'PROCESS-WAIT called when scheduling not allowed'

Hi all,

LispWorks (LW 4.3.7 on 32bit Linux x86).  I know its pretty old but its
what I have to work with until I get a chance to upgrade....

I'm writing a CAPI app that creates 4 "worker processes", each of which
uses open-pipe with :direction :io to invoke a child task.  Each each LW
worker process reads the stdout of its respective child task (the latter
being a little C app).  Commands to the child tasks are routed thru
mailboxs owned by the respective LW worker process, which writes them to
the child task's stdin.

The top-level app reads bmp files out of a hot directory and submits
them to worker processes.  Since there are 4 processes, each running the
same child task, then four bmp's can be in process at once- taking
advantage of the multicore system I'll be running it on.

So the problem is I'm getting this fatal error

"Wait function signalled non-repeating error
'PROCESS-WAIT called when scheduling not allowed'"

in each of the LW processes after the first few lines are read from
their child task's stdout.  I am using (peek-char nil) to detect the
presence of chars, then (read-line) to wait for & read an entire line of
text.  The child task guarantees line-oriented Lisp formatted sexp's
output to its stdio.  Note I do get correct behavior for the first few
lines, then the error is asserted.

The debugger shows the error is coming from the process-wait invocation-
not somewhere else.  Inside the process-wait polling function I use a
(peek-char nil) to detect chars coming from the child task, so the LW
task sleeps until it has something to process.

If I replace the (read-line) related code with a (when (peek-char nil
...) followed by a simple (read ..), then the error goes away.  The
problem then is the process seems to eventually get stuck in the (read),
I believe because there is whitespace that peek-char sees but there is
not a complete sexp to read.

I've tried variations with (read-char-no-hang) in a loop to accumulate a
line & had the same problem as (read-line). 

I was hoping that any experienced in the nuances of process-wait as it
relates to the reader might comment on the possibility of (read-line) or
(read-char*) causing stream-related or process-related side-effects that
(read) doesn't.  Since the child task's output is line-oriented,
(read-line) seems a better choice.

I'd be happy to supply code but figured it would be better to try and
ask useful questions first.

Thanks,

Greg

Re: 'PROCESS-WAIT called when scheduling not allowed'

Hi all,

Found the problem but I don't understand why it was a problem.  It seems
to have been caused by the (peek-char) inside the process-wait polling
function.  When I replaced it with a (read-char-no-hang) & saved the
char retrieved so I could prepend it to the results of the subsequent
(read-line), then all the stdio acts as expected.

Perhaps theres something inside (peek-char) that can interact with
(process-wait) which was messing me up.

Regards,

Greg



On 05/20/2011 04:00 AM, Greg Menke wrote:
> Hi all,
>
> LispWorks (LW 4.3.7 on 32bit Linux x86).  I know its pretty old but its
> what I have to work with until I get a chance to upgrade....
>
> I'm writing a CAPI app that creates 4 "worker processes", each of which
> uses open-pipe with :direction :io to invoke a child task.  Each each LW
> worker process reads the stdout of its respective child task (the latter
> being a little C app).  Commands to the child tasks are routed thru
> mailboxs owned by the respective LW worker process, which writes them to
> the child task's stdin.
>
> The top-level app reads bmp files out of a hot directory and submits
> them to worker processes.  Since there are 4 processes, each running the
> same child task, then four bmp's can be in process at once- taking
> advantage of the multicore system I'll be running it on.
>
> So the problem is I'm getting this fatal error
>
> "Wait function signalled non-repeating error
> 'PROCESS-WAIT called when scheduling not allowed'"
>
> in each of the LW processes after the first few lines are read from
> their child task's stdout.  I am using (peek-char nil) to detect the
> presence of chars, then (read-line) to wait for & read an entire line of
> text.  The child task guarantees line-oriented Lisp formatted sexp's
> output to its stdio.  Note I do get correct behavior for the first few
> lines, then the error is asserted.
>
> The debugger shows the error is coming from the process-wait invocation-
> not somewhere else.  Inside the process-wait polling function I use a
> (peek-char nil) to detect chars coming from the child task, so the LW
> task sleeps until it has something to process.
>
> If I replace the (read-line) related code with a (when (peek-char nil
> ..) followed by a simple (read ..), then the error goes away.  The
> problem then is the process seems to eventually get stuck in the (read),
> I believe because there is whitespace that peek-char sees but there is
> not a complete sexp to read.
>
> I've tried variations with (read-char-no-hang) in a loop to accumulate a
> line & had the same problem as (read-line). 
>
> I was hoping that any experienced in the nuances of process-wait as it
> relates to the reader might comment on the possibility of (read-line) or
> (read-char*) causing stream-related or process-related side-effects that
> (read) doesn't.  Since the child task's output is line-oriented,
> (read-line) seems a better choice.
>
> I'd be happy to supply code but figured it would be better to try and
> ask useful questions first.
>
> Thanks,
>
> Greg
>