On 2 août 2013, at 15:47, Esteban Lorenzano wrote:

On Aug 2, 2013, at 3:39 PM, Camille Teruel <camille.teruel@gmail.com> wrote:

...

Just a citation:

“I spent a few weeks ... trying to sort out the terminology of ‘strongly typed’, ‘statically typed’, ‘safe’, etc., and found it amazingly difficult ... The usage of these terms is so various as to render them almost useless.” -- Benjamin C. Pierce

If you use one of these terms, you should give a definition. And if you hear one, you shouldn't assume the author means what you think. For example some people could argue that Smalltalk is untyped (or unityped [1]), while you argued that it is strongly typed since there is no implicit type conversion. For most people, the static/dynamic and the weak/strong distinctions are supposed to be orthogonal. So, for most people, strong typing is not a synonymous of static typing.

I have to say that IMHO, you are wrong in this one. For researchers, maybe. But I have seen in "industry word" that the terms are often exchangeable and for most people I have known, strong=static :(

It depends on the people one knows then ;) All that shows that these terms are unclear for everyone and must be defined when used.

...

[1] http://existentialtype.wordpress.com/2011/03/19/dynamic-languages-are-static...

On 2 août 2013, at 06:03, btc@openInWorld.com wrote:

...

greetings all,

I'm in the final weeks of writing up my Masters dissertation and seeking some scholarly references to Smalltalk being "Strongly Typed."

I my review of Smalltalk I was surprised to find that [1] describes Smalltalk as Strongly Typed, since Smalltalk is sometimes denigrated as being untyped / weakly typed. From reviewing discussion forums this now makes sense, but I can only find one of scholarly reference that briefly mentions this [2]. The most enlightening is [3] which defines Type Strength as:

"A strongly typed language prevents any operation on the wrong type of data. In weakly typed languages there are ways to escape this restriction: type conversions"

meaning that getting a MNU is a form of Strong Typing since you can't make a Smalltalk object run a method that is not its own. The problem appears to be that Strong Typing has been synonymous with Static Typing for a long time, and Static Typing strongly ties types to variables, except in Dynamically Typed languages, I think types can be considered independently from variables, in which case the definition of [3] has some merit, hence Smalltalk is Strongly Typed.

Sounds controversial, so I'm just hoping for some peer reviewed backup - but only you have something easily to hand. This is just a small thing I can just leave out if necessary.

cheers -ben

[1] http://www.squeak.org/Features/ [2] p15, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.35.7507&rep=rep1&ty... [3] http://www.cs.kuleuven.ac.be/publicaties/rapporten/cw/CW415.pdf

I wonder if 'types' can be applied to smalltalk at all. Saying it is 'unityped' language (since everything is an object) is same as saying it has no types. Because where you need types? When you want to manipulate with data, but in smalltalk all data manipulation semantics is provided and implemented at primitive level, but not by the language itself. At language level the only semantics which is defined is message passing, closures, return and assignment. There is no any rule(s), in smalltalk at language level, saying that result of expression: 1 + 2 should be of integer type. Because this is provided by implementation but not defined by language itself e.g.: it is just a method in corresponding class, which implemented in such a way that result of expression will be instance of Integer (or SmallInteger to be precise). Now, if you change implementation so result will be float (or anything else), will such change allow you to say, that it is no longer smalltalk, or that now it is weakly typed or "anything-else typed"? Because smalltalk code do not operates directly with data, but only with references to it (which is objects). Take an assignment, for example. In staticly-typed language assignment copies the value of expression to variable, e.g.: a = 10 "now variable a has value = 10" in smalltalk, assignment is not copying value, but changing the reference: a := 10 "now variable a points (or refers) to object '10'" and thus, whatever 'type' used to represent the integer value 10 is completely irrelevant for assignment operation. -- Best regards, Igor Stasenko.

That is correct, there are no types in Smalltalk. Cheers, Hernán El 02/08/2013 15:55, Igor Stasenko escribió:

I wonder if 'types' can be applied to smalltalk at all. Saying it is 'unityped' language (since everything is an object) is same as saying it has no types.

Because where you need types? When you want to manipulate with data, but in smalltalk all data manipulation semantics is provided and implemented at primitive level, but not by the language itself. At language level the only semantics which is defined is message passing, closures, return and assignment.

There is no any rule(s), in smalltalk at language level, saying that result of expression:

1 + 2

should be of integer type. Because this is provided by implementation but not defined by language itself e.g.: it is just a method in corresponding class, which implemented in such a way that result of expression will be instance of Integer (or SmallInteger to be precise). Now, if you change implementation so result will be float (or anything else), will such change allow you to say, that it is no longer smalltalk, or that now it is weakly typed or "anything-else typed"?

Because smalltalk code do not operates directly with data, but only with references to it (which is objects).

Take an assignment, for example. In staticly-typed language assignment copies the value of expression to variable, e.g.:

a = 10

"now variable a has value = 10"

in smalltalk, assignment is not copying value, but changing the reference:

a := 10

"now variable a points (or refers) to object '10'"

and thus, whatever 'type' used to represent the integer value 10 is completely irrelevant for assignment operation.

" Smalltalk has types: a set of objects and the messages they understand" Nope thats not what a type is. A type is a category of data. Its that simple. Your set of objects and the messages they understand can do thousands of things. Emulating types is one of those things. Also what happens in implementation is not related to language. Implementation is one thing language is another. Also even if you see it philosophical still is very clear. Types in the end is a set of restrictions which tells "I am the language and I wont allow you to shoot yourself on the foot. I know better." Smalltalk on the other hand begs you "Hack me dude, hack me!". Smalltalk does not enforce such restrictions. But then it does not prevent you from implementing such restrictions. Afterall a static typed language does not prevents from implementing dynamic types. If it did it would be impossible to use C to create python or smalltalk. With any language there are a billion ways to skin a cat. What it differs is the things they offer out of the box .Nothing is set in stone. Carla F. Griggio wrote

Well, I've always defined a type as the a set of values and the operations you can do with them. For example, the set of values { 1, 2, 3, 4, ... } and the operations { + , - , * , ... } would be the type Integer. With that in mind, *Smalltalk has types: a set of objects and the messages they understand*. And thanks to polymorphism a type *is not* a synonym of a class, as instances of different objects can understand messages in common. So yes, "everything is an object" beacuse all objects are instances of a class that inherits from Object, but from my point of view it's not true that all objects are of one same type.

Everytime an object doesn't understand a message is telling you that you're doing an operation on the wrong type of object.

As some others have already said, I don't like the terms "strongly typed" or "weakly typed" because sometimes is ambiguous, and I rather say this: *Smalltalk has dynamic type-checking in runtime* (instead of compilation time), because it checks if an object can perform the operation you want (the message) in the moment of executing that operation (when you send the message).

Now, talking not-so-siriously, and maybe I'm getting a little code-philosofical here... If we take into account what I've just defined as a type, and knowing that all objects are instances of a class that inherits from Object, I think the term "unityped" should actually be replaced for "multityped", as the same object could respond to many different types depending on the interface its client wants it to understand. But that happens maybe in any object-oriented program, the thing is that in Smalltalk it's very natural and you don't even have to think about it (in Java you would have to explicitly use 'interfaces').

On Sat, Aug 3, 2013 at 4:02 AM, Hernán Morales Durand <

hernan.morales@

...

wrote:

...

That is correct, there are no types in Smalltalk. Cheers,

Hernán

El 02/08/2013 15:55, Igor Stasenko escribió:

I wonder if 'types' can be applied to smalltalk at all.

...

Saying it is 'unityped' language (since everything is an object) is same as saying it has no types.

Because where you need types? When you want to manipulate with data, but in smalltalk all data manipulation semantics is provided and implemented at primitive level, but not by the language itself. At language level the only semantics which is defined is message passing, closures, return and assignment.

There is no any rule(s), in smalltalk at language level, saying that result of expression:

1 + 2

should be of integer type. Because this is provided by implementation but not defined by language itself e.g.: it is just a method in corresponding class, which implemented in such a way that result of expression will be instance of Integer (or SmallInteger to be precise). Now, if you change implementation so result will be float (or anything else), will such change allow you to say, that it is no longer smalltalk, or that now it is weakly typed or "anything-else typed"?

Because smalltalk code do not operates directly with data, but only with references to it (which is objects).

Take an assignment, for example. In staticly-typed language assignment copies the value of expression to variable, e.g.:

a = 10

"now variable a has value = 10"

in smalltalk, assignment is not copying value, but changing the reference:

a := 10

"now variable a points (or refers) to object '10'"

and thus, whatever 'type' used to represent the integer value 10 is completely irrelevant for assignment operation.

-- View this message in context: http://forum.world.st/Smalltalk-strongly-typed-tp4701894p4702012.html Sent from the Pharo Smalltalk Developers mailing list archive at Nabble.com.

BTW, I especially like the idea that Smalltalk is Multityped, as an object can respond to several protocols/interfaces/type specs. So true! Quoting "Carla F. Griggio" <carla.griggio@gmail.com>:

Well, I've always defined a type as the a set of values and the operations you can do with them. For example, the set of values { 1, 2, 3, 4, ... } and the operations { + , - , * , ... } would be the type Integer. With that in mind, SMALLTALK HAS TYPES: A SET OF OBJECTS AND THE MESSAGES THEY UNDERSTAND. And thanks to polymorphism a type _is not_ a synonym of a class, as instances of different objects can understand messages in common. So yes, "everything is an object" beacuse all objects are instances of a class that inherits from Object, but from my point of view it's not true that all objects are of one same type.

Everytime an object doesn't understand a message is telling you that you're doing an operation on the wrong type of object.

As some others have already said, I don't like the terms "strongly typed" or "weakly typed" because sometimes is ambiguous, and I rather say this: SMALLTALK HAS DYNAMIC TYPE-CHECKING IN RUNTIME (instead of compilation time), because it checks if an object can perform the operation you want (the message) in the moment of executing that operation (when you send the message).

Now, talking not-so-siriously, and maybe I'm getting a little code-philosofical here... If we take into account what I've just defined as a type, and knowing that all objects are instances of a class that inherits from Object, I think the term "unityped" should actually be replaced for "multityped", as the same object could respond to many different types depending on the interface its client wants it to understand. But that happens maybe in any object-oriented program, the thing is that in Smalltalk it's very natural and you don't even have to think about it (in Java you would have to explicitly use 'interfaces').

On Sat, Aug 3, 2013 at 4:02 AM, Hernán Morales Durand <hernan.morales@gmail.com> wrote:

...

That is correct, there are no types in Smalltalk. Cheers,

Hernán

El 02/08/2013 15:55, Igor Stasenko escribió:

...

I wonder if 'types' can be applied to smalltalk at all. Saying it is 'unityped' language (since everything is an object) is same as saying it has no types.

Because where you need types? When you want to manipulate with data, but in smalltalk all data manipulation semantics is provided and implemented at primitive level, but not by the language itself. At language level the only semantics which is defined is message passing, closures, return and assignment.

There is no any rule(s), in smalltalk at language level, saying that result of expression:

1 + 2

should be of integer type. Because this is provided by implementation but not defined by language itself e.g.: it is just a method in corresponding class, which implemented in such a way that result of expression will be instance of Integer (or SmallInteger to be precise). Now, if you change implementation so result will be float (or anything else), will such change allow you to say, that it is no longer smalltalk, or that now it is weakly typed or "anything-else typed"?

Because smalltalk code do not operates directly with data, but only with references to it (which is objects).

Take an assignment, for example. In staticly-typed language assignment copies the value of expression to variable, e.g.:

a = 10

"now variable a has value = 10"

in smalltalk, assignment is not copying value, but changing the reference:

a := 10

"now variable a points (or refers) to object '10'"

and thus, whatever 'type' used to represent the integer value 10 is completely irrelevant for assignment operation.

Cheers, Juan Vuletich

btc wrote

btc@

wrote:

Kilon, I agree with most of you writeup, except that I think Smalltalk has types - except the types are not associated with variables, rather they are associated with objects. This aligns with Igor's comment "in smalltalk, assignment is not copying value, but changing the reference".

Actually I also mostly agree with Igor point. If I am not mistake we agree on the fact that smalltalk has no types. The only thing I would not agree, if my assumption is correct here, is that Igor appears here implying that smalltalk somehow is special to assign values by reference. If this assumption of mine is correct then I do disagree strongly , its actually a very common practice. Python also implement this, Lisp does, ruby and an array of dynamic type languages. I am not so sure about C and C++ but since they use pointers its a moot point anyway. I explain this concept in my book about python that I have published in blenderwiki , section "Reference" which can be found here -> http://wiki.blender.org/index.php/User:Kilon/Python_book_of_magic/05.Diction... However this concept is implementation specific and not language specific. Its obviously a massive diffirence because a coder rarely worries about the implementation itself the same way a user of an OS does not worry how and why an OS does the things it does, he only care about what it does. Also I am not sure how that prove the existence of types, for me it does not, it only describes how values are assigned. What prevents you from assigning by reference typeless data ? Unless once again we talk about how pharo itself implement this and somehow uses its own internal type system. The question that remain why I the coder should care about this , if at language level Smalltalk is typeless. Maybe Igor can clarify this more to make a more clear picture what happens at implementation level. But at language level yes I have seen nothing that proves the existence of types inside Smalltalk unless we treat types in a very vague kind way which for me it just defeats their purpose. I also agree on the remark of Duck Typing , it definitely smells like smalltalk , but once again the terminology is failed because it implies a type system. Once more I have to stress the importance of dichotomy between language specification and language implementation. Types obviously touch both areas but not necessarily for same reasons. -- View this message in context: http://forum.world.st/Smalltalk-strongly-typed-tp4701894p4702028.html Sent from the Pharo Smalltalk Developers mailing list archive at Nabble.com.

Hi Ben, Even if you adopt a type system or definition as part of Smalltalk, it is not coherent nor intuitive with the system usage. Types are not identifiable in Smalltalk, they belong to other systems semantics (both formal and informal) and probably scientific corpus. Because of that, you are having problems to cite those concepts. And type system definitions are inherently unstable because there is no scientific consensus. Does typing correspond with common reality in Smalltalk developement? Why would you need to think in terms of type checking in an object environment? It is ok if you don't want to break contractualism in your thesis, although to me talking about type checking in Smalltalk is the same as talking about L2 cache hits ;) Cheers, Hernán El 03/08/2013 14:35, btc@openinworld.com escribió:

btc@openinworld.com wrote:

...

greetings all,

I'm in the final weeks of writing up my Masters dissertation and seeking some scholarly references to Smalltalk being "Strongly Typed."

I my review of Smalltalk I was surprised to find that [1] describes Smalltalk as Strongly Typed, since Smalltalk is sometimes denigrated as being untyped / weakly typed. From reviewing discussion forums this now makes sense, but I can only find one of scholarly reference that briefly mentions this [2]. The most enlightening is [3] which defines Type Strength as:

"A strongly typed language prevents any operation on the wrong type of data. In weakly typed languages there are ways to escape this restriction: type conversions"

meaning that getting a MNU is a form of Strong Typing since you can't make a Smalltalk object run a method that is not its own. The problem appears to be that Strong Typing has been synonymous with Static Typing for a long time, and Static Typing strongly ties types to variables, except in Dynamically Typed languages, I think types can be considered independently from variables, in which case the definition of [3] has some merit, hence Smalltalk is Strongly Typed.

Sounds controversial, so I'm just hoping for some peer reviewed backup - but only you have something easily to hand. This is just a small thing I can just leave out if necessary.

cheers -ben

[1] http://www.squeak.org/Features/ [2] p15, http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.35.7507&rep=rep1&ty...

[3] http://www.cs.kuleuven.ac.be/publicaties/rapporten/cw/CW415.pdf

Thank you all for you responses. I am integrating it all into a section of my report, which I'll paste in a bit for you to pick holes in. I few quick responses....

I agree with Igor and Camille, statically-typed & strongly-typed should not be synonyms - but as Esteban says, it is common usage - but I think that doesn't make it right. I think this is due to statically-typed languages tightly associating types with variables, whereas types exist independently from variables. Milan, To clarify, it doesn't prevent the *error*. What it does is prevent the *operation*. Since a MNU is produced the specified operation did not execute. Igor gives a good example of why C is weakl typed.

Camille, I take your point about defining "Strongly Typed" , which I will do in my report.

Kilon, I agree with most of you writeup, except that I think Smalltalk has types - except the types are not associated with variables, rather they are associated with objects. This aligns with Igor's comment "in smalltalk, assignment is not copying value, but changing the reference".

Enrico, I will cover Duck Typing, thanks. The implication is, if Duck Typing applies to Smalltalk, then types *are* part of Smalltalk. Alain, Thanks for the link. To much to respond to here. Bits should end up in my report. I found this compelling, as it clarify my own thinking along the same lines "Statically typed languages assign types to variables, while dynamically typed languages assign types to values"

Hernan, I disagree that Smalltalk has no types. But they apply to objects rather than variables, which is where it breaks down for the explicitly/statically typed crowd.

Carla, I am thinking along those lines. A type groups data and valid operations. In non-OO languages, the data and operations are separated, so the type is associated with a variable, which defines what operations are valid on the variable, and hence on the data within the variable. In OO/Smalltalk, the data and operations a combined, so type is associated with the object, which defines what operations are valid on the object, and hence the data within the object.

Juan, I agree with your first post that Smalltalk is strongly typed. Discussing of multi-typed is beyond what I need to consider for my report.

German, That paper states "Smalltalk, although an untype language" but the goes on to indicate many times that Smalltalk has types eg "subtype polymorphism of Smalltalk" I think it is common to say "Smalltalk is untyped" but actually the correct statement should be "In Smalltalk, the *variables* are untyped"

I'll paste something in more detail soon.

Thanks again. All good points to round out my discussion.

cheers -ben