[Vm-dev] Unexpected #resume behavior

[Jaromir Matas]

Hi Eliot, all,

I'm back to my original question what is the *desirable* semantics of #resume and #suspend for Squeak:

We’ve discussed #suspend and you’ve implemented the very elegant backing of the blocked process before the send wait.

Now we have this unexpected #resume behavior plainly letting a process continue regardless of whether it’s being blocked.

I’ve checked VW and VA:

VW’s #suspend removes the process from the conditional variable’s list and #resume fails an attempt to resume a blocked process. The examples below behave as expected.

VA on the other hand treat blocked processes equally as suspended processes which seems to me as if they ignored/redefined the idea of conditional variables; i.e. suspending a blocked process is a no-op and resuming a blocked process simply removes it from the semaphore and puts it in the runnable list.

It seems to me Squeak sits inconsistently somewhere in-between. From #resume and #suspend comments being identical to VW’s I guess VW’s behavior was the original intention.

best,
~~~
^[^    Jaromir

From: Jaromir Matas<mailto:mail at jaromir.net>
Sent: Thursday, January 13, 2022 19:29
To: vm-dev at lists.squeakfoundation.org<mailto:vm-dev at lists.squeakfoundation.org>; Eliot Miranda<mailto:eliot.miranda at gmail.com>
Subject: Unexpected #resume behavior

Hi Eliot, all,

I'm baffled by this example:

p := [Semaphore new wait] fork.
Processor yield.
p resume explore

I expected to get an error but the process just gets out of the semaphore and finishes happily.

#resume comment says:
                Primitive. Allow the process that the receiver represents to continue. Put
                 the receiver in line to become the activeProcess.  *Fail if the receiver is
                 already waiting in a queue (in a Semaphore or ProcessScheduler)*.  Fail if
                the receiver's suspendedContext is not a context.

The following works as expected - p sits at the semaphore:

p := [Semaphore new wait] fork.
Processor yield.
p explore

This works as well (with the new VM) - p backs up and sits before the wait:

p := [Semaphore new wait] fork.
Processor yield.
p suspend.
p explore

And this works too indeed - p sits at the semaphore again after suspend.

p := [Semaphore new wait] fork.
Processor yield.
p suspend.
p resume explore


Is this a bug? Or is the comment just outdated?

I've been trying to figure out possible ways how to prevent resuming a process being terminated (other than setting its suspendedContext to nil) and this unexpected behavior gave me a real hard time :)

Many thanks for your help.

Best regards,
Jaromir

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.squeakfoundation.org/pipermail/vm-dev/attachments/20220114/d110ce9d/attachment.html>