Re: 64 bit images (was Re: [squeak-dev] The Trunk: Collections-dtl.268.mcz)

> Hi,
>
> I made a new version of the system-tracer (in attachment).
> It works fine with the latest pharo version.
>
> For wordSize, we decide to create a instance variable in SmalltalkImage, which is initialized at startup.
> The method initialPC is:
>
> ---
> initialPC
> "Answer the program counter for the receiver's first bytecode."
>
> ^ (self numLiterals + 1) * SmalltalkImage current wordSize + 1
> ---
>
> The fix will be integrated in Pharo soon.
> It is also in attachment.
>
> <addWordSizeInSystemDictionary.1.cs><System-Tracing-forPharo.cs>
>
> Cheers
>
> On Jan 3, 2010, at 06:46 , John M McIntosh wrote:
>
>> Ok, well I was hoping that Laval Jannik would review the changes/solution in http://bugs.squeak.org/view.php?id=7430
>> then update his original System-Tracing.2forPharo3.cs  Then test to confirm we can build a 64bit image from the current Pharo image
>>
>>
>> On 2010-01-02, at 6:46 PM, David T. Lewis wrote:
>>
>>> John, one additional note.
>>>
>>> The SystemTracer changesets currently in circulation have an obsolete
>>> implementation of SystemDictionary>>wordSize.The original 64-bit
>>> Squeak used vmParameterAt: 27 for the VM to answer its word size,
>>> but this was later changed to vmParameter at: 40. This is the reason
>>> that the original "dist3" 64-bit image does not work on current VMs.
>>>
>>> You will want to revert #wordSize back to a current version
>>> so that it uses vmParameter at: 40.
>>>
>>> This issue will go away when we implement the cached #wordSize fixes
>>> discussed separately (http://bugs.squeak.org/view.php?id=7430), but
>>> I wanted to mention it because I spotted the obsolete method in the
>>> System-Tracing2forPharo.cs change set.
>>>
>>
>> --
>> ===========================================================================
>> John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
>> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
>> ===========================================================================
>>
>>
>>
>>
>
> ---
> Jannik Laval
> ---
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> Jannik
>
> did you check the changeset in the mantis bug report?
>
> Stef
> On Jan 4, 2010, at 10:53 AM, Laval Jannik wrote:
>
>> Hi,
>>
>> I made a new version of the system-tracer (in attachment).
>> It works fine with the latest pharo version.
>>
>> For wordSize, we decide to create a instance variable in SmalltalkImage, which is initialized at startup.
>> The method initialPC is:
>>
>> ---
>> initialPC
>> "Answer the program counter for the receiver's first bytecode."
>>
>> ^ (self numLiterals + 1) * SmalltalkImage current wordSize + 1
>> ---
>>
>> The fix will be integrated in Pharo soon.
>> It is also in attachment.
>>
>> <addWordSizeInSystemDictionary.1.cs><System-Tracing-forPharo.cs>
>>
>> Cheers
>>
>> On Jan 3, 2010, at 06:46 , John M McIntosh wrote:
>>
>>> Ok, well I was hoping that Laval Jannik would review the changes/solution in http://bugs.squeak.org/view.php?id=7430
>>> then update his original System-Tracing.2forPharo3.cs  Then test to confirm we can build a 64bit image from the current Pharo image
>>>
>>>
>>> On 2010-01-02, at 6:46 PM, David T. Lewis wrote:
>>>
>>>> John, one additional note.
>>>>
>>>> The SystemTracer changesets currently in circulation have an obsolete
>>>> implementation of SystemDictionary>>wordSize.The original 64-bit
>>>> Squeak used vmParameterAt: 27 for the VM to answer its word size,
>>>> but this was later changed to vmParameter at: 40. This is the reason
>>>> that the original "dist3" 64-bit image does not work on current VMs.
>>>>
>>>> You will want to revert #wordSize back to a current version
>>>> so that it uses vmParameter at: 40.
>>>>
>>>> This issue will go away when we implement the cached #wordSize fixes
>>>> discussed separately (http://bugs.squeak.org/view.php?id=7430), but
>>>> I wanted to mention it because I spotted the obsolete method in the
>>>> System-Tracing2forPharo.cs change set.
>>>>
>>>
>>> --
>>> ===========================================================================
>>> John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
>>> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
>>> ===========================================================================
>>>
>>>
>>>
>>>
>>
>> ---
>> Jannik Laval
>> ---
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> Yes, but the changeset build the fix in SystemDictionary,
> In a previous discussion, we decide to create an instance variable in SmalltalkImage and to put methods in SmalltalkImage too.
>
> Cheers,
> Jannik
>
> On Jan 4, 2010, at 13:35 , Stéphane Ducasse wrote:
>
>> Jannik
>>
>> did you check the changeset in the mantis bug report?
>>
>> Stef
>> On Jan 4, 2010, at 10:53 AM, Laval Jannik wrote:
>>
>>> Hi,
>>>
>>> I made a new version of the system-tracer (in attachment).
>>> It works fine with the latest pharo version.
>>>
>>> For wordSize, we decide to create a instance variable in SmalltalkImage, which is initialized at startup.
>>> The method initialPC is:
>>>
>>> ---
>>> initialPC
>>> "Answer the program counter for the receiver's first bytecode."
>>>
>>> ^ (self numLiterals + 1) * SmalltalkImage current wordSize + 1
>>> ---
>>>
>>> The fix will be integrated in Pharo soon.
>>> It is also in attachment.
>>>
>>> <addWordSizeInSystemDictionary.1.cs><System-Tracing-forPharo.cs>
>>>
>>> Cheers
>>>
>>> On Jan 3, 2010, at 06:46 , John M McIntosh wrote:
>>>
>>>> Ok, well I was hoping that Laval Jannik would review the changes/solution in http://bugs.squeak.org/view.php?id=7430
>>>> then update his original System-Tracing.2forPharo3.cs  Then test to confirm we can build a 64bit image from the current Pharo image
>>>>
>>>>
>>>> On 2010-01-02, at 6:46 PM, David T. Lewis wrote:
>>>>
>>>>> John, one additional note.
>>>>>
>>>>> The SystemTracer changesets currently in circulation have an obsolete
>>>>> implementation of SystemDictionary>>wordSize.The original 64-bit
>>>>> Squeak used vmParameterAt: 27 for the VM to answer its word size,
>>>>> but this was later changed to vmParameter at: 40. This is the reason
>>>>> that the original "dist3" 64-bit image does not work on current VMs.
>>>>>
>>>>> You will want to revert #wordSize back to a current version
>>>>> so that it uses vmParameter at: 40.
>>>>>
>>>>> This issue will go away when we implement the cached #wordSize fixes
>>>>> discussed separately (http://bugs.squeak.org/view.php?id=7430), but
>>>>> I wanted to mention it because I spotted the obsolete method in the
>>>>> System-Tracing2forPharo.cs change set.
>>>>>
>>>>
>>>> --
>>>> ===========================================================================
>>>> John M. McIntosh <[hidden email]>   Twitter:  squeaker68882
>>>> Corporate Smalltalk Consulting Ltd.  http://www.smalltalkconsulting.com
>>>> ===========================================================================
>>>>
>>>>
>>>>
>>>>
>>>
>>> ---
>>> Jannik Laval
>>> ---
>>>
>>> _______________________________________________
>>> Pharo-project mailing list
>>> [hidden email]
>>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> On Mon, Jan 04, 2010 at 05:15:28PM +0100, Andreas Raab wrote:
>> Hi Jannik -
>>
>> A couple of comments. First, If wordSize is a constant, put it into a
>> class var. That's the best practice pattern for dealing with constants
>> and I see no reason why one would opt to use an ivar for a value that
>> never changes. See for example the EndianCache in SmalltalkImage and
>> other use of constants throughout the system.
>>
>> As for caching, if I understand your description correctly, then system
>> tracer is storing the correct value for the traced image. When the image
>> starts, your cache code invalidates the known to be correct value which
>> is later lazily filled in with the same value again. So what observable
>> effect does your cache invalidation have?
>>
>> #startUp isn't the first method sent, not by a very long shot - in
>> particular when you mess with the execution machinery you have to be
>> aware that a method like #initialPC might be called before you ever get
>> to the point where you invalidate that cache (I'm not sure if this can
>> happen in this concrete case). In any case you should trace through the
>> startup sequence to find out just how much other code is executed before
>> getting there.
>
> John and/or Bert previously pointed out that there is no need to ever
> set the cached value from the image, so this would be done from a system
> tracer only (and yes this should have a comment).
>
>> My recommendations would be to make WordSize a class variable, leave the
>> lazy initialization in (it might be helpful during install etc) but add
>> a nice comment explaining that only system tracer ever modifies that
>> value when a 32/64 bit image is written. And leave out the pointless
>> cache invalidation :-)
>
> I think that the change set on Mantis 7430 does what you describe. This
> puts the class variable in SystemDictionary in order to retain the current
> "Smalltalk wordSize" idiom, and uses the original Dan Ingalls #initialPC
> implementation from the "dist3" 64-bit image.
>
> It's a small change, so I'll copy it here:
>
> 'From Squeak3.10.2 of 16 December 2009 [latest update: #8496] on 18 December 2009 at 6:08:11 pm'!
> "Change Set: Smalltalk-wordSize-dtl-M7430
> Date: 18 December 2009
> Author: David T. Lewis
>
> Cache Smalltalk wordSize in class var in SystemDictionary..
>
> Update CompiledMethod>>initialPC to use #wordSize. This is the method as implemented in the original 64-bit image (author di)."!
>
> IdentityDictionary subclass: #SystemDictionary
> instanceVariableNames: 'cachedClassNames'
> classVariableNames: 'LastImageName LastQuitLogPosition LowSpaceProcess LowSpaceSemaphore MemoryHogs ShutDownList SpecialSelectors StartUpList StartupStamp SystemChanges WordSize'
> poolDictionaries: ''
> category: 'System-Support'!
>
> !CompiledMethod methodsFor: 'accessing' stamp: 'di 6/29/2004 12:28'!
> initialPC
> "Answer the program counter for the receiver's first bytecode."
>
> ^ (self numLiterals + 1) * Smalltalk wordSize + 1
> ! !
>
>
> !SystemDictionary methodsFor: 'sources, change log' stamp: 'dtl 12/18/2009 00:32'!
> wordSize
> "Answer the size (in bytes) of an object pointer."
> "Smalltalk wordSize"
>
> ^ WordSize ifNil: [WordSize := [SmalltalkImage current vmParameterAt: 40] on: Error do: [4]]! !
>
>
>
>

> On Mon, Jan 04, 2010 at 07:09:01PM +0100, Laval Jannik wrote:
>> Hi,
>>
>> I understand your comments,
>> So:
>> - i will put word size a class var,
>> - this class var will be modify only by systemTracer
>>
>> Now I have a question:
>> Why using Smalltalk wordSize whereas the vm parameter is getting by SmalltalkImage current ?
>
> Hi Jannik,
>
> Different people may answer this question in different ways, because it
> is a matter of style and personal opinion. But it may help to know that
> earlier versions of Squeak had the vm parameter query in SystemDictionary
> ("Smalltalk") rather than in SmalltalkImage. This was moved as part of
> a larger effort to reorganize SystemDictionary, which is a rather large
> class that has accumulated many functions over the years.
>
> As matter of function, it makes no difference either way. The part that
> is subject to opinion is whether it makes more sense for a person to
> say "Smalltalk wordSize" to refer to the size of a word in the object
> memory, or whether "SmalltalkImage current wordSize" is more meaningful.
>
> FWIW, my own opinion is that neither one is right, but "Smalltalk wordSize"
> is easier to remember. But no matter what approach is taken, you will be able
> to find someone who does not agree, so you should form your own opinion ;-)
>
> HTH,
>
> Dave
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> On Mon, Jan 04, 2010 at 07:09:01PM +0100, Laval Jannik wrote:
>> Hi,
>>
>> I understand your comments,
>> So:
>> - i will put word size a class var,
>> - this class var will be modify only by systemTracer
>>
>> Now I have a question:
>> Why using Smalltalk wordSize whereas the vm parameter is getting by SmalltalkImage current ?
>
> Hi Jannik,
>
> Different people may answer this question in different ways, because it
> is a matter of style and personal opinion. But it may help to know that
> earlier versions of Squeak had the vm parameter query in SystemDictionary
> ("Smalltalk") rather than in SmalltalkImage. This was moved as part of
> a larger effort to reorganize SystemDictionary, which is a rather large
> class that has accumulated many functions over the years.
>
> As matter of function, it makes no difference either way. The part that
> is subject to opinion is whether it makes more sense for a person to
> say "Smalltalk wordSize" to refer to the size of a word in the object
> memory, or whether "SmalltalkImage current wordSize" is more meaningful.
>
> FWIW, my own opinion is that neither one is right, but "Smalltalk wordSize"
> is easier to remember. But no matter what approach is taken, you will be able
> to find someone who does not agree, so you should form your own opinion ;-)
>
> HTH,
>
> Dave
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> but we will need another name.
> Now the point behind the current (which I do not like but live with it is that)
> it would be cool to be able to do from my image some query into your image).
> This was about thinking about that that SmalltalkImage current got birth.
>
> Now I would love to get some discussions about a way to
> - have a class management class = SystemDictionary but with clean contents and cool name => namespace?
> (which often people confused with having namespace at the language level and it not part of the proposal
> don't open the box please)
> - have a cool abstraction over image housekeeping
> - a cool abstraction about vm information
>
> so if you have ideas please let me know.
>
> Stef
>
>
> On Jan 4, 2010, at 9:16 PM, David T. Lewis wrote:
>
>> On Mon, Jan 04, 2010 at 07:09:01PM +0100, Laval Jannik wrote:
>>> Hi,
>>>
>>> I understand your comments,
>>> So:
>>> - i will put word size a class var,
>>> - this class var will be modify only by systemTracer
>>>
>>> Now I have a question:
>>> Why using Smalltalk wordSize whereas the vm parameter is getting by SmalltalkImage current ?
>>
>> Hi Jannik,
>>
>> Different people may answer this question in different ways, because it
>> is a matter of style and personal opinion. But it may help to know that
>> earlier versions of Squeak had the vm parameter query in SystemDictionary
>> ("Smalltalk") rather than in SmalltalkImage. This was moved as part of
>> a larger effort to reorganize SystemDictionary, which is a rather large
>> class that has accumulated many functions over the years.
>>
>> As matter of function, it makes no difference either way. The part that
>> is subject to opinion is whether it makes more sense for a person to
>> say "Smalltalk wordSize" to refer to the size of a word in the object
>> memory, or whether "SmalltalkImage current wordSize" is more meaningful.
>>
>> FWIW, my own opinion is that neither one is right, but "Smalltalk wordSize"
>> is easier to remember. But no matter what approach is taken, you will be able
>> to find someone who does not agree, so you should form your own opinion ;-)
>>
>> HTH,
>>
>> Dave
>>
>>
>> _______________________________________________
>> Pharo-project mailing list
>> [hidden email]
>> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project
>
>
> _______________________________________________
> Pharo-project mailing list
> [hidden email]
> http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-project

> david
>
> I would really like to continue the cleaning of SystemDictionary
> because I do not like the current state of affair.
>
> Now we can either have a namespace class (with only namespace behavior)
> and a imageManagement class
>
> Now we could have one variable Smalltalk to point to either of it.
> but we will need another name.
> Now the point behind the current (which I do not like but live with it is that)
> it would be cool to be able to do from my image some query into your image).
> This was about thinking about that that SmalltalkImage current got birth.
>
> Now I would love to get some discussions about a way to
> - have a class management class = SystemDictionary but with clean contents and cool name => namespace?
> (which often people confused with having namespace at the language level and it not part of the proposal
> don't open the box please)
> - have a cool abstraction over image housekeeping
> - a cool abstraction about vm information
>
> so if you have ideas please let me know.

> On Mon, Jan 04, 2010 at 09:30:45PM +0100, St?phane Ducasse wrote:
>> david
>>
>> I would really like to continue the cleaning of SystemDictionary
>> because I do not like the current state of affair.
>>
>> Now we can either have a namespace class (with only namespace behavior)
>> and a imageManagement class
>>
>> Now we could have one variable Smalltalk to point to either of it.
>> but we will need another name.
>> Now the point behind the current (which I do not like but live with it is that)
>> it would be cool to be able to do from my image some query into your image).
>> This was about thinking about that that SmalltalkImage current got birth.
>>
>> Now I would love to get some discussions about a way to
>> - have a class management class = SystemDictionary but with clean contents and cool name => namespace?
>> (which often people confused with having namespace at the language level and it not part of the proposal
>> don't open the box please)
>> - have a cool abstraction over image housekeeping
>> - a cool abstraction about vm information
>>
>> so if you have ideas please let me know.
>
> Stef,
>
> I think it would be good to step away from the details and just
> think about the high level objects - what do they represent and what
> are they responsible for (http://www.wirfs-brock.com/Design.html)?