🚀 go-pugleaf

In article <12964@ecs.soton.ac.uk> ams90@ecs.soton.ac.uk (SHEHAB AM) writes:
>
>Hi everyone,
>
>  I have come across a little problem; it goes as follows:
>
>I need to create lots of nearly identical worker processes
>dynamically.  Then once they have executed they return their
>results to their parent processes and should die.  It is when the die
>that my problem starts; how do I get rid of the process from the heap?
>The processes are not started by ProcPar...etc but by ProcRun for
>various reasons.  This means that I can not use ProcAllocClean which
>proved to crash the processors sometimes. I have tried free(process
>pointer) to no avail.

The next release of the ANSI C toolset relaxes the restriction that
ProcAllocClean may only be used with synchronous processes.  A new function
has been introduced which allows a parent process to wait for an asychronous
child process to terminate (similar to wait in a UNIX shell script) to ensure
that ProcAllocClean is not applied to an executing process.

Please contact your normal supplier for release dates of the next ANSI C
toolset as I'm afraid I don't know them.


-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Nigel Holder, Software Group, INMOS Ltd		Tel +44 454 616616
1000 Aztec West					Fax +44 454 617910
Almondsbury					UK: nigel@inmos.co.uk
Bristol BS12 4SQ, UK.				US: nigel@inmos.com

		"Lost in time like tears in rain.  Time to die"    Blade Runner

In <12964@ecs.soton.ac.uk> ams90@ecs.soton.ac.uk (SHEHAB AM) writes:


>I need to create lots of nearly identical worker processes
>dynamically.  Then once they have executed they return their
>results to their parent processes and should die.  It is when the die
>that my problem starts; how do I get rid of the process from the heap?
>The processes are not started by ProcPar...etc but by ProcRun for
>various reasons.  This means that I can not use ProcAllocClean which
>proved to crash the processors sometimes. I have tried free(process
>pointer) to no avail.

>Any ideas will be greatly appreciated.

Thanks for the resposes I have had from Mark Debbage,Alfred Rizzi, Rob Kurver, Peter Burgess and Andy Pepperdine who have supplied a number of ideas.  The consencus seeams to be an implementation as follows:

1. Pass into every process a channel, a different channel for each process.
Call this channel the "finish-channel".

2. At the end of each asynchronous process, write something to its
finish-channel.

3. When you wish to check whether a process is finished, do a ProcAlt on
its finish-channel. But when the finish-channel is selected, do NOT read
the input on it. Just take the response that it is ready to transfer.

4. At this point, the asynchronous process is known to be locked waiting to
complete an output. If the corresponding input is never performed, the process
cannot continue. So it is at this point that all the process's memory can
be freed. It is not in any execution queue.

5. In order to free all the memory, the processes must not use any default
allocations, so use ProcInit and specify the actual memory area it should
use for the heap.

6. Each asynchronous process must also free all the memory it uses for
its own processing before writing to its finish-channel, ie. put its own
house in order before leaving.

many thanks to Andy Pepperdine for the above steps.

AMS....