IPC problems

[Henrik Johansson]

I'm trying to write an interface to Grace (a plotting tool) in SKILL.
The communication with the Grace process is handled by SKILL's IPC
functions.

The new plot functions are working beautifully unless they sends too
much data too fast to the child process. Since the ipcWriteProcess
function is non-blocking I suspect that the data is buffered at the
parent side when the child gets more data than it can chew. If the
parent sleeps for a second and then sends another command to the child,
the missing data is written out before the new command.

Anyone know of a blocking alternative to ipcWriteProcess?

Regards

Henrik Johansson

 

[Andrew Beckett]

Hi Henrik,

I'm not sure that would solve your problem. ipcWriteProcess() does an
unbuffered write (or at least it flushes its buffers after a write) - but any
data which is unread by the child would still be sitting there in the pipe
waiting to be read, I'd have thought.

It sounds like the problem is on the child side - the child is not correctly
reading its input commands if a new command is sent whilst it is still
processing.

You may be able to get around this by getting the child to send something back
to DFII (i.e. some sort of handshaking) so that your SKILL code doesn't try
to send a command until it knows it is OK to do so.

Or you may be able to put some kind of wrapper around grace (perhaps using
expect?) to make it behave better?

Andrew.

On Thu, 02 Sep 2004 19:22:53 +0200, Henrik Johansson <henjo@ieee.org> wrote:

I'm trying to write an interface to Grace (a plotting tool) in SKILL.
The communication with the Grace process is handled by SKILL's IPC
functions.

The new plot functions are working beautifully unless they sends too
much data too fast to the child process. Since the ipcWriteProcess
function is non-blocking I suspect that the data is buffered at the
parent side when the child gets more data than it can chew. If the
parent sleeps for a second and then sends another command to the child,
the missing data is written out before the new command.

Anyone know of a blocking alternative to ipcWriteProcess?

Regards

Henrik Johansson

--
Andrew Beckett
Senior Technical Leader
Custom IC Solutions
Cadence Design Systems Ltd

 

[Henrik Johansson]

Andrew Beckett wrote:

You may be able to get around this by getting the child to send something back
to DFII (i.e. some sort of handshaking) so that your SKILL code doesn't try
to send a command until it knows it is OK to do so.

First I wrote a wrapper in perl and it worked, but with each line being
acknowledged it became very slow (like 3 lines per second).

So I tried a different approach with named pipes instead. And now with
the pipe handling the bufferering, things works perfectly. So if anyone
is interrested in a Grace interface for Skill I can publish the code in
the newsgroup.

Regards

Henrik

 

[Erik Wanta]

Henrik:
Please publish.
---
Erik

Henrik Johansson <henjo@ieee.org> wrote in message news:<chaeek$7ad$1@news.island.liu.se>...

Andrew Beckett wrote:

You may be able to get around this by getting the child to send something back
to DFII (i.e. some sort of handshaking) so that your SKILL code doesn't try
to send a command until it knows it is OK to do so.

First I wrote a wrapper in perl and it worked, but with each line being
acknowledged it became very slow (like 3 lines per second).

So I tried a different approach with named pipes instead. And now with
the pipe handling the bufferering, things works perfectly. So if anyone
is interrested in a Grace interface for Skill I can publish the code in
the newsgroup.

Regards

Henrik

 

Hi Henrik, can you please publish your solution?

Thanks.

Noong Sabado, Setyembre 04 2004 21:43:39 UTC+8, si Erik Wanta ay sumulat:

Henrik:
Please publish.
---
Erik

Henrik Johansson <henjo@ieee.org> wrote in message news:<chaeek$7ad$1@news.island.liu.se>...
Andrew Beckett wrote:

You may be able to get around this by getting the child to send something back
to DFII (i.e. some sort of handshaking) so that your SKILL code doesn't try
to send a command until it knows it is OK to do so.

First I wrote a wrapper in perl and it worked, but with each line being
acknowledged it became very slow (like 3 lines per second).

So I tried a different approach with named pipes instead. And now with
the pipe handling the bufferering, things works perfectly. So if anyone
is interrested in a Grace interface for Skill I can publish the code in
the newsgroup.

Regards

Henrik