[squeak-dev] Multiple issues with class definition

I just found, that when i tried to redefine previously declared
regular class to a variable one,
the ClassBuilder bugs me with error, that 'class can't be recompiled'.

Just two doits to reproduce the problem:

Object subclass: #Blabla
        instanceVariableNames: ''
        classVariableNames: ''
        poolDictionaries: ''
        category: 'Whatever'.

Object variableSubclass: #Blabla
        instanceVariableNames: ''
        classVariableNames: ''
        poolDictionaries: ''
        category: 'Whatever'.

i wonder what the reason of preventing me to redefine a mistakenly
defined class, which having 0 methods? And even if i have the methods,
so what?
I think that this limitation is too strict.
I find it quite inconvenient , since now, to redefine a class, i have
to rename old one, then define a new one which is variable from a
start and then manually copy methods from old one.


Another funny thing that i found, that you can actually do:

| string |
string := 'Blabla'.
Object subclass: string
        instanceVariableNames: ''
        classVariableNames: ''
        poolDictionaries: ''
        category: 'Whatever'.

Smalltalk at: string " returns a class you just defined."

Its fun, but having no sense. What is interesting that OB can display
this class (but not in a regular way, you can find it in smart groups
pane, after searching for 'Blabla').

The above is true for both 3.10 and  Pharo-10178 (and i think a more
recent Pharo have same behavior).


--
Best regards,
Igor Stasenko AKA sig.

Btw, the latter proves, that class naming has nothing to do with
system internals, it only provides a means for developer to identify
the class so, then he can use a browsing tools for finding it.
This is, of course, not something that i didn't knew before :)
Just a thought, that by altering the browser (and some/many other
places), one could have classes named as:

'MyPackage.MyClassName'
or
'MyNamespase.MyClassName'

--
Best regards,
Igor Stasenko AKA sig.

On 10.04.2009, at 05:34, Igor Stasenko wrote:

Yes, reflectively generating classes is what the browser does. And in  
addition,
it is useful for tests. Or in general meta-programming tricks of the  
evil kind.
(e.g. eToys used it to create on-the-fly classes for it's prototype-
like object
model. In most of these cases, it is better to use anonymous  
Behaviors, than you don't
see them in the browser at all).

> Its fun, but having no sense. What is interesting that OB can display
> this class (but not in a regular way, you can find it in smart groups
> pane, after searching for 'Blabla').


I think this is because the new class is not packaged. The old browser  
shows
it in the class category 'Whatever', indistinguishable from a class  
created with
the browser.

        Marcus

--
Marcus Denker  --  [hidden email]
http://www.marcusdenker.de

2009/4/10 Marcus Denker <[hidden email]>:
Right, but for common case, this should be errorneous? No?
At least Smalltalk>>#at:put: should check that key is symbol.

I'm not sure what the difference from class created by browser..
A browser IMO should use very same message(s), sent to a class which
should be a superclass for newly created new class(es). Doing
otherwise would lead to multiple entry points into classes internals,
which leads to exponential growth of complexity.



--
Best regards,
Igor Stasenko AKA sig.

Igor Stasenko wrote:
This is a bug. I could track this to a change somewhere between Squeak
3.2 and Squeak 3.6 (i.e., the bug is not in 3.2 but is in 3.6) where a
consistence check was changed from #subclassesDo: to #withAllSubclassesDo:.

The test now does a little more than it used to do but it also applies a
check that should only be run between two *different* classes (i.e., the
subclass and its superclass) to the *same* classes (the superclass and
itself). To fix it, change
ClassBuilder>>validateSubclass:canKeepLayoutFrom:forSubclassFormat: to
have a line saying:

        subclass == oldClass ifTrue:[^true]. "this ain't no subclass"

Please file a mantis bug for this problem.

Yeah, I'd say that's another bug. Somewhere there needs to be an
#asSymbol in the middle. Can you file another bug for this?

Cheers,
   - Andreas

2009/4/10 Andreas Raab <[hidden email]>: http://bugs.squeak.org/view.php?id=7332
or maybe 'self assert: name isSymbol' ?

> Can you file another bug for this?
>

http://bugs.squeak.org/view.php?id=7333

> Cheers,
>  - Andreas
>
>
>



--
Best regards,
Igor Stasenko AKA sig.

On 10.04.2009, at 11:29, Igor Stasenko wrote:

Ups. I did not realize the string vs. symbol problem... I blame that on
the jet-lag ;-)

Maybe Smalltalk>>#at:put: could do an #asSymbol?

But in general, the whole structural model is not checked too much for  
consistency.
If you do wrong, it will be wrong.

        Marcus

--
Marcus Denker - http://marcusdenker.de
PLEIAD Lab - Computer Science Department (DCC) - University of Chile