we often use symbols for identity keys. Now I was just dreaming.

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Hmm.. i don't see obvious advantages. Do you? Most symbols (90% of cases), used as selectors, so it is likely that such refactoring will be nothing more, but just renaming the class :)

...

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

that a fun one. BTW I know that david simmons introduced == and === on symbol and === was checking the namespace in which the symbol was defined and this way he implemented selector namespace. On Aug 24, 2010, at 4:00 PM, Igor Stasenko wrote:

On 24 August 2010 16:16, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

we often use symbols for identity keys. Now I was just dreaming.

I had a long discussion with my friend, who was learning smalltalk, where we argued about 'selector is not a symbol'.

The difference between a bare string and symbol is uniqueness. Literally, a symbol is just a string, which having additional, implied semantics, that same symbolic strings refer to a single, unique object. (Another minor one is immutability).

The difference between a symbol and selector is less significant. A simple example. Suppose you have two different classes, one Symbol, other is Selector. Now, suppose that compiler creating an instance of Symbol each time it sees following:

self foo.

where 'foo' is selector (and instance of Selector).

but now, lets put it in another way:

self perform: #foo.

A #foo is, obviously a symbol, and must be an instance of Symbol class, isnt?

So, in #perform: you must somehow convert a symbol into selector and then find the implementation of method with given selector. This obviously kills an idea of having distinction between selectors and symbols, because implementation of #perform: will be awfully slow and awkward.

...

...

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Hmm.. i don't see obvious advantages. Do you? Most symbols (90% of cases), used as selectors, so it is likely that such refactoring will be nothing more, but just renaming the class :)

...

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

:) Your previous mail was what the bee I had in my bonnet :) Stef On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

But then eliminating those methods introduces a circularity problem: What if a symbol starts to be used as a selector part of the way through the life of that object? It's not a fixed property of a symbol to be, or not be, a selector. On Wed, Aug 25, 2010 at 7:11 PM, Hernan Wilkinson < hernan.wilkinson@10pines.com> wrote:

From the design point of view, I like the idea, a selector is not a symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea. This idea brings implementation challenges of course, but nothing new :-)

On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse < stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Marcin Tustin Mobile: 07773787105 Office: 020 3400 3259

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

From the design point of view, I like the idea, a selector is not a symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code. I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability. String + uniqueness + immutability => Symbol so, there is a strong reason for having a separate class for it, but Symbol + ??? => Selector give me a strong reason. i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places. 'foo' = #foo ->true #foo = 'foo' -> true #foo asSelector = #foo ?? if so, then #foo hash = #foo asSelector hash ? and , obviously #foo == #foo asSelector => false which means, that mixing selectors and symbols in a single collection will be very bad idea, and source of bugs and confusion. But what gives us a uniqueness of symbols is, that we don't need to use #= for comparing them, but #== , which is much faster. Right? Now, if you introduce selectors, there is no uniqueness anymore, since selector is a symbol (since it is a subclass of it), and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule. Objects could play many roles, depending on their use. Just take a look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

This idea brings implementation challenges of course, but nothing new :-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

On Wed, Aug 25, 2010 at 4:33 PM, Igor Stasenko <siguctua@gmail.com> wrote:

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

From the design point of view, I like the idea, a selector is not a symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code.

I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability.

String + uniqueness + immutability => Symbol

so, there is a strong reason for having a separate class for it, but

Symbol + ??? => Selector

give me a strong reason.

meaning, semantic... that I thing are stronger reasons that uniqueness for example, because uniqueness is an implementation issue, meaning is not. Immutability is not an implementation issue is part of the semantic of the object.

i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

and , obviously #foo == #foo asSelector => false

yes, and #foo = #foo asSelector --> false

which means, that mixing selectors and symbols in a single collection will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo' where asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc. But what gives us a uniqueness of symbols is, that we don't need to

use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really want two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

Now, if you introduce selectors, there is no uniqueness anymore,

why?

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to reuse code it is a bad design decision from my point of view, Selector should be a class by itself whose instances know a name, that are instances of Symbol and must honor

the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

Objects could play many roles, depending on their use. Just take a look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

Not a bad idea!! Object is completely out of control and we should remove a lot of the methods it has! I completely agree there :-)

...

This idea brings implementation challenges of course, but nothing new :-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

On Wed, Aug 25, 2010 at 7:25 PM, Igor Stasenko <siguctua@gmail.com> wrote:

On 25 August 2010 23:08, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

On Wed, Aug 25, 2010 at 4:33 PM, Igor Stasenko <siguctua@gmail.com>

wrote:

...

...

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com

...

wrote:

...

From the design point of view, I like the idea, a selector is not a symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code.

I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability.

String + uniqueness + immutability => Symbol

so, there is a strong reason for having a separate class for it, but

Symbol + ??? => Selector

give me a strong reason.

meaning, semantic... that I thing are stronger reasons that uniqueness for example, because uniqueness is an implementation issue, meaning is not. Immutability is not an implementation issue is part of the semantic of the object.

Meaning could be anything. It depends from context where i using this object.

hmmm, for me the meaning of an object is given by the messages it knows how to respond, no context related. I'm talking about objects defined in a formal language, not of entities described in natural language where context influences meaning...

From machine's point of view there is no meaning, only semantic.

hehe, I thought that machines only treated symbols, therefore they are incapable to provide semantic to them... only we human can provide meaning (that for me is the same as semantic in this context :-) ), so we have basic differences in fundamental concepts... I don't think we will agree therefore on high level ones...

So, i am asking, what _technical_ benefits it would brings us.

I don't care about technical benefits, I care about modeling benefits. I leave the technical issues to the computer that does not get tired when doing things over and over again... :-). Asking only for technical benefits of a solution is like asking why do we have garbage collection? what is the technical benefit of it? it takes more time that manually deleting objects, it needs more memory, etc. so what is its technical benefit? none... the benefit is for us, because it allow us to delegate the memory management responsibility to the computer (a repetitive process) and free our mind to concentrate on important issues Because from meaning point of view, a symbol #foo reads foo,

as well as selector foo, and only we (humans) know that it is selector in one context, or just a simple 'unique string' in another one.

...

...

i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

...

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

...

and , obviously #foo == #foo asSelector => false

yes, and #foo = #foo asSelector --> false

...

which means, that mixing selectors and symbols in a single collection will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo' where asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc.

it could have a sense to use different entity, when you would need to hold additional state. I can imagine a class, like 'measure unit', with two variables - value and units. So, you can express 1 cm, 1 mm, 1 km, 100 kg using such object. But in case of selectors , you expressing a symbol, and there is no need for additional state.

Different objects are not only created because they have different "state" but because they have different responsibilities, for me that is the central issue, and as I said before I see that we do not agree on fundamental issues so...

...

...

But what gives us a uniqueness of symbols is, that we don't need to use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really want two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

...

Now, if you introduce selectors, there is no uniqueness anymore,

why?

well, because two things which having same name should be equal. Otherwise it makes no sense to use same name for them. Keep in mind, that symbols are made for humans. Machine feels pretty well with pointers and don't needs to know anything about names.

but programs are made by humans, read by humans and maintain by humans, so that is why we have symbol instead of pointers and true instead of 1 and false instead of 0... for us, not for the machine. I don't understand why you say that there will be no more uniqueness, a selector can be unique per its name or not, that is an implementation issue, that is, how much memory you want to consume... the important thing is that selectors with the same name should be equal, who cares if they share the same memory position? only the implementor of the vm... I don't care if "Date today == Date today", I do care that "Date today = Date today"... if tomorrow the Date implementor decides to do "Date today == Date today", good, that is an implementation issue, could be good could be bad, and I as a user of Date should not care.

...

...

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to reuse code it is a bad design decision from my point of view, Selector should be a class by itself whose instances know a name, that are instances of Symbol

...

and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

So, you want a completely separate class for selectors, which by sending #name (for instance) will answer a symbol which identifies them?

yes

I see, use delegation instead of inheritance.

yes

But then, technically, this means that for each symbol, which used as selector, you will need to add an additional instance of Selector, which wraps around symbol.

no, it does not wraps around... it has a completely different set of messages that it responds to, it does not wrap it, because if in that situation we say that selector wraps a symbol, then a date wraps a number (its day number), and Character wraps a number (its value), etc. Symbols and selectors are not completely polymorphic under the model I propose, the same way Character is not completely polymorphic with Number, etc. If I understand your point, you should not have Character because it only holds a Number ? is that right?

Sounds not very good. But maybe there are benefits, which outweights such memory increase? Tell me more.

well I think I made my point very clear, you care about memory usage I care about object responsibilities, you care about implementation I care about design that encapsulates implementation... so I don't think we will agree until either one changes his vision on what it is more important... and I will not hehe (just kidding)

It reminds me a MethodReference. It also identifies a CompiledMethod, without referring directly to class or method. But if you look closer at its implementation, you'll see that it is used as a dumb state holder, grouping different aspects of a method's state. It is used extensively by browser tools, even in places, where you could use and talk with compiled methods directly. Needless to say, that i can't understand such approach: why one would need to wrap things with a proxy, only to access the state of an object of interest though it? Why an object of interest can't give same answers directly?

I would understand that tools could be made so, that they can work with real objects as well as with proxies, which both honoring same protocol. But blindly wrapping each object with proxy?? This is beyond my understanding abilities.

I can't comment based on your description, I don't know MethodReference, what represents and what it is used for... sorry, cant help here.

...

...

Objects could play many roles, depending on their use. Just take a look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

Not a bad idea!! Object is completely out of control and we should remove a lot of the methods it has! I completely agree there :-)

Hehe. Good luck with that :)

thanks! but I'm not going to do it... I would start from zero than trying to fix something that is already broken :-)

Sure thing, i'd like to see less methods in Object class.

But "Object class" has no methods! it is Object that has a lot of methods! (just kidding, I know you mean the class Object and no "Object class") Bye! Hernan.

...

...

...

This idea brings implementation challenges of course, but nothing new :-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

...

...

...

...

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

as I said in my first mail, I think it is a good idea from the design point of view, should it be implemented now? I dont think so, as you say, there are other more important things to do in pharo, but the idea is good. Bye! Hernan. On Thu, Aug 26, 2010 at 12:27 AM, Igor Stasenko <siguctua@gmail.com> wrote:

On 26 August 2010 04:09, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

On Wed, Aug 25, 2010 at 7:25 PM, Igor Stasenko <siguctua@gmail.com>

wrote:

...

...

On 25 August 2010 23:08, Hernan Wilkinson <hernan.wilkinson@10pines.com

...

wrote:

...

On Wed, Aug 25, 2010 at 4:33 PM, Igor Stasenko <siguctua@gmail.com> wrote:

...

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

From the design point of view, I like the idea, a selector is not a symbol and we can see that clearly in the Symbol protocol, like Mariano

and

...

...

...

Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code.

I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability.

String + uniqueness + immutability => Symbol

so, there is a strong reason for having a separate class for it, but

Symbol + ??? => Selector

give me a strong reason.

meaning, semantic... that I thing are stronger reasons that uniqueness for example, because uniqueness is an implementation issue, meaning is not. Immutability is not an implementation issue is part of the semantic of the object.

Meaning could be anything. It depends from context where i using this object.

hmmm, for me the meaning of an object is given by the messages it knows how to respond, no context related. I'm talking about objects defined in a formal language, not of entities described in natural language where context influences meaning...

I don't know what you mean by 'meaning of an object in a formal language'. For me, an object's meaning (or in other words, its role) always derives from context, where it is used, not from its class or implementation. Classes are not 'specification of roles'. They are orthogonal to them and just providing an object's behavior. Classes do not restricting me from using objects for different roles, even not intended ones originally. That's why i see nothing wrong when symbols in one context plays role as selector, in other as dictionary keys, or class names , a collection element etc etc. It is pointless to try and predict all possible roles, which object may play in different contexts, and try to implement all of them in separate class(es).

...

...

From machine's point of view there is no meaning, only semantic.

hehe, I thought that machines only treated symbols, therefore they are incapable to provide semantic to them... only we human can provide meaning (that for me is the same as semantic in this context :-) ), so we have basic differences in fundamental concepts... I don't think we will agree therefore on high level ones...

...

So, i am asking, what _technical_ benefits it would brings us.

I don't care about technical benefits, I care about modeling benefits. I leave the technical issues to the computer that does not get tired when doing things over and over again... :-).

you know, why i don't like rebooting my PC? i am tired waiting, for system booting, since over last 20 years, computers became 10 times faster (or even more), while at the same time, their boot time increasing and increasing over and over. So, please, don't say to me, that technical side is not worth to consider.

...

Asking only for technical benefits of a solution is like asking why do we have garbage collection? what is the technical benefit of it? it takes more time that manually deleting objects, it needs more memory, etc. so what is its technical benefit? none... the benefit is for us, because it allow us to delegate the memory management responsibility to the computer (a repetitive process) and free our mind to concentrate on important issues

Well, i went to technical benefits only because i don't see any strong modeling ones, which worth time spent on: - refactoring class/metaclass model - changing compiler - changing all tools to use selectors instead of symbols - etc etc etc

really, think about how much must be done to introduce it over existing system, and only to dissect a minor subset of behavior from Symbol class.. Come on. Do we really have nothing else to do, more important to change/improve?

I agree, such change may help developers in better understanding the system. But for most developers, who busy doing own projects, not related with depths of a system organization and compiler/browser framework, i really don't see big quality shift comparing to current scheme. The fact that currently, Symbols playing role as selectors doesn't prevents me from writing good code, and i doubt that even if you change that, it will help me in anything except mentioned above.

Good model is a balance between simplicity, clarity and technical sides. While adding Selector improves clarity, at same time it increasing complexity, and worsening a technical side, since it will lead to increased memory footprint.

...

...

Because from meaning point of view, a symbol #foo reads foo, as well as selector foo, and only we (humans) know that it is selector in one context, or just a simple 'unique string' in another one.

...

...

i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

...

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

...

and , obviously #foo == #foo asSelector => false

yes, and #foo = #foo asSelector --> false

...

which means, that mixing selectors and symbols in a single collection will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo'

where

...

...

asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc.

it could have a sense to use different entity, when you would need to hold additional state. I can imagine a class, like 'measure unit', with two variables - value and units. So, you can express 1 cm, 1 mm, 1 km, 100 kg using such object. But in case of selectors , you expressing a symbol, and there is no need for additional state.

Different objects are not only created because they have different "state" but because they have different responsibilities, for me that is the central issue, and as I said before I see that we do not agree on fundamental issues so...

i don't feel that selectors are so different from symbols, that its worth making a separate class.

...

...

...

...

But what gives us a uniqueness of symbols is, that we don't need to use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really

want

...

...

two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

...

Now, if you introduce selectors, there is no uniqueness anymore,

why?

well, because two things which having same name should be equal. Otherwise it makes no sense to use same name for them. Keep in mind, that symbols are made for humans. Machine feels pretty well with pointers and don't needs to know anything about names.

but programs are made by humans, read by humans and maintain by humans, so that is why we have symbol instead of pointers and true instead of 1 and false instead of 0... for us, not for the machine. I don't understand why you say that there will be no more uniqueness, a selector can be unique per its name or not, that is an implementation issue, that is, how much memory you want to consume... the important thing is that selectors with the same name should be equal, who cares if they share the same memory position? only the implementor of the vm... I don't care if "Date today == Date today", I do care that "Date today = Date today"... if tomorrow the Date implementor decides to do "Date today == Date today", good, that is an implementation issue, could be good could be bad, and I as a user of Date should not care.

...

...

...

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to reuse code it is a bad design decision from my point of view, Selector should be

a

...

...

class by itself whose instances know a name, that are instances of Symbol

...

and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

So, you want a completely separate class for selectors, which by sending #name (for instance) will answer a symbol which identifies them?

yes

...

I see, use delegation instead of inheritance.

yes

...

But then, technically, this means that for each symbol, which used as selector, you will need to add an additional instance of Selector, which wraps around symbol.

no, it does not wraps around... it has a completely different set of messages that it responds to, it does not wrap it, because if in that situation we say that selector wraps a symbol, then a date wraps a number (its day number), and Character wraps a number (its value), etc. Symbols and selectors are not completely polymorphic under the model I propose, the same way Character is not completely polymorphic with Number, etc. If I understand your point, you should not have Character because it only holds a Number ? is that right?

Character and Number is quite different by nature. As to me, for Symbol/Selector a better analogy would be Fruit and Apple.

...

...

Sounds not very good. But maybe there are benefits, which outweights such memory increase? Tell me more.

well I think I made my point very clear, you care about memory usage I care about object responsibilities, you care about implementation I care about design that encapsulates implementation... so I don't think we will agree until either one changes his vision on what it is more important... and I will not hehe (just kidding)

I care about all sides of it. And i'm not ready to sacrifice a memory usage for that. There's a lot more better places where i could put a free memory in use :)

...

...

It reminds me a MethodReference. It also identifies a CompiledMethod, without referring directly to class or method. But if you look closer at its implementation, you'll see that it is used as a dumb state holder, grouping different aspects of a method's state. It is used extensively by browser tools, even in places, where you could use and talk with compiled methods directly. Needless to say, that i can't understand such approach: why one would need to wrap things with a proxy, only to access the state of an object of interest though it? Why an object of interest can't give same answers directly?

I would understand that tools could be made so, that they can work with real objects as well as with proxies, which both honoring same protocol. But blindly wrapping each object with proxy?? This is beyond my understanding abilities.

I can't comment based on your description, I don't know MethodReference, what represents and what it is used for... sorry, cant help here.

...

...

...

Objects could play many roles, depending on their use. Just take a look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

Not a bad idea!! Object is completely out of control and we should remove a lot of the methods it has! I completely agree there :-)

Hehe. Good luck with that :)

thanks! but I'm not going to do it... I would start from zero than trying to fix something that is already broken :-)

...

Sure thing, i'd like to see less methods in Object class.

But "Object class" has no methods! it is Object that has a lot of methods! (just kidding, I know you mean the class Object and no "Object class") Bye! Hernan.

...

...

...

...

This idea brings implementation challenges of course, but nothing

new

...

...

...

...

:-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

...

...

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

an optimisation? for me this change looks for exactly what you plea, simplicity and uniformity... at the end this change makes symbol and selector more cohesive, and we all know that the more cohesive an object is the more simply and uniform it is, because it represents just one thing, no like symbols now that they can be symbols or selectors (therefore they are not as cohesive as having two different abstractions) and we, human, have to decide what they are instead of letting the objects decide... On Thu, Aug 26, 2010 at 8:43 AM, Marcin Tustin <mm3@zepler.net> wrote:

Can I make a last plea for the simplicity and uniformity of the current model? I suspect that this change is a premature optimisation that will lose us flexibility, for no apparent benefit.

On Thu, Aug 26, 2010 at 12:28 PM, Hernan Wilkinson < hernan.wilkinson@10pines.com> wrote:

...

as I said in my first mail, I think it is a good idea from the design point of view, should it be implemented now? I dont think so, as you say, there are other more important things to do in pharo, but the idea is good.

Bye! Hernan.

On Thu, Aug 26, 2010 at 12:27 AM, Igor Stasenko <siguctua@gmail.com>wrote:

...

On 26 August 2010 04:09, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

On Wed, Aug 25, 2010 at 7:25 PM, Igor Stasenko <siguctua@gmail.com>

wrote:

...

...

On 25 August 2010 23:08, Hernan Wilkinson <

hernan.wilkinson@10pines.com>

...

wrote:

...

On Wed, Aug 25, 2010 at 4:33 PM, Igor Stasenko <siguctua@gmail.com> wrote:

...

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

From the design point of view, I like the idea, a selector is not

a

...

...

...

symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code.

I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability.

String + uniqueness + immutability => Symbol

so, there is a strong reason for having a separate class for it, but

Symbol + ??? => Selector

give me a strong reason.

meaning, semantic... that I thing are stronger reasons that uniqueness for example, because uniqueness is an implementation issue, meaning is not. Immutability is not an implementation issue is part of the semantic of the object.

Meaning could be anything. It depends from context where i using this object.

hmmm, for me the meaning of an object is given by the messages it knows how to respond, no context related. I'm talking about objects defined in a formal language, not of entities described in natural language where context influences meaning...

I don't know what you mean by 'meaning of an object in a formal language'. For me, an object's meaning (or in other words, its role) always derives from context, where it is used, not from its class or implementation. Classes are not 'specification of roles'. They are orthogonal to them and just providing an object's behavior. Classes do not restricting me from using objects for different roles, even not intended ones originally. That's why i see nothing wrong when symbols in one context plays role as selector, in other as dictionary keys, or class names , a collection element etc etc. It is pointless to try and predict all possible roles, which object may play in different contexts, and try to implement all of them in separate class(es).

...

...

From machine's point of view there is no meaning, only semantic.

hehe, I thought that machines only treated symbols, therefore they are incapable to provide semantic to them... only we human can provide meaning (that for me is the same as semantic in this context :-) ), so we have basic differences in fundamental concepts... I don't think we will agree therefore on high level ones...

...

So, i am asking, what _technical_ benefits it would brings us.

I don't care about technical benefits, I care about modeling benefits. I leave the technical issues to the computer that does not get tired when doing things over and over again... :-).

you know, why i don't like rebooting my PC? i am tired waiting, for system booting, since over last 20 years, computers became 10 times faster (or even more), while at the same time, their boot time increasing and increasing over and over. So, please, don't say to me, that technical side is not worth to consider.

...

Asking only for technical benefits of a solution is like asking why do we have garbage collection? what is the technical benefit of it? it takes more time that manually deleting objects, it needs more memory, etc. so what is its technical benefit? none... the benefit is for us, because it allow us to delegate the memory management responsibility to the computer (a repetitive process) and free our mind to concentrate on important issues

Well, i went to technical benefits only because i don't see any strong modeling ones, which worth time spent on: - refactoring class/metaclass model - changing compiler - changing all tools to use selectors instead of symbols - etc etc etc

really, think about how much must be done to introduce it over existing system, and only to dissect a minor subset of behavior from Symbol class.. Come on. Do we really have nothing else to do, more important to change/improve?

I agree, such change may help developers in better understanding the system. But for most developers, who busy doing own projects, not related with depths of a system organization and compiler/browser framework, i really don't see big quality shift comparing to current scheme. The fact that currently, Symbols playing role as selectors doesn't prevents me from writing good code, and i doubt that even if you change that, it will help me in anything except mentioned above.

Good model is a balance between simplicity, clarity and technical sides. While adding Selector improves clarity, at same time it increasing complexity, and worsening a technical side, since it will lead to increased memory footprint.

...

...

Because from meaning point of view, a symbol #foo reads foo, as well as selector foo, and only we (humans) know that it is selector in one context, or just a simple 'unique string' in another one.

...

...

i like the idea of shrinking a Symbol's protocol and putting it

into

...

...

...

lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

...

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

...

and , obviously #foo == #foo asSelector => false

yes, and #foo = #foo asSelector --> false

...

which means, that mixing selectors and symbols in a single

collection

...

will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo' where asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc.

it could have a sense to use different entity, when you would need to hold additional state. I can imagine a class, like 'measure unit', with two variables - value and units. So, you can express 1 cm, 1 mm, 1 km, 100 kg using such object. But in case of selectors , you expressing a symbol, and there is no need for additional state.

Different objects are not only created because they have different "state" but because they have different responsibilities, for me that is the central issue, and as I said before I see that we do not agree on fundamental issues so...

i don't feel that selectors are so different from symbols, that its worth making a separate class.

...

...

...

...

But what gives us a uniqueness of symbols is, that we don't need to use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really

want

...

...

two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

...

Now, if you introduce selectors, there is no uniqueness anymore,

why?

well, because two things which having same name should be equal. Otherwise it makes no sense to use same name for them. Keep in mind, that symbols are made for humans. Machine feels pretty well with pointers and don't needs to know anything about names.

but programs are made by humans, read by humans and maintain by humans, so that is why we have symbol instead of pointers and true instead of 1 and false instead of 0... for us, not for the machine. I don't understand why you say that there will be no more uniqueness, a selector can be unique per its name or not, that is an implementation issue, that is, how much memory you want to consume... the important thing is that selectors with the same name should be equal, who cares if they share the same memory position? only the implementor of the vm... I don't care if "Date today == Date today", I do care that "Date today = Date today"... if tomorrow the Date implementor decides to do "Date today == Date today", good, that is an implementation issue, could be good could be bad, and I as a user of Date should not care.

...

...

...

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to

reuse

...

...

code it is a bad design decision from my point of view, Selector should be a class by itself whose instances know a name, that are instances of Symbol

...

and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

So, you want a completely separate class for selectors, which by sending #name (for instance) will answer a symbol which identifies them?

yes

...

I see, use delegation instead of inheritance.

yes

...

But then, technically, this means that for each symbol, which used as selector, you will need to add an additional instance of Selector, which wraps around symbol.

no, it does not wraps around... it has a completely different set of messages that it responds to, it does not wrap it, because if in that situation we say that selector wraps a symbol, then a date wraps a number (its day number), and Character wraps a number (its value), etc. Symbols and selectors are not completely polymorphic under the model I propose, the same way Character is not completely polymorphic with Number, etc. If I understand your point, you should not have Character because it only holds a Number ? is that right?

Character and Number is quite different by nature. As to me, for Symbol/Selector a better analogy would be Fruit and Apple.

...

...

Sounds not very good. But maybe there are benefits, which outweights such memory increase? Tell me more.

well I think I made my point very clear, you care about memory usage I care about object responsibilities, you care about implementation I care about design that encapsulates implementation... so I don't think we will agree until either one changes his vision on what it is more important... and I will not hehe (just kidding)

I care about all sides of it. And i'm not ready to sacrifice a memory usage for that. There's a lot more better places where i could put a free memory in use :)

...

...

It reminds me a MethodReference. It also identifies a CompiledMethod, without referring directly to class or method. But if you look closer at its implementation, you'll see that it is used as a dumb state holder, grouping different aspects of a method's state. It is used extensively by browser tools, even in places, where you could use and talk with compiled methods directly. Needless to say, that i can't understand such approach: why one would need to wrap things with a proxy, only to access the state of an object of interest though it? Why an object of interest can't give same answers directly?

I would understand that tools could be made so, that they can work with real objects as well as with proxies, which both honoring same protocol. But blindly wrapping each object with proxy?? This is beyond my understanding abilities.

I can't comment based on your description, I don't know MethodReference, what represents and what it is used for... sorry, cant help here.

...

...

...

Objects could play many roles, depending on their use. Just take a look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

Not a bad idea!! Object is completely out of control and we should remove a lot of the methods it has! I completely agree there :-)

Hehe. Good luck with that :)

thanks! but I'm not going to do it... I would start from zero than trying to fix something that is already broken :-)

...

Sure thing, i'd like to see less methods in Object class.

But "Object class" has no methods! it is Object that has a lot of methods! (just kidding, I know you mean the class Object and no "Object class") Bye! Hernan.

...

...

...

...

This idea brings implementation challenges of course, but nothing

new

...

...

...

...

:-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in this list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new

class:

...

...

...

...

Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Marcin Tustin Mobile: 07773787105 Office: 020 3400 3259

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

On Thu, Aug 26, 2010 at 2:02 PM, Marcin Tustin <mm3@zepler.net> wrote:

Uniformity is when we represent our various significands using the same data type

well... I don't agree... starting from the fact that there are no data types in the object paradigm, just objects.

(in smalltalk using objects with the same protocol). For example, this allows us to treat all things in our image as objects, which gives us the flexibility to do everything in the object protocol to every kind of thing, without having to decide that we do only some things to classes, only other things to objects, and only other things again to constants (as in Java).

sorry, I don't understand.

This change will introduce heterogeneity, which prevents us using the (now two) things in the same way.

but, for me, a symbol is not a selector. What is the meaning of asking a selector for its size? for me makes more sense to ask the selector's name for its size. what is the meaning of asking a symbol for the number of arguments? if that symbol does not represent a selector there is no reason for the symbol to respond that question.

The reason to do this is if it would be a category error to do symbol operations on a selector, as it locks us into a particular view of the way the world should work, which is what the designers of java did when they posited classes, objects, and constants as being entirely separate.

I dont understand if this is good or bad for you... for me the separation that java does with object, classes and data types is a mess, it is just not uniform, I prefer the smalltalk way :-)

This has the "advantage" that you can't accidentally put a constant in an array, for example, because they wanted to restrict flexibility, in order to protect users from mistakes.

I never made that mistake in Smalltalk and Smalltalk does not prevent me for that...

I don't see a risk of our current approach causing errors, and I don't see that it creates unnecessary complexity, so I am confused as to why this change is desirable.

ok, we don't have to agree, but just answer yourself this question: the proposed change is more cohesive or not? and remember that more cohesive objects means more simplicity and uniformity... if you don't agree that cohesion implies that, well.... we should be discussing other things first :-)

On Thu, Aug 26, 2010 at 5:52 PM, Hernan Wilkinson < hernan.wilkinson@10pines.com> wrote:

...

an optimisation? for me this change looks for exactly what you plea, simplicity and uniformity... at the end this change makes symbol and selector more cohesive, and we all know that the more cohesive an object is the more simply and uniform it is, because it represents just one thing, no like symbols now that they can be symbols or selectors (therefore they are not as cohesive as having two different abstractions) and we, human, have to decide what they are instead of letting the objects decide...

On Thu, Aug 26, 2010 at 8:43 AM, Marcin Tustin <mm3@zepler.net> wrote:

...

Can I make a last plea for the simplicity and uniformity of the current model? I suspect that this change is a premature optimisation that will lose us flexibility, for no apparent benefit.

On Thu, Aug 26, 2010 at 12:28 PM, Hernan Wilkinson < hernan.wilkinson@10pines.com> wrote:

...

as I said in my first mail, I think it is a good idea from the design point of view, should it be implemented now? I dont think so, as you say, there are other more important things to do in pharo, but the idea is good.

Bye! Hernan.

On Thu, Aug 26, 2010 at 12:27 AM, Igor Stasenko <siguctua@gmail.com>wrote:

...

On 26 August 2010 04:09, Hernan Wilkinson < hernan.wilkinson@10pines.com> wrote:

...

On Wed, Aug 25, 2010 at 7:25 PM, Igor Stasenko <siguctua@gmail.com>

wrote:

...

...

On 25 August 2010 23:08, Hernan Wilkinson <

hernan.wilkinson@10pines.com>

...

wrote:

...

On Wed, Aug 25, 2010 at 4:33 PM, Igor Stasenko <

siguctua@gmail.com>

...

wrote:

...

On 25 August 2010 21:11, Hernan Wilkinson <hernan.wilkinson@10pines.com> wrote:

...

From the design point of view, I like the idea, a selector is

not a

...

...

symbol and we can see that clearly in the Symbol protocol, like Mariano and Gabriel said, implements messages that are related to selector behavior... Why symbols respond to numArgs:, precedence, isUnary, etc? those are a selectors responsibilities... not all symbols are selectors... so for me, it is a good idea.

not all symbols are selectors... but you never tell, that they could not be, once you load new package or write more code.

I see the need of having a separate subclass for Symbol , instead of stopping at String, because it have to maintain uniqueness and immutability.

String + uniqueness + immutability => Symbol

so, there is a strong reason for having a separate class for it, but

Symbol + ??? => Selector

give me a strong reason.

meaning, semantic... that I thing are stronger reasons that uniqueness for example, because uniqueness is an implementation issue, meaning is not. Immutability is not an implementation issue is part of the semantic of the object.

Meaning could be anything. It depends from context where i using this object.

hmmm, for me the meaning of an object is given by the messages it knows how to respond, no context related. I'm talking about objects defined in a formal language, not of entities described in natural language where context influences meaning...

I don't know what you mean by 'meaning of an object in a formal language'. For me, an object's meaning (or in other words, its role) always derives from context, where it is used, not from its class or implementation. Classes are not 'specification of roles'. They are orthogonal to them and just providing an object's behavior. Classes do not restricting me from using objects for different roles, even not intended ones originally. That's why i see nothing wrong when symbols in one context plays role as selector, in other as dictionary keys, or class names , a collection element etc etc. It is pointless to try and predict all possible roles, which object may play in different contexts, and try to implement all of them in separate class(es).

...

...

From machine's point of view there is no meaning, only semantic.

hehe, I thought that machines only treated symbols, therefore they are incapable to provide semantic to them... only we human can provide meaning (that for me is the same as semantic in this context :-) ), so we have basic differences in fundamental concepts... I don't think we will agree therefore on high level ones...

...

So, i am asking, what _technical_ benefits it would brings us.

I don't care about technical benefits, I care about modeling benefits. I leave the technical issues to the computer that does not get tired when doing things over and over again... :-).

you know, why i don't like rebooting my PC? i am tired waiting, for system booting, since over last 20 years, computers became 10 times faster (or even more), while at the same time, their boot time increasing and increasing over and over. So, please, don't say to me, that technical side is not worth to consider.

...

Asking only for technical benefits of a solution is like asking why do we have garbage collection? what is the technical benefit of it? it takes more time that manually deleting objects, it needs more memory, etc. so what is its technical benefit? none... the benefit is for us, because it allow us to delegate the memory management responsibility to the computer (a repetitive process) and free our mind to concentrate on important issues

Well, i went to technical benefits only because i don't see any strong modeling ones, which worth time spent on: - refactoring class/metaclass model - changing compiler - changing all tools to use selectors instead of symbols - etc etc etc

really, think about how much must be done to introduce it over existing system, and only to dissect a minor subset of behavior from Symbol class.. Come on. Do we really have nothing else to do, more important to change/improve?

I agree, such change may help developers in better understanding the system. But for most developers, who busy doing own projects, not related with depths of a system organization and compiler/browser framework, i really don't see big quality shift comparing to current scheme. The fact that currently, Symbols playing role as selectors doesn't prevents me from writing good code, and i doubt that even if you change that, it will help me in anything except mentioned above.

Good model is a balance between simplicity, clarity and technical sides. While adding Selector improves clarity, at same time it increasing complexity, and worsening a technical side, since it will lead to increased memory footprint.

...

...

Because from meaning point of view, a symbol #foo reads foo, as well as selector foo, and only we (humans) know that it is

selector

...

in one context, or just a simple 'unique string' in another one.

...

...

i like the idea of shrinking a Symbol's protocol and putting it

into

...

lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

...

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

...

and , obviously #foo == #foo asSelector => false

yes, and #foo = #foo asSelector --> false

...

which means, that mixing selectors and symbols in a single

collection

...

will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo' where asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc.

it could have a sense to use different entity, when you would need to hold additional state. I can imagine a class, like 'measure unit', with two variables - value and units. So, you can express 1 cm, 1 mm, 1 km, 100 kg using such object. But in case of selectors , you expressing a symbol, and there is no need for additional state.

Different objects are not only created because they have different "state" but because they have different responsibilities, for me that is the central issue, and as I said before I see that we do not agree on fundamental issues so...

i don't feel that selectors are so different from symbols, that its worth making a separate class.

...

...

...

...

But what gives us a uniqueness of symbols is, that we don't need

to

...

...

...

use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really want two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

...

Now, if you introduce selectors, there is no uniqueness anymore,

why?

well, because two things which having same name should be equal. Otherwise it makes no sense to use same name for them. Keep in mind, that symbols are made for humans. Machine feels pretty well with pointers and don't needs to know anything about names.

but programs are made by humans, read by humans and maintain by humans, so that is why we have symbol instead of pointers and true instead of 1 and false instead of 0... for us, not for the machine. I don't understand why you say that there will be no more uniqueness, a selector can be unique per its name or not, that is an implementation issue, that is, how much memory you want to consume... the important thing is that selectors with the same name should be equal, who cares if they share the same memory position? only the implementor of the vm... I don't care if "Date today == Date today", I do care that "Date today = Date today"... if tomorrow the Date implementor decides to do "Date today == Date today", good, that is an implementation issue, could be good could be bad, and I as a user of Date should not care.

...

...

...

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to

reuse

...

...

code it is a bad design decision from my point of view, Selector should be a class by itself whose instances know a name, that are instances of Symbol

...

and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

So, you want a completely separate class for selectors, which by sending #name (for instance) will answer a symbol which identifies them?

yes

...

I see, use delegation instead of inheritance.

yes

...

But then, technically, this means that for each symbol, which used

as

...

selector, you will need to add an additional instance of Selector, which wraps around symbol.

no, it does not wraps around... it has a completely different set of messages that it responds to, it does not wrap it, because if in that situation we say that selector wraps a symbol, then a date wraps a number (its day number), and Character wraps a number (its value), etc. Symbols and selectors are not completely polymorphic under the model I propose, the same way Character is not completely polymorphic with Number, etc. If I understand your point, you should not have Character because it only holds a Number ? is that right?

Character and Number is quite different by nature. As to me, for Symbol/Selector a better analogy would be Fruit and Apple.

...

...

Sounds not very good. But maybe there are benefits, which outweights such memory increase? Tell me more.

well I think I made my point very clear, you care about memory usage I care about object responsibilities, you care about implementation I care about design that encapsulates implementation... so I don't think we will agree until either one changes his vision on what it is more important... and I will not hehe (just kidding)

I care about all sides of it. And i'm not ready to sacrifice a memory usage for that. There's a lot more better places where i could put a free memory in use :)

...

...

It reminds me a MethodReference. It also identifies a CompiledMethod, without referring directly to class or method. But if you look closer at its implementation, you'll see that it is used as a dumb state holder, grouping different aspects of a method's state. It is used extensively by browser tools, even in places, where you could use and talk with compiled methods directly. Needless to say, that i can't understand such approach: why one

would

...

need to wrap things with a proxy, only to access the state of an object of interest though it? Why an object of interest can't give same answers directly?

I would understand that tools could be made so, that they can work with real objects as well as with proxies, which both honoring same protocol. But blindly wrapping each object with proxy?? This is beyond my understanding abilities.

I can't comment based on your description, I don't know MethodReference, what represents and what it is used for... sorry, cant help here.

...

...

...

Objects could play many roles, depending on their use. Just take

a

...

...

...

look at Object protocol, and answer yourself, how many of these methods is userful in some specific case (such as an object used to identify a message - selector). So, if you want to dissect all roles onto a set of specific (sub)classes, i propose to start from an Object class, and then, gradually get to a Symbol.

Not a bad idea!! Object is completely out of control and we should remove a lot of the methods it has! I completely agree there :-)

Hehe. Good luck with that :)

thanks! but I'm not going to do it... I would start from zero than trying to fix something that is already broken :-)

...

Sure thing, i'd like to see less methods in Object class.

But "Object class" has no methods! it is Object that has a lot of methods! (just kidding, I know you mean the class Object and no "Object class") Bye! Hernan.

...

...

...

...

This idea brings implementation challenges of course, but

nothing new

...

...

...

...

:-) On Wed, Aug 25, 2010 at 2:05 PM, Stéphane Ducasse <stephane.ducasse@inria.fr> wrote:

...

:)

Your previous mail was what the bee I had in my bonnet :)

Stef

On Aug 25, 2010, at 6:57 PM, Hernán Morales Durand wrote:

...

Actually I think is one of the better ideas I've read in

this

...

...

list. Too bad nobody seems interested in semiotics :(

2010/8/25 Stéphane Ducasse <stephane.ducasse@inria.fr>:

...

I;m sorry that I raised a stupid idea. Don;t bash people for me :)

On Aug 23, 2010, at 10:26 PM, stephane ducasse wrote:

...

Hi guys

I'm thinking aloud... I was wondering if we could not take advantage of a new class: Selector. MethodDict would only contain selectors and not symbols.

I remember that I talked about that with hernan.

Stef

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

...

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr

http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Best regards, Igor Stasenko AKA sig.

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Marcin Tustin Mobile: 07773787105 Office: 020 3400 3259

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- Marcin Tustin Mobile: 07773787105 Office: 020 3400 3259

_______________________________________________ Pharo-users mailing list Pharo-users@lists.gforge.inria.fr http://lists.gforge.inria.fr/cgi-bin/mailman/listinfo/pharo-users

-- *Hernán Wilkinson Agile Software Development, Teaching & Coaching Mobile: +54 - 11 - 4470 - 7207 email: hernan.wilkinson@10Pines.com site: http://www.10Pines.com <http://www.10pines.com/>*

2010/8/25 Igor Stasenko <siguctua@gmail.com>:

...

Meaning could be anything. It depends from context where i using this object. From machine's point of view there is no meaning, only semantic.

Igor, there are *TONS* of books about meaning (Saussure, Eco, Wittgenstein, etc), meaning is everything but anything. The richer understanding of meaning the better you are understanding systems. And there is no semantic from a machine's point of view (it was a claim of strong AI for many years).

So, i am asking, what _technical_ benefits it would brings us.

From a functionalist point of view, there are no additional benefits.

Because from meaning point of view, a symbol #foo reads foo, as well as selector foo, and only we (humans) know that it is selector in one context, or just a simple 'unique string' in another one.

...

...

i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol. But this is not strong enough. In practice, such shrinking, could lead to explosion of complexity in many other places.

'foo' = #foo  ->true #foo = 'foo' -> true

#foo asSelector = #foo ??

clearly no, we are saying that a symbol should be different from a selector, therefore they should never be equal no matter if their "printable" representation is the same (as in this example #foo)

...

if so, then #foo hash = #foo asSelector hash ?

could or could not... depends on the implementation, who cares?

...

and , obviously #foo == #foo asSelector  => false

yes, and #foo = #foo asSelector --> false

...

which means, that mixing selectors and symbols in a single collection will be very bad idea, and source of bugs and confusion.

why? It is not the same the relationship between a symbol and a string that with a symbol and a selector. We could say that a symbol and a string represent a sequence of characters, so if they represent the same sequence of characters they have to be equal, that is why #foo = 'foo', (that means 'foo' asUniqueString = 'foo' where asUniqueString is the real meaning of the # in front of the symbols...), but a selector does not represent a sequence of character, it represents the name of a message. Therefore for me a selector has a "name" that is a symbol, but a selector IS NOT a symbol, because can answer messages like numOfArgs:, sendTo: etc.

it could have a sense to use different entity, when you would need to hold additional state. I can imagine a class, like 'measure unit', with two variables - value and units. So, you can express 1 cm, 1 mm, 1 km, 100 kg using such object. But in case of selectors , you expressing a symbol, and there is no need for additional state.

A selector could have an intention, a duration or time of life (have you ever heard about mRNA?), an impact factor, a context (to validate its application), etc. A lot of very interesting and new things could be done! Ok, may be there is no need for this at all, but there is no need for hundreds of natural or artifficial languages too.

...

...

But what gives us a uniqueness of symbols is, that we don't need to use #= for comparing them, but #== , which is much faster. Right?

well, I think that using #== is a wrong design decision, for me you should only use #== if you are dealing with "memory" issues, if you really want two objects to be in the same memory position, therefore it is an implementation issue, but no if you are designing for extensibility....

...

Now, if you introduce selectors, there is no uniqueness anymore,

why?

well, because two things which having same name should be equal. Otherwise it makes no sense to use same name for them.

And what we do with polysemic names? :)

Keep in mind, that symbols are made for humans. Machine feels pretty well with pointers and don't needs to know anything about names.

...

...

since selector is a symbol (since it is a subclass of it),

why? I would not make it a subclass... using a subclass only to reuse code it is a bad design decision from my point of view, Selector should be a class by itself whose instances know a name, that are instances of Symbol

...

and must honor the rules of its ancestor.. but its not. The only way how to deal with that is to make particular symbol (sub)instance to be either an instance of Symbol or Selector but not both. This means that in system you could have either #foo or #foo asSelector but not both, otherwise you will violate the uniqueness rule.

the problem you see is, as I said above, because you are subclassing symbol which is not what I would do.

So, you want a completely separate class for selectors, which by sending #name (for instance) will answer a symbol which identifies them? I see, use delegation instead of inheritance.

Sorry I cannot comment about implementation because I'm not too interested in super technical issues, but I recognize the difficulty :) Cheers, Hernán

Igor Stasenko wrote

Symbol + ??? => Selector

give me a strong reason.

i like the idea of shrinking a Symbol's protocol and putting it into lean and clean place - Symbol.

While fixing Issue 15523: "Code Cruft Rule Only Matches One-Liners", I reeeeeally wanted: Selector UnarySelector BinarySelector KeywordSelector so that I could double dispatch. It made me remember this old interesting thread. So, in addition to the conceptual clarity reification would provide, this provided a clear (to me) case of technical benefit. We could eliminate the ~20 or so senders of isUnary, ~20 of isBinary, and ~20 isKeyword in the Kernel for a start... ----- Cheers, Sean -- View this message in context: http://forum.world.st/funny-idea-Symbol-vs-Selector-tp2335774p4826193.html Sent from the Pharo Smalltalk Users mailing list archive at Nabble.com.