19:57:36 G Guy Harrison <test@swampd wrote: 

----- snip -----> re: OS independent pipe implementation

>So, before I go and make it OS dependant here's what it does:

Please don't make it OS dependent at all, since we need it to function
identically in all TOS versions to be of maximum use for STiK.


>1)	upto 2^15 pipes - is this enough? ;-)

It certainly should be.


>2)	pipe writes are block orientated. You can stuff varying sized 
>	blocks into a pipe and the blocks will remain distinct within 
>	the pipe.

Ok, but due to point 3 below this should be transparent to an APP which
reads less than an entire block, but that raises some questions.

What happens as you reach the end of a block ?
What happens if you ask for more data than one block contains ?


>3)	pipe reads are block orientated but you can read less than a 
>	block over successive calls in case you stuffed 1 meg in but 
>	only have a 50 byte receive buffer!

Neat, but to repeat one question from above:
  What happens if you ask for an amount available only by combining blocks ?
  

>4)	no restrictions on who stuffs blocks into a pipe or who gets 
>	them out.

Good.


>5)	next read block pending can be exchanged with the one waiting 
>	behind it. Allows unwanted blocks to be fended off until 
>	desired.

Huh !?!  Aren't pipe ids sufficient to ensure correct recipients ?
Surely you don't mean that a single pipe should be used for multiple
simultaneous purposes ?
I'd think such needs would be better handled by using more pipes.

This reminds me a bit of the message system we discussed a while ago, where
the APP ids where used, thus requiring mixed usage of each channel, but with
genuine pipe ids this should not be needed.


>I've slapped a front end on it so it preserves regs, sets its own 
>stack etc. Only the method of getting at the front end address is 
>still undefined - could be via a cookie, stik or a message. Options 
>are open.
> 
>
>Any interest?

Sure, but we still have the problem of where/how we can best interweave such
'piper' code with the event loops of clients.
Have you had any new ideas on that ?

-------------------------------------------------------------------------
Regards:  Ronald Andersson                     mailto:dlanor@oden.se
                                               http://www.oden.se/~dlanor
-------------------------------------------------------------------------
