Hi!
1) I think we are failing also at communicating one point better. It is
not that people is arguing against #asClass because it's ugly and bad
and a terrible villain. Ok, maybe a bit, but also:
The point is that #asClass, as it looks handy and easy to use, it
may not work in the future.
Why?
Because if you imaging a Pharo with modules, explicit imports, and
even namespaces, then you may have several classes with the same name.
And then you may have name clashes. And #asClass will not have a single
obvious result. And that makes #asClass both impractical and with no
sense at the same time.
One other reality:
- We are thinking about a problem we do not yet have... :)
2) Then, I agree with Doru, with Luc and with (E)(Ste(f|ban)). But I
also agree with Phil and myself. And my position says:
- Let the kernel be clean. There should not be an #asClass or similar
implementation as part of the kernel. This extension should exist not as
part of the kernel but as part of other package (I'd vote for
ScriptingExtensions for example).
- Let the kernel be clean (bis). We should not have users of such
*scripting nicety* inside the kernel. We should put in place a lint rule
to validate that.
- But we should let people use #asClass if they want in their code! And
thus we should not deprecate it. We should not control what Phil does to
get business. I agree that Pharo itself should not use #asClass, but
also that (a good implementation of) #asClass or similar should be
available for him. At the end, there will be so many packages and
libraries out there that we have to realize we can only guarantee that
Pharo gives you an empowering environment and people will use it and do
whatever they want :).
I updated the issue with some of these ideas
https://pharo.fogbugz.com/f/cases/18987/Extract-asClass-and-friends-in-a-separate-package-and-deprecate
Guille
-------- Original Message --------
Hi,
On Aug 26, 2016, at 9:10 AM, Esteban Lorenzano <[email protected]> wrote:
On 26 Aug 2016, at 08:49, Luc Fabresse <[email protected]> wrote:
Hi,
My point of view is:
1) in code/core, we should use (we already said that with Camille in the past
;-)):
self environmentAt: #Blah
Object>>environmentAt: aSymbol
^ self class environmentAt: aSymbol
Object class>>environmentAt: aSymbol
...
The idea is that we can then customize name resolution, per object, per class
and per module in the future.
+1
2) For scripting purpose, asClass is indeed useful (GTInspector, ...).
I would start simple as said before, re-package it and add a rule as suggested
by Denis.
Now, I am not sure that making asClass supporting name resolution in another
environment is really useful.
And if we do it using thisContext, some developers may use that instead of "obj
environmentAt: XXX" which would be bad.
I dislike a lot the thisContext resolution idea.
If is bad is bad… and it will be bad also for scripting. I know, now it does
not looks like adding value, but think about: #asClass is monolithic and
#asClass with thisContext “looks monolithic”, IMO promoting a bad way of
thinking problems in Pharo thus inducing confusion for non expert users.
No problem from my side. It was a proposal. I wanted to learn a bit and I was
looking for concrete arguments to learn from because I am likely missing
something. I still do not know why it is bad, but it is really not important
for the current issue.
So, we agree that:
- move asClass together with all other in a separate package.
- introduce deprecation.
I created an issue:
https://pharo.fogbugz.com/f/cases/18987/Extract-asClass-and-friends-in-a-separate-package-and-deprecate
Doru
Esteban
my 2Kč,
#Luc
2016-08-26 6:56 GMT+02:00 Tudor Girba <[email protected]>:
Hi,
On Aug 26, 2016, at 6:37 AM, stepharo <[email protected]> wrote:
Thanks doru.
I do not like when people think that we are complaining just because something
changes.
It should change for the better and we all agree on that.
Certainly. There are many points of view and many constraints. This is why it
is so important that we all bring forward those constraints because only like
this we can reach a global maximum.
So, about asClass, everyone agrees that it should be moved to another package.
The open questions are:
- do we add an automatic deprecation for those that use it in code, or
- do we make use of thisContext to retrieve the environment?
(or both)
Also, what about the asClassInEnvironment: method? Can it be used in code, or
do we better discourage its usage altogether? I am thinking that if we have to
write:
#MyClass asClassInEnvironment: self class environment
is even longer than:
self environment at: #MyClass
so, I think there is little point to it.
In fact, for scripting what I find useful is not so much less characters, but
the lack of parentheses, hence unary methods. That is why asClass is worth
being salvaged for scripting (even with a solution that is slower with
thisContext), but the rest maybe can be removed. What do you think?
Cheers,
Doru
Stef
Hi,
There exists already a method for that:
Symbol>>asClassInEnvironment:
But, what if we introduce:
Symbol>>asClassFrom: anObject
^ self asClassInEnvironment: anObject class environment
?
The problem is asClass unary.
All the tools should be parametrized by an environment.
Yes, but asClassFrom: would not be unary but would save us from typing an extra
"class environment" :).
Yes I see.
But inside Pharo core tools we are ready to type
environment as a message that dispatch to something else than a symbol.
Sure.
This would allow us to still script and be dynamic.
Furthermore, as #asClass is meant to be mainly used for convenience, not
performance, I would also propose to make it lookup in thisContext and take the
environment from there. I know that his might sound like magic, but it would be
the default that we are looking for (to always lookup through the current
environment dynamically).
argh I will die....:)
No use of thisContext or only in the scripting package.
:D
Yes, yes. I just talked with Guille. Moving these scripting methods outside of
the Kernel is clearly a must.
I think that each time you use them we will preempt cross compilation and
others.
Yes, we agree that this method should not be used inside code.
I was just thinking that we can make it so that we do not break any code while
still making it dynamic.
I do not like your definition of dynamic. Sending a message to an object is
dynamic.
What you imply is compact. I can understand it when typing in playground.
By dynamic I meant the dispatch through “self class environment” or “self
environment” which is what people will use by default.
Like with scripting solutions there is a performance penalty, but that is fine if
people choose to pay it (like in the case of Symbol>>#value:).
Yes for scripts. Not for core code.
Since people tend to be a bit lazy I think that having rules will make sense.
Definitely.
Doru
Cheers,
Doru
What do you think?
Cheers,
Doru
On Aug 25, 2016, at 8:34 AM, Yuriy Tymchuk <[email protected]> wrote:
Just my 2 cents:
instead of
#name asClass
we have to use
self class environment at: #name.
Maybe instead of #at: we can have #classNamed:? Or something similar? Because
1) it’s not obvious that the method will give you a class, what if in the
future and environment can also have a mapping of something else like packages?
Uko
On 25 Aug 2016, at 07:21, stepharo <[email protected]> wrote:
Hi guys
We got a meeting at ESUG with all the compiler guys and james from gemstone.
Our goal is to have a full tool suite that can be parametrized by environments
(so that
we can compile code in other space, or compile other code inside pharo).
I personnally started this effort one decade ago. Now the introduction
of #asClass and friend is simply destroying all our efforts. There was a
discussion
in the past but we are not listened.
We will
- packaged these extensions in a separate package
- add rules to ban the use of such method in Pharo
- fix all the use (again) to use the correct way to do it.
I can understand that for scripting this is easier but it cannot be at that
cost and impact.
I hope that we will understand but we have to do something else than
fixing code that breaks our effort.
Stef, Marcus, Guille and Luc
--
www.tudorgirba.com
www.feenk.com
"It's not what we do that matters most, it's how we do it."
--
www.tudorgirba.com
www.feenk.com
"Quality cannot be an afterthought."
--
www.tudorgirba.com
www.feenk.com
"Every thing has its own flow."
--
www.tudorgirba.com
www.feenk.com
"Every thing should have the right to be different."
--
www.tudorgirba.com
www.feenk.com
"Be rather willing to give than demanding to get."
