[squeak-dev] About squeak mac vm

Hi john

I got a following behavior (while been in the train = no net  
connection).
I tried to load a squeakmap package and I press on apple - .
after a while I got a wallback window. then I could only move the apple
window containing squeak is strange and extremely slow way. I had to  
kill it.

At the same time I got some other Squeak running well and which with I  
could
interact well and moving the mac ain window.

I use 3.9.1 (in fact exactly sapphire-...
        Sapphire0.1-10016-2008-03-22 (which is based on 3.9.1) and is  
available at http://sapphire.gforge.inria.fr/

and the VM 3.8.18beta1U

The bad news is that of course I could not reproduce it :(

Stef

I believe you get that behavior if ioProcessEvents is not called  
enough. What happens is checkForInterrupts has to invoke  
ioProcessEvents so that
we can grab events off the os-x event queue and pass to handler, plus  
we peform any pending display flushes.  So if this does not happen  
fast enough
you get delay in macintosh menu interaction and in moving the window.

So either there is a bug in the checkForInterrupts related to clock  
rollover?  See code below.

Or the other places we call ioProcessEvents
is via ioGetNextEvent, ioGetButtonState, ioGetKeystroke, ioMousePoint,  
ioPeekKeystroke

now the last four here  ioGetButtonState, ioGetKeystroke,  
ioMousePoint, ioPeekKeystroke  date from pre event driven VM days.
So I'll bet the Morphic polling loop which calls ioGetNextEvent was  
killed. This would result in the VM sleeping because if there
is no work to be done, like waiting on socket traffic and there is no  
morphic polling we would sleep until woken, or the next timer pop.
This would affect how often checkForInterrupts is called since it  
relies on decrementing the interruptCheckCounter,
but that only happens when Squeak is running, not sleeping, so the  
frequency of calling ioProcessEvents could be quite low.


checkForInterrupts
        "Check for possible interrupts and handle one if necessary."
        | sema now |
        self inline: false.

        "Mask so same wrapping as primitiveMillisecondClock"
        now := self ioMSecs bitAnd: MillisecondClockMask.

        self interruptCheckForced ifFalse: [
                "don't play with the feedback if we forced a check. It only makes  
life difficult"
                now - lastTick < interruptChecksEveryNms
                        ifTrue: ["wrapping is not a concern, it'll get caught quickly
                                enough. This clause is trying to keep a reasonable
                                guess of how many times per interruptChecksEveryNms we are calling
                                quickCheckForInterrupts. Not sure how effective it really is."
                                interruptCheckCounterFeedBackReset :=  
interruptCheckCounterFeedBackReset + 10]
                        ifFalse: [interruptCheckCounterFeedBackReset <= 1000
                                        ifTrue: [interruptCheckCounterFeedBackReset := 1000]
                                        ifFalse: [interruptCheckCounterFeedBackReset :=  
interruptCheckCounterFeedBackReset - 12]]].

        "reset the interrupt check counter"
        interruptCheckCounter := interruptCheckCounterFeedBackReset.

        signalLowSpace
                ifTrue: [signalLowSpace := false. "reset flag"
                        sema := self splObj: TheLowSpaceSemaphore.
                        sema = nilObj ifFalse: [self synchronousSignal: sema]].

        now < lastTick
                ifTrue: ["millisecond clock wrapped so correct the nextPollTick"
                        nextPollTick := nextPollTick - MillisecondClockMask - 1].
        now >= nextPollTick
                ifTrue: [self ioProcessEvents.
                        "sets interruptPending if interrupt key pressed"
                        nextPollTick := now + 200
                        "msecs to wait before next call to ioProcessEvents.
                        Note that strictly speaking we might need to update
                        'now' at this point since ioProcessEvents could take a
                        very long time on some platforms"].
        interruptPending
                ifTrue: [interruptPending := false.
                        "reset interrupt flag"
                        sema := self splObj: TheInterruptSemaphore.
                        sema = nilObj
                                ifFalse: [self synchronousSignal: sema]].

        nextWakeupTick ~= 0
                ifTrue: [now < lastTick
                                ifTrue: ["the clock has wrapped. Subtract the wrap
                                        interval from nextWakeupTick - this might just
                                        possibly result in 0. Since this is used as a flag
                                        value for 'no timer' we do the 0 check above"
                                        nextWakeupTick := nextWakeupTick - MillisecondClockMask - 1].
                        now >= nextWakeupTick
                                ifTrue: [nextWakeupTick := 0.
                                        "set timer interrupt to 0 for 'no timer'"
                                        sema := self splObj: TheTimerSemaphore.
                                        sema = nilObj ifFalse: [self synchronousSignal: sema]]].

        "signal any pending finalizations"
        pendingFinalizationSignals > 0
                ifTrue: [sema := self splObj: TheFinalizationSemaphore.
                        (self fetchClassOf: sema) = (self splObj: ClassSemaphore)
                                ifTrue: [self synchronousSignal: sema].
                        pendingFinalizationSignals := 0].

        "signal all semaphores in semaphoresToSignal"
        (semaphoresToSignalCountA > 0 or: [semaphoresToSignalCountB > 0])
                ifTrue: [self signalExternalSemaphores].

        "update the tracking value"
        lastTick := now



On Mar 28, 2008, at 2:15 AM, stephane ducasse wrote:

--
=
=
=
========================================================================
John M. McIntosh <[hidden email]>
Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
=
=
=
========================================================================