2014-06-23 22:50 GMT+02:00 Nicolas Cellier <nicolas.cellier.aka.nice@gmail.com>:



2014-06-23 22:38 GMT+02:00 Norbert Hartl <norbert@hartl.name>:

Hmmm,

I'm trying to write a test for it and I wonder why��

testDynamicVariableAccessFromDifferentProcess
| process ��sem1 result �� ��|
sem1 := Semaphore new.
process := [ ��
TestDynamicVariable��
value: 123��
during: [ sem1 wait ] ] fork.
Processor activeProcess��
evaluate: [ ��result := TestDynamicVariable value ]��
onBehalfOf: process.
sem1 signal.
self assert: result = 123

does not work. The variable accessing is in��

DynamicVariable>>#value: anObject during: aBlock
| p oldValue |
p := Processor activeProcess.
oldValue := (p psValueAt: index) ifNil: [ self default ].
^ [��
p psValueAt: index put: anObject.
aBlock value ] ensure: [ p psValueAt: index put: oldValue ]

Any ideas?

But what happens when you immediately fork, are you sure the forked process immediately pre-empts the activeProcess?
What if you introduce Processor activeProcess yield after the fork?
��

Ah, I just tried, it's Processor yield.

[Transcript cr; show: 'A' ] fork.
Processor yield.
Transcript cr; show: 'B' .

produces the required transcripting order.
Without yielding, B comes before A.

��
��
Norbert

Am 23.06.2014 um 21:29 schrieb Eliot Miranda <eliot.miranda@gmail.com>:




On Mon, Jun 23, 2014 at 12:05 PM, Norbert Hartl��<norbert@hartl.name>��wrote:
Hi Eliot,

thank you very much. I imported your changes in a pharo 3 image and it works awesome. So I'm preparing a slice for pharo.��

Thanks again, a very annoying problem seems to be solved,

glad to hear it, and thanks for your prompting me as it's now in Squeak trunk too.
��

Norbert

Am 23.06.2014 um 19:29 schrieb Eliot Miranda <eliot.miranda@gmail.com>:

and here are the changes I've just committed to Squeak trunk.


On Mon, Jun 23, 2014 at 10:05 AM, Eliot Miranda<eliot.miranda@gmail.com>��wrote:
Hi Norbert,

�� ����[ let me try again. ��never try and get code out too early in the morning ;-) ]

�� ����it is the debugger that needs fixing, not your code !! :-). ��The debugger needs to respect process identity. ��Andreas and I (mostly Andreas) came up with the following changes at Qwaq. ��Your message is a good reminder that I need to add this to Squeak asap.

The idea is for Process to have an additional inst var 'effectiveProcess' that holds the actual process running code. ��For the most part this is self, but in the debugger we substitute the process being debugged:

Process methods for accessing
effectiveProcess
"effectiveProcess is a mechanism to allow process-faithful debugging. ��The debugger executes code
��on behalf of processes, so unless some effort is made the identity of Processor activeProcess is not
��correctly maintained when debugging code. ��The debugger uses evaluate:onBehalfOf: to assign the
��debugged process as the effectiveProcess of the process executing the code, preserving process
��identity."
^effectiveProcess ifNil: [self]

then the relevant methods in Process and processorScheduler defer to effectiveProcess, e.g.

ProcessorScheduler methods for process state change
terminateActive
"Terminate the process that is currently running."

activeProcess effectiveProcess terminate

and the debugging methods use evaluate:onBehalfOf: to install the process being debugged:

Process methods for private
evaluate: aBlock onBehalfOf: aProcess
"Evaluate aBlock setting effectiveProcess to aProcess. ��Used
��in the execution simulation machinery to ensure that
��Processor activeProcess evaluates correctly when debugging."
| oldEffectiveProcess |
oldEffectiveProcess := effectiveProcess.
effectiveProcess := aProcess.
^aBlock ensure: [effectiveProcess := oldEffectiveProcess]

Process methods for changing suspended state
step

^Processor activeProcess
evaluate: [suspendedContext := suspendedContext step]
onBehalfOf: self

stepToCallee
"Step until top context changes"

Processor activeProcess
evaluate:
[| ctxt |
ctxt := suspendedContext.
[ctxt == suspendedContext] whileTrue: [
suspendedContext := suspendedContext step]]
onBehalfOf: self.
^suspendedContext

etc. ��Changes from a Qwaq image attached.

HTH


On Mon, Jun 23, 2014 at 4:50 AM, Norbert Hartl��<norbert@hartl.name>wrote:
In my code I'm using a DynamicVariable to request a context object when needed. Until now I knew the name DynamicVariable only from seaside. There it is called WADynamicVariable and it is an exception. So I blindly assumed the pharo DynamicVariable works the same.
I thought this might be a good optimization not to travel the stack all the time but put in the process.
Now that I am using it I can see the difference. I find it real hard using it because I don't know how to debug/step in code. DynamicVariable is a process specific variable but as soon as a debugger opens it is very likely to be in another process. This makes stepping in method using the DynamicVariable impossible. The only way round is to set break points after the dynamic lookup and step from there. But this feels just wrong.
What would be the best way to have DynamicVariable and be able to debug anything? Or is there a variant that uses the stack instead of the "active" process?

thanks,

Norbert





--��
best,
Eliot



--��
best,
Eliot
<trunk4.6EffectiveProcessMethods.st>




--��
best,
Eliot