On Wednesday, 96/07/10, Leslie Rhorer wrote to David Noon about "Named
Pipes" as follows:

LR>  DN> I think the problem is that he has no buffer space for sending
LR>  DN> messages, only for receiving.
LR> 
LR>         That's odd.  I wonder why he set it up that way.  I 
LR> can see how this could possibly be a problem, but then why 
LR> does it work on his machine, and not mine?

Hi Les,

Does he have some networking software that offers "value added" pipe
support? Something along those lines might provide buffers where
vanilla OS/2 doesn't.

LR> required.  At the moment, I simply use trial and error and 
LR> a hand-set delay to have the apps start simultaneously.  

There is, I believe, a DOS interrupt available in a VDM that can
access an event semaphore created by an OS/2 process. [I don't have my
copy of Ralf Brown's list handy.] This would have much less latency for
synching DOS and OS/2 processes than a named pipe transmission.

LR> The long term answer may be to incorporate both codes into 
LR> a single code under a single compiler, and use directly 

This would be the ideal solution. The event semaphores would simply
operate across threads.

LR> interim measure, I may have the music code write to I/O 
LR> ports, which should be faster and more uniform than named 
LR> pipes.

Device driver calls are preferable to hitting I/O ports, since you
don't need to poll. Let the IRQ do the work for you. It also doesn't
need IOPL-enabled code. [Ugly 16-bit stuff.]

Regards

Dave


 * KWQ/2 1.2i * FIDO = Flames In, Drivel Out?

--- Maximus/2 3.01