The Trunk: Kernel-nice.1386.mcz

Nicolas Cellier uploaded a new version of Kernel to project The Trunk:
http://source.squeak.org/trunk/Kernel-nice.1386.mcz

==================== Summary ====================

Name: Kernel-nice.1386
Author: nice
Time: 14 April 2021, 2:56:48.575456 pm
UUID: e7612c01-14d8-0940-87fe-45e2fb4d2719
Ancestors: Kernel-mt.1385, Kernel-ct.1383, Kernel-ct.1364

Merge commit. Note that Kernel-ct.1383 supersedes Kernel-ct.1382 and removes the annoying warning.

Kernel-mt.1385:
        Clean up and comment on the idiom #perform:orSendTo:.

Kernel-ct.1383:
        Fixes simulation of the #perform:... primitives 83, 84, and 100 for all edge cases. If the primitive is called with the wrong arguments, the primitive must fail but not the simulator.

Kernel-ct.1364:
        Fixes primitive comments in BlockClosure >> (cull:){3,4}.

=============== Diff against Kernel-mt.1385 ===============

Item was changed:
  ----- Method: BlockClosure>>cull:cull:cull: (in category 'evaluating') -----
  cull: firstArg cull: secondArg cull: thirdArg
  "Activate the receiver, with three or less arguments."
+ <primitive: 204> "Handle the three argument case primitively"
- <primitive: 204> "Handle the two argument case primitively"
  numArgs >= 2 ifTrue: [
  numArgs >= 3 ifTrue: [ ^self value: firstArg value: secondArg value: thirdArg ].
  ^self value: firstArg value: secondArg ].
  numArgs = 1 ifTrue: [ ^self value: firstArg ].
  ^self value!

Item was changed:
  ----- Method: BlockClosure>>cull:cull:cull:cull: (in category 'evaluating') -----
  cull: firstArg cull: secondArg cull: thirdArg cull: fourthArg
  "Activate the receiver, with four or less arguments."
+ <primitive: 205> "Handle the four argument case primitively"
- <primitive: 205> "Handle the two argument case primitively"
  numArgs >= 3 ifTrue: [
  numArgs >= 4 ifTrue: [
  ^self value: firstArg value: secondArg value: thirdArg value: fourthArg ].
  ^self value: firstArg value: secondArg value: thirdArg ].
  numArgs = 2 ifTrue: [ ^self value: firstArg value: secondArg ].
  numArgs = 1 ifTrue: [ ^self value: firstArg ].
  ^self value!

Item was changed:
  ----- Method: Context>>doPrimitive:method:receiver:args: (in category 'private') -----
  doPrimitive: primitiveIndex method: meth receiver: receiver args: arguments
  "Simulate a primitive method whose index is primitiveIndex.  The simulated receiver and
  arguments are given as arguments to this message. If successful, push result and return
  resuming context, else ^ {errCode, PrimitiveFailToken}. Any primitive which provokes
  execution needs to be intercepted and simulated to avoid execution running away."
 
  | value |
  "Judicious use of primitive 19 (a null primitive that doesn't do anything) prevents
  the debugger from entering various run-away activities such as spawning a new
  process, etc.  Injudicious use results in the debugger not being able to debug
  interesting code, such as the debugger itself.  Hence use primitive 19 with care :-)"
  "SystemNavigation new browseAllSelect: [:m| m primitive = 19]"
+ primitiveIndex = 19 ifTrue: [
- (primitiveIndex = 19 or: [primitiveIndex >= 85 and: [primitiveIndex <= 88] "control primitives"]) ifTrue: [
  [self notify: ('The code being simulated is trying to control a process ({1}). Process controlling cannot be simulated. If you proceed, things may happen outside the observable area of the simulator.' translated format: {meth reference})]
  ifCurtailed: [self push: nil "Cheap fix of the context's internal state"]].
 
  ((primitiveIndex between: 201 and: 222)
  and: [(self objectClass: receiver) includesBehavior: BlockClosure]) ifTrue:
  [(primitiveIndex = 206
   or: [primitiveIndex = 208]) ifTrue: "[Full]BlockClosure>>valueWithArguments:"
  [^receiver simulateValueWithArguments: arguments first caller: self].
  ((primitiveIndex between: 201 and: 209) "[Full]BlockClosure>>value[:value:...]"
   or: [primitiveIndex between: 221 and: 222]) ifTrue: "[Full]BlockClosure>>valueNoContextSwitch[:]"
  [^receiver simulateValueWithArguments: arguments caller: self]].
 
  primitiveIndex = 83 ifTrue: "afr 9/11/1998 19:50" "Object>>perform:[with:...]"
+ [| selector |
+ selector := arguments at: 1 ifAbsent:
+ [^ self class primitiveFailTokenFor: #'bad argument'].
+ arguments size - 1 = selector numArgs ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad number of arguments'].
+ ^self send: selector to: receiver with: arguments allButFirst].
- [^self send: arguments first to: receiver with: arguments allButFirst].
  primitiveIndex = 84 ifTrue: "afr 9/11/1998 19:50 & eem 8/18/2009 17:04" "Object>>perform:withArguments:"
+ [| selector args |
+ arguments size = 2 ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad argument'].
+ selector := arguments first.
+ args := arguments second.
+ args isArray ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad argument'].
+ args size = selector numArgs ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad number of arguments'].
+ ^self send: selector to: receiver with: args].
- [^self send: arguments first to: receiver with: (arguments at: 2) lookupIn: (self objectClass: receiver)].
  primitiveIndex = 100 ifTrue: "eem 8/18/2009 16:57" "Object>>perform:withArguments:inSuperclass:"
+ [| rcvr selector args superclass |
+ arguments size
+ caseOf: {
+ [3] -> [
+ rcvr := receiver.
+ selector := arguments first.
+ args := arguments second.
+ superclass := arguments third].
+ [4] -> ["mirror primitive"
+ rcvr := arguments first.
+ selector := arguments second.
+ args := arguments third.
+ superclass := arguments fourth] }
+ otherwise: [^ self class primitiveFailTokenFor: #'bad argument'].
+ args isArray ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad argument'].
+ args size = selector numArgs ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad number of arguments'].
+ ((self objectClass: rcvr) includesBehavior: superclass) ifFalse:
+ [^ self class primitiveFailTokenFor: #'bad argument'].
+ ^self send: selector to: rcvr with: args lookupIn: superclass].
- [^self send: arguments first to: receiver with: (arguments at: 2) lookupIn: (arguments at: 3)].
 
  "Mutex>>primitiveEnterCriticalSection
  Mutex>>primitiveTestAndSetOwnershipOfCriticalSection"
  (primitiveIndex = 186 or: [primitiveIndex = 187]) ifTrue:
+ [| effective |
+ effective := Processor activeProcess effectiveProcess.
+ "active == effective"
+ value := primitiveIndex = 186
+ ifTrue: [receiver primitiveEnterCriticalSectionOnBehalfOf: effective]
+ ifFalse: [receiver primitiveTestAndSetOwnershipOfCriticalSectionOnBehalfOf: effective].
- [value := primitiveIndex = 186
- ifTrue: [receiver primitiveEnterCriticalSectionOnBehalfOf: Processor activeProcess]
- ifFalse: [receiver primitiveTestAndSetOwnershipOfCriticalSectionOnBehalfOf: Processor activeProcess].
  ^(self isPrimFailToken: value)
  ifTrue: [value]
  ifFalse: [self push: value]].
 
  primitiveIndex = 188 ifTrue: "Object>>withArgs:executeMethod:
  CompiledMethod class>>receiver:withArguments:executeMethod:
  VMMirror>>ifFail:object:with:executeMethod: et al"
  [| n args methodArg thisReceiver |
  ((n := arguments size) between: 2 and: 4) ifFalse:
  [^self class primitiveFailTokenFor: #'unsupported operation'].
  ((self objectClass: (args := arguments at: n - 1)) == Array
   and: [(self objectClass: (methodArg := arguments at: n)) includesBehavior: CompiledMethod]) ifFalse:
  [^self class primitiveFailTokenFor: #'bad argument'].
  methodArg numArgs = args size ifFalse:
  [^self class primitiveFailTokenFor: #'bad number of arguments'].
  thisReceiver := arguments at: n - 2 ifAbsent: [receiver].
  methodArg primitive > 0 ifTrue:
  [methodArg isQuick ifTrue:
  [^self push: (methodArg valueWithReceiver: thisReceiver arguments: args)].
  ^self doPrimitive: methodArg primitive method: meth receiver: thisReceiver args: args].
  ^Context
  sender: self
  receiver: thisReceiver
  method: methodArg
  arguments: args].
 
  primitiveIndex = 118 ifTrue: "[receiver:]tryPrimitive:withArgs:; avoid recursing in the VM"
  [(arguments size = 3
   and: [(self objectClass: arguments second) == SmallInteger
   and: [(self objectClass: arguments last) == Array]]) ifTrue:
  [^self doPrimitive: arguments second method: meth receiver: arguments first args: arguments last].
  (arguments size = 2
  and: [(self objectClass: arguments first) == SmallInteger
  and: [(self objectClass: arguments last) == Array]]) ifFalse:
  [^self class primitiveFailTokenFor: nil].
  ^self doPrimitive: arguments first method: meth receiver: receiver args: arguments last].
 
  value := primitiveIndex = 120 "FFI method"
  ifTrue: [(meth literalAt: 1) tryInvokeWithArguments: arguments]
  ifFalse:
  [primitiveIndex = 117 "named primitives"
  ifTrue: [self tryNamedPrimitiveIn: meth for: receiver withArgs: arguments]
  ifFalse: "should use self receiver: receiver tryPrimitive: primitiveIndex withArgs: arguments but this is only in later VMs (and appears to be broken)"
  [receiver tryPrimitive: primitiveIndex withArgs: arguments]].
 
  ^(self isPrimFailToken: value)
  ifTrue: [value]
  ifFalse: [self push: value]!

Hi Nicolas,

I see that you reverted Kernel-ct.1382 (which has been merged into the Trunk
earlier via Kernel-mt.1383). Was this a merge error or an intended rollback?

Assuming the latter, I would like to discuss this again, please. :-)
I already noticed by myself that the control primitive warning occurred a
bit too often in situations where you would not expect it (for instance,
when stepping through "Object sourceCodeAt: #at:"). Still, do we indeed do
not want to get informed about any attempt by the debugged process to
execute logic in another process?
Also, the simulator is not only used by the debugger but also in
miscellaneous non-interactive situations, including MessageTally, some
KernelTests, and SimulationStudio where the warning might be even more
relevant because the assumption or the contract is that the passed logic
will be simulated completely.
Would it be a compromise for you to signal a custom
EscapeFromSimulationWarning (better name proposals are welcome :-) ) instead
and to ignore them in the Debugger?

I am looking forward to your reply! :-)

Best,
Christoph



-----
Carpe Squeak!
--
Sent from: http://forum.world.st/Squeak-Dev-f45488.html

PS: I also see that you touched the simulation of criticalSection primitives
(primitiveIndex = 186 ...) again. Again, was this intended or an incident?
:-)

(OT: We are really treading on each other's toes in this method ... What can
we learn from this? Is it too large and should be split up? Or should we
improve our tooling support for resolving merge conflicts?)

Best,
Christoph



-----
Carpe Squeak!
--
Sent from: http://forum.world.st/Squeak-Dev-f45488.html

Hi Christoph,
Hmm, there must have been some confusion on my side.
I've tried to carefully review the remaining contributions in the inbox.
To my understanding Kernel-ct.1383 fixes more simulation cases than
Kernel-ct.1382.
Since
- the tests simulating primitives 83, 84 & 100 were failing,
- and that a more recent inbox contribution was trying to fix that,
I thought that merging this contribution was a good idea.
It apparently was fixing two tests out of three after trial.

Also, to my experience, generating a warning when we simulate
    primitiveIndex >= 85 and: [primitiveIndex <= 88] "control primitives"])
is super annoying! It always happens when running some kernel tests
and forces me to switch warning off altogether.
But switching warning off is not a good solution (and it makes some
more tests fail !).
Kernel-ct.1383 being more recent, I thought that the removal of
warnings was intentional...

I saw later that you removed <primitive: 19> from newProcess... SHould we?

My understanding now is that you need this warning in some specific
context. Is that it?
We can't fix one usage by breaking another. We have to find a solution
that fits both purposes.

Le ven. 16 avr. 2021 à 15:40, Marcel Taeumel <[hidden email]> a écrit :

Hi Christoph,
Hmm, there must have been some confusion on my side.
I've tried to carefully review the remaining contributions in the inbox.
To my understanding Kernel-ct.1383 fixes more simulation cases than
Kernel-ct.1382.
Since
- the tests simulating primitives 83, 84 & 100 were failing,
- and that a more recent inbox contribution was trying to fix that,
I thought that merging this contribution was a good idea.
It apparently was fixing two tests out of three after trial.

Also, to my experience, generating a warning when we simulate
    primitiveIndex >= 85 and: [primitiveIndex <= 88] "control primitives"])
is super annoying! It always happens when running some kernel tests
and forces me to switch warning off altogether.
But switching warning off is not a good solution (and it makes some
more tests fail !).
Kernel-ct.1383 being more recent, I thought that the removal of
warnings was intentional...

I saw later that you removed <primitive: 19> from newProcess... SHould we?

My understanding now is that you need this warning in some specific
context. Is that it?
We can't fix one usage by breaking another. We have to find a solution
that fits both purposes.

Le ven. 16 avr. 2021 à 15:40, Marcel Taeumel <[hidden email]> a écrit :
>
> Hi Christoph,
>
> the commit message in Kernel-nice.1386 reads: "[...] Note that Kernel-ct.1383 supersedes Kernel-ct.1382 and removes the annoying warning. [...]"
>
> I think the problem was that both ct.1382 and ct.1382 change the same method but do not actually consider each other. With our current tools, sub-method merging is rather tricky and hence the author should take care that this does not happen. :-)
>
> Please review the current state of Context >> #doPrimitive:method:receiver:args: and upload the missing lines again into the inbox if necessary.
>
> Best,
> Marcel
>
> Am 16.04.2021 15:16:18 schrieb Christoph Thiede <[hidden email]>:
>
> PS: I also see that you touched the simulation of criticalSection primitives
> (primitiveIndex = 186 ...) again. Again, was this intended or an incident?
> :-)
>
> (OT: We are really treading on each other's toes in this method ... What can
> we learn from this? Is it too large and should be split up? Or should we
> improve our tooling support for resolving merge conflicts?)
>
> Best,
> Christoph
>
>
>
> -----
> Carpe Squeak!
> --
> Sent from: http://forum.world.st/Squeak-Dev-f45488.html
>
>

Hi Christoph,

Le lun. 19 avr. 2021 à 13:32, Thiede, Christoph
<[hidden email]> a écrit :
>
> Hi Nicolas,
>
>
> in general, I use to submit one inbox version per proposal/fix/change. They all branch from the main Trunk line so I consider each of them something that you would call a Pull Request in the GitHub world. This also means that no inbox proposal is to be considered deprecated only because I upload a newer version. When I indeed intend to replace the former version, I usually add a notice "replaces version Package-ct.xxx" to the message of the new version. :-)
>
> I think our special problem in this case was that it is not a straightforward task to merge multiple changes of a single method with our tooling.
>

Yes, changing several times the same method is necessarily a conflict
for our tools.
IMO it's a good thing, because it will require human attention for resolution.
Merging based on line editions is not the right solution IMO.
If both editions modify the temps, they will conflict anyway.

Even if methods are atomic pieces, two modifications of the same
method are not necessarily exclusive.
It happens that we updated a comment in one branch, or renamed a
message, inst var or class in the other, etc...
But here, I failed to identify this situation.
I might have been biased by the desire to see the warnings go ;)

>
> I see the problem with annoying warnings, which is why I proposed to create a new subclass for these warnings! Then we could easily ignore these warnings when they are raised inside of the debugger. For other use cases such as automated tests or SimulationStudio, however, we would still see them.
>

Yes, this could work.
I also wondered if we could disable warnings at specific send sites on
user request (instead of disabling all the Warnings).
A good way to exercise new super powers for the Sorcerer's Apprentices ;)

Hi Christoph,

Le lun. 19 avr. 2021 à 13:32, Thiede, Christoph
<[hidden email]> a écrit :
>
> Hi Nicolas,
>
>
> in general, I use to submit one inbox version per proposal/fix/change. They all branch from the main Trunk line so I consider each of them something that you would call a Pull Request in the GitHub world. This also means that no inbox proposal is to be considered deprecated only because I upload a newer version. When I indeed intend to replace the former version, I usually add a notice "replaces version Package-ct.xxx" to the message of the new version. :-)
>
> I think our special problem in this case was that it is not a straightforward task to merge multiple changes of a single method with our tooling.
>

Yes, changing several times the same method is necessarily a conflict
for our tools.
IMO it's a good thing, because it will require human attention for resolution.
Merging based on line editions is not the right solution IMO.
If both editions modify the temps, they will conflict anyway.

Even if methods are atomic pieces, two modifications of the same
method are not necessarily exclusive.
It happens that we updated a comment in one branch, or renamed a
message, inst var or class in the other, etc...
But here, I failed to identify this situation.
I might have been biased by the desire to see the warnings go ;)

>
> I see the problem with annoying warnings, which is why I proposed to create a new subclass for these warnings! Then we could easily ignore these warnings when they are raised inside of the debugger. For other use cases such as automated tests or SimulationStudio, however, we would still see them.
>

Yes, this could work.
I also wondered if we could disable warnings at specific send sites on
user request (instead of disabling all the Warnings).
A good way to exercise new super powers for the Sorcerer's Apprentices ;)

>
> > I saw later that you removed <primitive: 19> from newProcess... SHould we?
>
>
> I think we should. Either a simulator client is interested to be informed about attempts of the code to escape from simulation, or it isn't. In every case, creating a process is an absolutely boring event with regard to that interest, but all actions such as resuming, terminating, or signaling a process could be of interest.
>
> > Please review the current state of Context >> #doPrimitive:method:receiver:args: and upload the missing lines again into the inbox if necessary.
>
> I'll put that onto my list, but I've myself lost track of the changes regarding primitive 186 et al. I'm not the author of these changes ...
>
> Best,
> Christoph
>
> ________________________________
> Von: Squeak-dev <[hidden email]> im Auftrag von Nicolas Cellier <[hidden email]>
> Gesendet: Freitag, 16. April 2021 19:34:20
> An: The general-purpose Squeak developers list
> Betreff: Re: [squeak-dev] The Trunk: Kernel-nice.1386.mcz
>
> Hi Christoph,
> Hmm, there must have been some confusion on my side.
> I've tried to carefully review the remaining contributions in the inbox.
> To my understanding Kernel-ct.1383 fixes more simulation cases than
> Kernel-ct.1382.
> Since
> - the tests simulating primitives 83, 84 & 100 were failing,
> - and that a more recent inbox contribution was trying to fix that,
> I thought that merging this contribution was a good idea.
> It apparently was fixing two tests out of three after trial.
>
> Also, to my experience, generating a warning when we simulate
>     primitiveIndex >= 85 and: [primitiveIndex <= 88] "control primitives"])
> is super annoying! It always happens when running some kernel tests
> and forces me to switch warning off altogether.
> But switching warning off is not a good solution (and it makes some
> more tests fail !).
> Kernel-ct.1383 being more recent, I thought that the removal of
> warnings was intentional...
>
> I saw later that you removed <primitive: 19> from newProcess... SHould we?
>
> My understanding now is that you need this warning in some specific
> context. Is that it?
> We can't fix one usage by breaking another. We have to find a solution
> that fits both purposes.
>
> Le ven. 16 avr. 2021 à 15:40, Marcel Taeumel <[hidden email]> a écrit :
> >
> > Hi Christoph,
> >
> > the commit message in Kernel-nice.1386 reads: "[...] Note that Kernel-ct.1383 supersedes Kernel-ct.1382 and removes the annoying warning. [...]"
> >
> > I think the problem was that both ct.1382 and ct.1382 change the same method but do not actually consider each other. With our current tools, sub-method merging is rather tricky and hence the author should take care that this does not happen. :-)
> >
> > Please review the current state of Context >> #doPrimitive:method:receiver:args: and upload the missing lines again into the inbox if necessary.
> >
> > Best,
> > Marcel
> >
> > Am 16.04.2021 15:16:18 schrieb Christoph Thiede <[hidden email]>:
> >
> > PS: I also see that you touched the simulation of criticalSection primitives
> > (primitiveIndex = 186 ...) again. Again, was this intended or an incident?
> > :-)
> >
> > (OT: We are really treading on each other's toes in this method ... What can
> > we learn from this? Is it too large and should be split up? Or should we
> > improve our tooling support for resolving merge conflicts?)
> >
> > Best,
> > Christoph
> >
> >
> >
> > -----
> > Carpe Squeak!
> > --
> > Sent from: http://forum.world.st/Squeak-Dev-f45488.html
> >
> >
>
>

Hi all,

I just wanted to pick up this item from my list again and add a
SimulationSideEffectWarning (I think this is a better name than my former
proposal, EscapeFromSimulationWarning) to
#doPrimitive:method:receiver:args:.

As discussed below, it should allow pedantic clients of the simulator to
avoid situations in which the code to be executed escapes from the control
of the simulator, e.g. when it uses #fork. On the other hand, unaware
clients such as the debugger should not be disturbed by unimportant
warnings. However, critical attempts should still be reported, i.e. when a
debugger reaches a simulation guard (primitive 19).

However, I'm not yet sure in which exact situations the warning should be
caught/suppressed and in which not. Here is a list of possible situations:

- Use the stepping buttons in a debugger: SUPPRESS (except simulation guard)
- Return entered value in a debugger: SUPPRESS (except simulation guard)
- Terminate a process: SUPPRESS
- Context class >> #trace.../#tally...: NOTIFY
- MessageTally: NOTIFY
- Context class >> #runSimulated:: NOTIFY
- Morphic "debug action" (of a button, menu item, etc.): SUPPRESS
- Debug it: SUPPRESS (what about simulation guards?)
- (external project) SimulationStudio/Sandbox: NOTIFY (!)

Based on this collection, I considered adding a handler in the public
Process selectors for debugging (protocols: *System-debugging and 'changing
suspended state'). This would also emphasize the different roles of Context
and Process in the debugging machinery (Process also holds the
responsibility for process-faithful debugging).

Still, there is one edge case that would violate this rule: Imagine a
subclass of Debugger that would be side-effect aware, i.e. care about
EscapeFromSimulationWarnings. Since the debugger only talks to the Process,
it would not be able to handle these warnings following my proposal from
above. This is not only a fictive experiment but I'm indeed considering a
SandboxDebugger at the moment which would actually need this. For this
concrete case, of course, I could hack my exception handler into my Context
subclass, too, but the existence of this example makes we wonder whether my
proposal from above is actually useful.

What is your opinion on this matter? Would you agree to my list of aware and
non-aware simulation clients? When do you think we should suppress
EscapeFromSimulationWarnings and when not?

Best,
Christoph



-----
Carpe Squeak!
--
Sent from: http://forum.world.st/Squeak-Dev-f45488.html

Hi all!

After rethinking the case again, here is my proposed solution (please find
it in the attachment).
SimulationSideEffectWarning.cs
<http://forum.world.st/file/t372205/SimulationSideEffectWarning.cs>  

In a nutshell, I decided on a strict separation between the different causes
for a side effect warning since control primitives and simulation guards
actually reside on different levels of abstraction. They only have in common
that the user of the simulator should maybe know about them. The detailed
changelog is in the postscript of the changelog, here is a copy:


SimulationSideEffectWarning.cs wrote
Please let us know how I can improve it or just merge it directly! :-)

Best,
Christoph



-----
Carpe Squeak!
--
Sent from: http://forum.world.st/Squeak-Dev-f45488.html

It is extremely confusing that Nabble strips of the revision number of the changeset upon upload. :-)

---

Community service, here is the inlined diff:

"Change Set:        SimulationSideEffectWarning
Date:            9 May 2021
Author:            Christoph Thiede

<your descriptive text goes here>"

Warning subclass: #SimulationSideEffectWarning
    instanceVariableNames: 'primitiveIndex sender suppressed'
    classVariableNames: ''
    poolDictionaries: ''
    category: 'Kernel-Exceptions'

I am signaled to notify the client of a simulation operation (i.e., a sender of Context) about potential side effects that might occur when resuming the simulation. See Context >> #doPrimitive:method:receiver:args:, #messageText, and Parser >> #simulationGuard for more information.

doPrimitive: primitiveIndex method: meth receiver: receiver args: arguments
    "Simulate a primitive method whose index is primitiveIndex. The simulated receiver and
     arguments are given as arguments to this message. If successful, push result and return
     resuming context, else ^ {errCode, PrimitiveFailToken}. Any primitive which provokes
     execution needs to be intercepted and simulated to avoid execution running away."

    | value |
    "Judicious use of primitive 19 (a null primitive that doesn't do anything) prevents
     the debugger from entering various run-away activities such as spawning a new
     process, etc. Injudicious use results in the debugger not being able to debug
     interesting code, such as the debugger itself. Hence use primitive 19 with care :-)"
    "SystemNavigation new browseAllSelect: [:m| m primitive = 19]"
    primitiveIndex = 19 ifTrue: [
        [self notify: ('The code being simulated is trying to control a process ({1}). Process controlling cannot be simulated. If you proceed, things may happen outside the observable area of the simulator.' translated format: {meth reference})]
            ifCurtailed: [self push: nil "Cheap fix of the context's internal state"]].
    "Test for unsimulatable side effects (that is, code that will be triggered in the image outside of the simulator range). This includes simulation guards, which are traditionally flagged using primitive 19 (a null primitive that doesn't do anything), as well as certain control primitives that might trigger code on other processes. If a side effect is detected, raise a warning to give the user a chance to cancel the operation."
    "#(19 87) do: [:primitive | self systemNavigation browseAllSelect: [:m | m primitive = primitive]]"
    (primitiveIndex = 19 "simulationGuard" or: [primitiveIndex = 87 "primitiveResume"]) ifTrue: [
        [SimulationSideEffectWarning signalForPrimitive: primitiveIndex sender: self]
            ifCurtailed: [self push: nil "Cheap fix of the context's internal state. Note that unwinding the receiver -- so that the next step would invoke the primitive again -- would be challenging due to to the variety of senders to this method."]].
    
    ((primitiveIndex between: 201 and: 222)
     and: [(self objectClass: receiver) includesBehavior: BlockClosure]) ifTrue:
        [(primitiveIndex = 206
         or: [primitiveIndex = 208]) ifTrue:                        "[Full]BlockClosure>>valueWithArguments:"
            [^receiver simulateValueWithArguments: arguments first caller: self].
         ((primitiveIndex between: 201 and: 209)             "[Full]BlockClosure>>value[:value:...]"
         or: [primitiveIndex between: 221 and: 222]) ifTrue: "[Full]BlockClosure>>valueNoContextSwitch[:]"
            [^receiver simulateValueWithArguments: arguments caller: self]].

    primitiveIndex = 83 ifTrue: "afr 9/11/1998 19:50" "Object>>perform:[with:...]"
        [| selector |
        selector := arguments at: 1 ifAbsent:
            [^ self class primitiveFailTokenFor: #'bad argument'].
        arguments size - 1 = selector numArgs ifFalse:
            [^ self class primitiveFailTokenFor: #'bad number of arguments'].
        ^self send: selector to: receiver with: arguments allButFirst].
    primitiveIndex = 84 ifTrue: "afr 9/11/1998 19:50 & eem 8/18/2009 17:04" "Object>>perform:withArguments:"
        [| selector args |
        arguments size = 2 ifFalse:
            [^ self class primitiveFailTokenFor: #'bad argument'].
        selector := arguments first.
        args := arguments second.
        args isArray ifFalse:
            [^ self class primitiveFailTokenFor: #'bad argument'].
        args size = selector numArgs ifFalse:
            [^ self class primitiveFailTokenFor: #'bad number of arguments'].
        ^self send: selector to: receiver with: args].
    primitiveIndex = 100 ifTrue: "eem 8/18/2009 16:57" "Object>>perform:withArguments:inSuperclass:"
        [| rcvr selector args superclass |
        arguments size
            caseOf: {
                [3] -> [
                    rcvr := receiver.
                    selector := arguments first.
                    args := arguments second.
                    superclass := arguments third].
                [4] -> ["mirror primitive"
                    rcvr := arguments first.
                    selector := arguments second.
                    args := arguments third.
                    superclass := arguments fourth] }
            otherwise: [^ self class primitiveFailTokenFor: #'bad argument'].
        args isArray ifFalse:
            [^ self class primitiveFailTokenFor: #'bad argument'].
        args size = selector numArgs ifFalse:
            [^ self class primitiveFailTokenFor: #'bad number of arguments'].
        ((self objectClass: rcvr) includesBehavior: superclass) ifFalse:
            [^ self class primitiveFailTokenFor: #'bad argument'].
        ^self send: selector to: rcvr with: args lookupIn: superclass].

    "Mutex>>primitiveEnterCriticalSection
     Mutex>>primitiveTestAndSetOwnershipOfCriticalSection"
    (primitiveIndex = 186 or: [primitiveIndex = 187]) ifTrue:
        [| effective |
         effective := Processor activeProcess effectiveProcess.
         "active == effective"
         value := primitiveIndex = 186
                    ifTrue: [receiver primitiveEnterCriticalSectionOnBehalfOf: effective]
                    ifFalse: [receiver primitiveTestAndSetOwnershipOfCriticalSectionOnBehalfOf: effective].
         ^(self isPrimFailToken: value)
            ifTrue: [value]
            ifFalse: [self push: value]].

    primitiveIndex = 188 ifTrue:    "Object>>withArgs:executeMethod:
                                    CompiledMethod class>>receiver:withArguments:executeMethod:
                                    VMMirror>>ifFail:object:with:executeMethod: et al"
        [| n args methodArg thisReceiver |
         ((n := arguments size) between: 2 and: 4) ifFalse:
            [^self class primitiveFailTokenFor: #'unsupported operation'].
         ((self objectClass: (args := arguments at: n - 1)) == Array
         and: [(self objectClass: (methodArg := arguments at: n)) includesBehavior: CompiledMethod]) ifFalse:
            [^self class primitiveFailTokenFor: #'bad argument'].
         methodArg numArgs = args size ifFalse:
            [^self class primitiveFailTokenFor: #'bad number of arguments'].
         thisReceiver := arguments at: n - 2 ifAbsent: [receiver].
         methodArg primitive > 0 ifTrue:
            [methodArg isQuick ifTrue:
                [^self push: (methodArg valueWithReceiver: thisReceiver arguments: args)].
             ^self doPrimitive: methodArg primitive method: meth receiver: thisReceiver args: args].
         ^Context
            sender: self
            receiver: thisReceiver
            method: methodArg
            arguments: args].

    primitiveIndex = 118 ifTrue: "[receiver:]tryPrimitive:withArgs:; avoid recursing in the VM"
        [(arguments size = 3
         and: [(self objectClass: arguments second) == SmallInteger
         and: [(self objectClass: arguments last) == Array]]) ifTrue:
            [^self doPrimitive: arguments second method: meth receiver: arguments first args: arguments last].
         (arguments size = 2
         and: [(self objectClass: arguments first) == SmallInteger
         and: [(self objectClass: arguments last) == Array]]) ifFalse:
            [^self class primitiveFailTokenFor: nil].
         ^self doPrimitive: arguments first method: meth receiver: receiver args: arguments last].

    value := primitiveIndex = 120 "FFI method"
                ifTrue: [(meth literalAt: 1) tryInvokeWithArguments: arguments]
                ifFalse:
                    [primitiveIndex = 117 "named primitives"
                        ifTrue: [self tryNamedPrimitiveIn: meth for: receiver withArgs: arguments]
                        ifFalse: "should use self receiver: receiver tryPrimitive: primitiveIndex withArgs: arguments but this is only in later VMs (and appears to be broken)"
                            [receiver tryPrimitive: primitiveIndex withArgs: arguments]].

    ^(self isPrimFailToken: value)
        ifTrue: [value]
        ifFalse: [self push: value]

invokeSimulationGuard
    <simulationGuard>
    "Nothing to see here, please move along!"
    ^ 42

testSimulationSideEffectWarningControl

    | warning |
    [Context runSimulated: [[] fork]] on: SimulationSideEffectWarning do: [:ex |
        warning := ex].
    
    self assert: warning notNil.
    self assert: warning isControlPrimitive.
    self assert: warning suppressed.

testSimulationSideEffectWarningGuard

    | warning |
    [Context runSimulated: [self invokeSimulationGuard]] on: SimulationSideEffectWarning do: [:ex |
        warning := ex].
    
    self assert: warning notNil.
    self assert: warning isSimulationGuard.
    self deny: warning suppressed.

testSimulationSideEffectWarningSuppress

    self
        shouldnt: [(SimulationSideEffectWarning forPrimitive: 42 sender: thisContext)
            suppress;
            defaultAction] raise: UnhandledWarning;
        should: [(SimulationSideEffectWarning forPrimitive: 42 sender: thisContext)
            unsuppress;
            defaultAction] raise: UnhandledWarning.

activeController: aController
    "Set aController to be the currently active controller. Give the user
    control in it."
    <primitive: 19> "Simulation guard"
    "Set aController to be the currently active controller. Give the user control in it."
    <simulationGuard>

    activeController := aController.
    (activeController == screenController)
        ifFalse: [self promote: activeController].
    activeControllerProcess :=
            [activeController startUp.
            self searchForActiveController] newProcess.
    activeControllerProcess priority: Processor userSchedulingPriority.
    activeControllerProcess resume

scheduleActive: aController
    "Make aController be scheduled as the active controller. Presumably the
    active scheduling process asked to schedule this controller and that a
    new process associated this controller takes control. So this is the last act
    of the active scheduling process."
    <primitive: 19> "Simulation guard"
    "Make aController be scheduled as the active controller. Presumably the active scheduling process asked to schedule this controller and that a new process associated this controller takes control. So this is the last act of the active scheduling process."
    <simulationGuard>

    self scheduleActiveNoTerminate: aController.
    Processor terminateActive

handleLabelUpdatesIn: aBlock whenExecuting: aContext
    "Send the selected message in the accessed method, and regain control
    after the invoked method returns."
    
    ^aBlock
        on: Notification
        do: [:ex|
            (ex tag isArray
             and: [ex tag size = 2
             and: [(ex tag first == aContext or: [ex tag first hasSender: aContext])]])
                ifTrue:
                    [self labelString: ex tag second description.
                     ex resume]
                ifFalse:
                    [ex pass]]
                    [ex pass]]
        on: SimulationSideEffectWarning
        do: [:ex |
            ex isControlPrimitive ifTrue: [ex unsuppress].
            ex pass]

simulationGuard
    "primitive 19 is a null primitive that always fails. Just a marker for the simulator."
    <pragmaParser>

    self addPragma: (Pragma keyword: #primitive: arguments: #(19)).
    
    self advance.
    ^ true

isControlPrimitive
    "See StackInterpreter class>>#initializePrimitiveTable."

    ^ self primitive between: 80 and: 89

isSimulationGuard
    "See Parser >> #simulationGuard."

    ^ self primitive = 19

primitive

    ^ primitiveIndex

sender

    ^ sender

suppress

    suppressed := true.

suppressed

    ^ suppressed ifNil: [self isSimulationGuard not]

unsuppress

    suppressed := false.

primitive: anInteger sender: senderContext

    primitiveIndex := anInteger.
    sender := senderContext.

messageText

    ^ messageText ifNil: [
        'The code being simulated is trying to control a process ({1}). {2}' translated format: {
            self sender method reference.
            self isSimulationGuard
                ifTrue: ['If you proceed, your image may become unusable. Continue at own risk, and better save your image before.' translated]
                ifFalse: ['Process controlling cannot be simulated. If you proceed, side effects may occur outside the observable area of the simulator.' translated]}]

defaultAction

    ^ self suppressed ifFalse: [super defaultAction]

forPrimitive: primitiveIndex sender: senderContext

    ^ self new primitive: primitiveIndex sender: senderContext

signalForPrimitive: primitiveIndex sender: senderContext

    ^ (self forPrimitive: primitiveIndex sender: senderContext) signal

('instance creation' forPrimitive:sender:)
('signaling' signalForPrimitive:sender:)


('testing' isControlPrimitive isSimulationGuard)
('accessing' primitive sender suppress suppressed unsuppress)
('initialize-release' primitive:sender:)
('printing' messageText)
('priv handling' defaultAction)


"Postscript:
CHANGELOG*:

- Replace generic Warning in Context >> #doPrimitive:method:receiver:args: by specific warning of new class SimulationSideEffectWarning.
- Also signal SimulationSideEffectWarning if primitive 87 (primitiveResume) is hit.
- SimulationSideEffectWarning contains logic to detect the type (simulation guard/control primitive) of the side effect. It can also be suppressed or unsuppressed along the handler chain using the '*suppress*' selectors. Control primitive side effects are suppressed by default.
- Add tests for the changes above.
- In the debugger, unsuppress control primitive warnings.
- Replace definitions of primitive 19 (currently only in ControlManager) by a named alias pragma, <simulationGuard>, which is implemented on Parser.

For more information, see: http://forum.world.st/The-Trunk-Kernel-nice-1386-mcz-td5128636.html


(* Sorry, this should be in the preamble, not in the postscript, I know, but the preamble editor in the ChangeSorter is currently broken. ¯\_(?)_/¯)
"

---

Sent from the Squeak - Dev mailing list archive at Nabble.com.