On 2011/01/11 01:40, Levente Uzonyi wrote:
>
>
> Quoting Frank Shearar <frank.shearar at angband.za.org>:
>
>> On 2011/01/02 11:13, Levente Uzonyi wrote:
>>> On Fri, 31 Dec 2010, Frank Shearar wrote:
>>>
>>>> On 2010/12/31 16:12, Levente Uzonyi wrote:
>>>>> On Fri, 31 Dec 2010, Frank Shearar wrote:
>>>>>
>>>>>> On 2010/12/31 13:59, Enrico Spinielli wrote:
>>>>>>> Hi,
>>>>>>> I still have the following applied in my image and I see it is still
>>>>>>> not yet in Trunk.
>>>>>>> Any core developer willing to commit it?
>>>>>>
>>>>>> It's not in the Inbox anymore and you're right, it wasn't applied. I
>>>>>> have a local copy I can put back in the Inbox, if it's convenient? Or
>>>>>> if preferred I can remake the submission based on Network's latest
>>>>>> version.
>>>>>
>>>>> I moved it to the treated inbox, because another solution was applied.
>>>>> SchemeRegistry has HttpUrl at nil, so Url urlClassForScheme: nil
>>>>> should
>>>>> return HttpUrl.
>>>>
>>>> Right, I see so. My confusion in reconstructing history came from
>>>> Andreas suggesting the fix I implemented in Network-fbs.91, after
>>>> suggesting we either fix the call sites causing the issue or what you
>>>> then did on 2010/10/28, Levente, and then not seeing anything happen
>>>> after that (in my possibly too-brief investigation).
>>>>
>>>> (Why, by the way, do we _not_ call super initialize?)
>>>
>>>> From which method?
>>
>> Url class>>initialize. In a git-like diff we have
>>
>> +++ SchemeRegistry := Dictionary new
>> --- super initialize.
>> --- SchemeRegistry := Dictionary new.
>> --- SchemeRegistry
>> at: 'browser' put: BrowserUrl;
>> at: 'file' put: FileUrl;
>> at: 'ftp' put: FtpUrl;
>> at: 'http' put: HttpUrl;
>> at: 'https' put: HttpUrl;
>> at: 'mailto' put: MailtoUrl;
>> at: nil put: HttpUrl;
>> yourself
>> --- at: 'mailto' put: MailtoUrl.
>
> There are a few reasons why we shouldn't send #initialize to super from
> the class side. In this case it doesn't make a real difference, because
> Url is a subclass of Object, and Object implements #initialize on the
> class side relatively safely. What are the problems with the super sends?
> - if your class is not a subclass of Object, but ProtoObject, then
> sending #initialize to super will mess up your class, because you will
> execute Behavior >> #initialize.
> - a superclass may initialize class variables which are shared by it's
> subclasses, doing that from the subclasses may break some stuff.

Right. The subclass can re-initialise all those variables that contain 
things the superclass needs. Classes are also long-lived objects, and 
might contain state gathered over time.

That means that if one wants to protect one's class against subclasses 
doing a super-send of #initialize one ought to lazily initialise one's 
class instance variables, right? (I hope I got the terminology right 
there: instance variables of the class.) Like so:

MyVar := MyVar ifNil: [MyVar := MyVar new]

Maybe we ought to put a warning in Behavior>>initialize then? Something 
like what we do with Undeclared:

Transcript showln: 'Class ', self className, ' should not call 
Behavior>>initialize.'.

frank

> Levente
>
>>
>> frank
>>>
>>>
>>> Levente
>>>
>>>>
>>>> frank
>>>>>
>>>>>
>>>>> Levente
>>>>>
>>>>>>
>>>>>> frank
>>>>>>
>>>>>>>
>>>>>>> Hope it helps
>>>>>>> Bye
>>>>>>> Enrico
>>>>>>>
>>>>>>> On Wed, Sep 8, 2010 at 08:31, Frank Shearar
>>>>>>> <frank.shearar at angband.za.org> wrote:
>>>>>>>> On 2010/09/08 08:04, Andreas Raab wrote:
>>>>>>>>>
>>>>>>>>> On 9/7/2010 11:01 PM, Frank Shearar wrote:
>>>>>>>>>>
>>>>>>>>>> There's a quicker test: Do It with: Url absoluteFromText: 'foo'
>>>>>>>>>>
>>>>>>>>>> Up until my submission (which is causing this trouble), "no
>>>>>>>>>> scheme"
>>>>>>>>>> meant "use an HttpUrl".
>>>>>>>>>>
>>>>>>>>>> I can see two ways forward: (a) fix the calling sites like C. A.
>>>>>>>>>> Oliver's submission, or (b) add "at: nil put: HttpUrl" in the
>>>>>>>>>> SchemeRegistry.
>>>>>>>>>>
>>>>>>>>>> I prefer (a) myself: 'foo' isn't any kind of valid _absolute_
>>>>>>>>>> URI.
>>>>>>>>>>
>>>>>>>>>> The Inbox contains Network-fbs.90, which contains (b)'s fix.
>>>>>>>>>
>>>>>>>>> Perhaps even easier:
>>>>>>>>>
>>>>>>>>> ^SchemeRegistry at: (scheme ifNil:['http']) ifAbsent:[GenericUrl]
>>>>>>>>
>>>>>>>> I prefer this to my submission: it doesn't pollute SchemeRegistry.
>>>>>>>>
>>>>>>>> Network-fbs.91 contains it.
>>>>>>>>
>>>>>>>> frank
>>>>>>>>
>>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>
>>>>>>
>>>>>>
>>>>>
>>>>>
>>>>>
>>>>
>>>>
>>>>
>>>
>>>
>>>
>>
>>
>>
>
>
>
>
>
>
>