On 26 ao�t 2013, at 13:31, Igor Stasenko wrote:

the intent was to replace
Smalltalk at:
Smalltalk globals at:

idioms with shorter one, and get rid of referencing Smalltalk global.

The are not for solving future problems with introduction of environments.

We want to get rid of Smalltalk globals because it currently resolve to an instance of SmalltalkImage, a class that does way too much things, including holding the currently unique environment, while a global could directly refer to it.

Also the references to Smalltalk globals that where really wrong are already gone and were in compiler.

And even if you introduce them, i still think you need a way to refer to globals (they are likely to stay)

Globals are global to their environment. We could also call them environment variable.

Whatever we will use for binding symbols in local environment, this is orthogonal.

Now you proposing introduce
ThisEnvironment at:

which getting back where we started from.

No because ThisEnvironment would be an instance of Environment (basically a refactored SystemDictionary) not of SmalltalkImage.

So who should be in charge of knowing how to handle accessible variable?

An instance of Environment with a nice api or an instance of Symbol by herself?

Unless , in your vision ThisEnvironment handled by compiler (like thisContext)..

so there will be no new global, but ThisEnvironment bound at compile time.

anyways, it will be semantically different from what we have now,
it will lookup for symbol in local environment, instead of global one.

ThisEnvironment would not be different than any other global.

There is no semantic change, no magic, nothing.

That's what the compiler already does.

The compiler use the environment of the class of the method being compiled to resolve globals.

It happens that all classes currently share the same environment, that's something else.

And this is not going to happen (you cannot replace
Smalltalk globals at: with ThisEnvironment at: without overlooking all the places where it used)...

while replacing it with #asClass is fine (because semantics stays same).

Also, #asClass means 'give me the class or throw error', while
Smalltalk globals at: , obviously don't do that and gives you any object bound to given symbol, if it there.

Any control?

at:ifAbsent:

at:ifPresent:

at:ifAbsent:ifPresent:

are already there and if you want to ensure the global is a class you could have: classNamed:ifAbsent:ifPresent:

On 26 August 2013 12:31, Camille Teruel <camille.teruel@gmail.com> wrote:

On 26 ao�t 2013, at 10:56, Esteban Lorenzano wrote:

>
> On Aug 25, 2013, at 9:40 AM, Camille Teruel <camille.teruel@gmail.com> wrote:
>
>>
>> On 24 ao�t 2013, at 19:20, Camillo Bruni wrote:
>>
>>> We have now:
>>>
>>> String >> #asClass
>>> String >> #asClassIfAbsent:
>>> String >> #asClassIfPresent:
>>
>> I don't understand why we need this new way.
>> Is it just to avoid calling 'Smalltalk globals at: #MyClass'?
>
> exactly
>
>> Because these names are confusing, and the 'as' prefix suggests a conversion while it's an access (with indirection but still a mere access).
>> And #asClass has no sender.
>> If you think 'Smalltalk globals at: #MyClass' is really too long to type, lets just create a new global ThisEnvironment := Smalltalk globals.
>
> this was already discussed. With #asClass and relatives what you have is a better abstraction jut because you are decoupled of "Smalltalk globals", it is not a big win now, but it open doors to better designs with environments, etc.
> At least, that was my understanding when the issue arise at the beginning.

It's not a better abstraction, it's a worse one.
Sending #asClass to a symbol will ever execute the same method that anyway as to rely on a global: Smalltalk.
And this global is solved in only one environment: the one of the class Symbol where #asClass and co. are implemented.
So if you want separated environments you know what kind of thing you'll end up writing?

'#Foo asClassInEnvitonment: self class environment'

or implementing:

Symbol>>#asClass
^ self asClassInEnvironment: thisContext sender receiver class environment

However, as soon as the compiler use the environment of the compiled method's class to solves the globals it contains, Smalltalk can refer to anything.
So the only problem with: 'Smalltalk globals at:' is that you send a message #globals to the global Smalltalk to fetch the environment while we could just have a global ThisEnvironment pointing to itself.

Writting 'ThisEnvironment at: #Foo' is much more natural.

If we want separated environments, each environment just have to define a binding ThisEnvironment pointing to itself and everything works well.
What I don't like with #asClass and co. is the style: you talk to a symbol instead of an environment.
It's like saying: '#key valueIn: aDict' instead of: 'aDict at: #key', 'letter beSentBy: postman to: receiver' instead of: 'postman send: letter to: receiver', etc...

In Smalltalk we already have self to refer to the current receiver and thisContext to refer to the current stack frame. If we want different environments, is it that weird to have a ThisEnvironment to refer to the current environment?

>
> Esteban
>
>>
>>
>>>
>>> On 2013-08-24, at 17:55, Fernando Olivero <fernando.olivero@usi.ch> wrote:
>>>> I prefer to evaluate
>>>>
>>>> Smalltalk globals classNamed: #MyClass
>>>>
>>>> Fernando
>>>>
>>>> On Sat, Aug 24, 2013 at 11:57 AM, St�phane Ducasse
>>>> <stephane.ducasse@inria.fr> wrote:
>>>>>
>>>>> Is it
>>>>>
>>>>> asClass?
>>>>>
>>>>> Stef
>>>>>
>>>>
>>>
>>>
>>
>>
>
>

--
Best regards,
Igor Stasenko.