2014-07-27 7:09 GMT+02:00 Ben Coman <btc@openinworld.com>:

2014-07-25 16:56 GMT+02:00 Ben Coman <btc@openinworld.com>:

...

Over the last few days I have been looking deeper into the image locking when suspending a process. It is an interesting rabbit hole [1] that leads to pondering the Delay machinery, that leads to some VM questions.

When pressing the interrupt key it seems to always opens the debugger with the following call stack. Semaphore>>critical: 'self wait' BlockClosure>>ensure: 'self valueNoContextSwitch' Semaphore>>critical: 'ensure: [ caught ifTrue: [self signal]] Delay>>schedule 'AccessProtect critical: [' Delay>>wait 'self schedule' WorldState>>interCyclePause:

Nicolai Hess wrote:

Hi Ben,

I am on Windows too :( So, the fixes does not work (not always) on winddows too. But at least they make it less probable to occure, but it still happens. The most distracting thing is, after the first ui lock, pressing alt+dot, closing the debuggers, pressing alt+dot .... and trying to close the very first debugger, after that, it all works. The UI is responsive again and suspending the process does not block the ui anymore. It "looks like" supsending the process reactivates another process that blocks the UI. And as soon as I terminate this process (alt+dot, close debugger ...) all works. But I really don't know.

Nicolai

Hi Nicolai, I have something for you to try.

With the problem seeming to be stuck in the semaphore critical section, I read around on mutexes and semaphores and found [1] enlightening, which says: "The correct use of a semaphore is for signaling from one task to another. A mutex is meant to be taken and released, always in that order, by each task that uses the shared resource it protects. By contrast, tasks that use semaphores either signal or wait—not both. [...] Any two RTOS tasks that operate at **different priorities** and coordinate via a mutex, create the opportunity for priority inversion [...while...] semaphores [...] do not cause priority inversion when used for signaling." [1]

To summarize with an example, here's how to use a mutex: /* Task 1 */ mutexWait(mutex_mens_room); // Safely use shared resource mutexRelease(mutex_mens_room);

/* Task 2 */ mutexWait(mutex_mens_room); // Safely use shared resource mutexRelease(mutex_mens_room);

By contrast, you should always use a semaphore like this: /* Task 1 - Producer */ semPost(sem_power_btn); // Send the signal

/* Task 2 - Consumer */ semPend(sem_power_btn); // Wait for signal

I am out of my depth here, but in light [1] the way the current code seems to mix semaphore/mutex concepts with Semaphore>>forMutalExclusion and Semaphore>>critical: gave me some concern. Anyway, the delay scheduling code seemed more like a producer/consumer signaling problem than a shared resource mutal exclusion problem, so I rewrote it like the former and fixes the problem without causing any *obvious* problems. However I lack experience here and it could definitely use some review. Can you take the attached zip file and...

1. File in "Delay class-new delay code part 1.st" Defines #enableNewDelayCode and #enableNewDelayCode: Manually define the referenced class variables

2. File in "Delay class-new delay code part 2.st" Modifies #handleTimerEvent, guarded by enableNewDelayCode

3. File in "Delay-new delay code part 3.st" Modifies #schedule and #unschedule, guarded by enableNewDelayCode

4. File in "BackgroundWorkDisplayMorph.st"

5. Save checkpoint Image to return to between trials.

6. Open System Browser on BackgroundWorkDisplayMorph>>initialize

Trial 1. BackgroundWorkDisplayMorph new openInWorld.

Trial 2. Delay enableNewDelayCode: true. BackgroundWorkDisplayMorph new openInWorld.

If if its deemed worthwhile, I'll open a ticket to track this more formally. One thing needed is to considered performance impact, I haven't so far.

cheers -ben

[1] http://www.barrgroup.com/Embedded-Systems/How-To/RTOS-Mutex-Semaphore [2] http://nerdfortress.com/2011/02/18/you-say-semaphore-i-say-mutex/

I notice...

...

Delay class >> initialize TimingSemaphore := (Smalltalk specialObjectsArray at: 30). and... Delay class >> startTimerEventLoop TimingSemaphore := Semaphore new. which seems incongruous that TimingSemaphore is set in differently. So while I presume this critical stuff all works fine, just in an exotic way, my entropy-guarding-neuron would just like confirm this is so.

--------------

In Delay class >> handleTimerEvent the comment says... "Handle a timer event.... -a timer signal (not explicitly specified)" ...is that event perhaps a 'tick' generated periodically by the VM via that item from specialObjectArray ? Or is there some other mechanism ?

--------------

[1] http://www.urbandictionary.com/define.php?term=Rabbit+Hole cheers -ben

P.S. I've left the following for some initial context as I change the subject. btw Nicolai, I confirm that my proposed fixes only work on Windows, not Mavericks (and I haven't checked Linux).

Nicolai Hess wrote:

Hi ben, thank you for looking at this.

2014-07-22 20:17 GMT+02:00 <btc@openinworld.com>:

...

I thought this might be interesting to learn, so I've gave it a go. I had some success at the end, but I'll give a progressive report.

First I thought I'd try moving the update of StringMorph outside the worker-process using a Morph's #step method as follows...

Morph subclass: #BackgroundWorkDisplayMorph instanceVariableNames: 'interProcessString stringMorph' classVariableNames: '' category: 'BenPlay' "---------"

BackgroundWorkDisplayMorph>>initializeMorph self color: Color red. stringMorph := StringMorph new. self addMorphBack: stringMorph. self extent:(300@50). "---------"

BackgroundWorkDisplayMorph>>newWorkerProcess ^[ | work | work := 0. [ 20 milliSeconds asDelay wait. work := work + 1. interProcessString := work asString. ] repeat. ] newProcess. "---------"

BackgroundWorkDisplayMorph>>step stringMorph contents: interProcessString. "---------"

BackgroundWorkDisplayMorph>>stepTime ^50 "---------"

BackgroundWorkDisplayMorph>>initialize | workerProcess running | super initialize. self initializeMorph.

workerProcess := self newWorkerProcess. running := false.

self on: #mouseUp send: #value to: [ (running := running not) ifTrue: [ workerProcess resume. self color: Color green. ] ifFalse: [ workerProcess suspend. self color: Color red. ] ] "---------"

But evaluating "BackgroundWorkDisplayMorph new openInWorld" found this exhibited the same problematic behavior you reported... Clicking on the morph worked a few times and then froze the UI until Cmd-. pressed a few times.

...

However I found the following never locked the GUI.

BackgroundWorkDisplayMorph>>initialize "BackgroundWorkDisplayMorph new openInWorld" | workerProcess running | super initialize. self initializeMorph.

workerProcess := self newWorkerProcess. running := false.

[ [ (running := running not) ifTrue: [ workerProcess resume. self color: Color green ] ifFalse: [ workerProcess suspend. self color: Color red ]. 10 milliSeconds asDelay wait. ] repeat ] fork. "---------"

This locks the UI as well. Not every timet hough. I did this 5 times, every time in a freshly loaded image and it happens two times.

...

So the problem seemed to not be with #suspend/#resume or with the shared variable /interProcessString/. Indeed, since in the worker thread /interProcessString/ is atomically assigned a copy via #asString, and the String never updated, I think there is no need to surround use of it with a critical section.

The solution then was to move the "#resume/#suspend" away from the "#on: #mouseUp send: #value to:" as follows...

BackgroundWorkDisplayMorph>>initialize "BackgroundWorkDisplayMorph new openInWorld" | workerProcess running lastRunning | super initialize. self initializeMorph.

workerProcess := self newWorkerProcess. lastRunning := running := false.

[ [ lastRunning = running ifFalse: [ running ifTrue: [ workerProcess resume ] ifFalse: [ workerProcess suspend ]. lastRunning := running. ]. 10 milliSeconds asDelay wait. ] repeat ] fork.

self on: #mouseUp send: #value to: [ (running := running not) ifTrue: [ self color: Color green. ] ifFalse: [ self color: Color red. ] ] "---------"

And this too :(

...

And finally remove the busy loop.

BackgroundWorkDisplayMorph>>initialize "BackgroundWorkDisplayMorph new openInWorld" | workerProcess running lastRunning semaphore | super initialize. self initializeMorph.

workerProcess := self newWorkerProcess. lastRunning := running := false. semaphore := Semaphore new.

[ [ semaphore wait. running ifTrue: [ workerProcess resume ] ifFalse: [ workerProcess suspend ]. ] repeat ] fork.

self on: #mouseUp send: #value to: [ (running := running not) ifTrue: [ self color: Color green. ] ifFalse: [ self color: Color red. ]. semaphore signal. ] "---------"

And this locks the UI too. (Loaded the code 20 times, every time after a fresh image start up. Two times I got a locked ui after the first two clicks). And I don't understand this code :)

...

Now I can't say how close that is to how it "should" be done. Its the first time I used sempahores and just what I discovered hacking around. But hey! it works :)

cheers -ben

Nicolai Hess wrote:

I am still struggling with it.

Any ideas?

2014-07-09 11:19 GMT+02:00 Nicolai Hess <nicolaihess@web.de>:

...

2014-07-09 2:07 GMT+02:00 Eliot Miranda <eliot.miranda@gmail.com>:

Hi Nicolai,

...

On Tue, Jul 8, 2014 at 7:19 AM, Nicolai Hess <nicolaihess@web.de> wrote:

...

I want to create a process doing some work and call #changed on a Morph. I want to start/suspend/resume or stop this process. But sometimes, suspending the process locks the UI-Process, and I don't know why. Did I miss something or do I have to care when to call suspend?

Wrapping the "morph changed" call in UIManager default defer:[ morph changed]. Does not change anything.

Here is an example to reproduce it. Create the process, call resume, call supsend. It works, most of the time, but sometimes, calling suspend locks the ui.

p:=[[true] whileTrue:[ Transcript crShow: (DateAndTime now asString). 30 milliSeconds asDelay wait]] newProcess.

p resume.

...

p suspend.

If you simply suspend this process at random form a user-priority process you'll never be able to damage the Delay machinery you're using, but chances are you'll suspend the process inside the critical section that Transcript uses to make itself thread-safe, and that'll lock up the Transcript.

Thank you Eliot yes I guessed it locks up the critical section, but I hoped with would not happen if I the use UIManager defer call.

...

ThreadSafeTranscript>>nextPutAll: value accessSemaphore critical: [stream nextPutAll: value]. ^value

So instead you need to use a semaphore. e.g.

| p s wait | s := Semaphore new. p:=[[true] whileTrue:[wait ifTrue: [s wait]. Transcript crShow: (DateAndTime now asString). 30 milliSeconds asDelay wait]] newProcess. wait := true. 30 milliSeconds asDelay wait. wait := false. s signal

etc...

Is this a common pattern I can find in pharos classes. Or I need some help understanding this. The semaphore wait/signal is used instead of process resume/suspend?

What I want is a process doing repeatly some computation, calls or triggers an update on a morph, and I want to suspend and resume this process.

I would stop this discussion if someone tells me, "No your are doing it wrong, go this way ..", BUT what strikes me: in this example, that reproduces my problem more closely:

|p m s running| running:=false. m:=Morph new color:Color red. s:= StringMorph new. m addMorphBack:s. p:=[[true]whileTrue:[20 milliSeconds asDelay wait. s contents:(DateAndTime now asString). m changed]] newProcess. m on:#mouseUp send:#value to:[ running ifTrue:[p suspend. m color:Color red.] ifFalse:[p resume.m color:Color green.]. running := running not]. m extent:(300@50). m openInWorld

clicking on the morph will stop or resume the process, if it locks up I can still press alt+dot -> - a Debugger opens but the UI is still not responsive. I can click with the mouse on the debuggers close icon. - nothing happens, as the UI is still blocked. - pressing alt+Dot again, the mouse click on the close icon is processed and the first debugger window closes - maybe other debuggers open.

Repeating this steps, at some time the system is *fully* responsive again! And miraculously, it works after that without further blockages. What's happening here?

Nicolai

...

HTH

regards

...

Nicolai

-- best, Eliot

Interesting, it works. So, it is the Delay>>schedule implementation? Because it uses a #critical block for the wait instead of wait/signal only? Nicolai