[OpenLayers-Dev] 2.10 and 2.11-RC1 OpenLayers.Class behavior changes

[Eric Lemoine]

On Sun, Aug 7, 2011 at 12:01 AM, Eric Lemoine
<eric.lemoine at camptocamp.com> wrote:
> On Sat, Aug 6, 2011 at 9:32 PM, RICHARD Didier <didier.richard at ign.fr> wrote:
>>
>>>
>>>>>>> 2/ in test 4, if you remove the initialize function from B and
>>>>>>> overload
>>>>>>> A,
>>>>>>> the B constructor is still the previous A's constructor.
>>>>>>
>>>>>> Yes. To address this one I see no other solution than patching
>>>>>> OpenLayers.Class. See the patch attached to this email, and my
>>>>>> test_overload_5 test function. The Class.html tests continue to pass
>>>>>> with my patch.
>>>>>>
>>>>>
>>>>> this works in OL 2.11, not in OL 2.10 for me !-(
>>>>> I guess there is something similar to do in OL 2.10 ?
>>>>
>>>> Your initial mail was about OpenLayers trunk/2.11 introducing
>>>> regressions. Now the failing test that remains is OL 2.10-only, so OL
>>>> trunk/2.11 works better than OL 2.10 for you. Yippee! :-)
>>>>
>>>
>>> My initial mail was really about regression between 2.10 and 2.11 !-)
>>>
>>>> (Not sure it is worth "fixing" OL 2.10 for that.)
>>>>
>>>
>>> Well, one of my test is versus the current release (2.10) and I am sure
>>> there are users that will migrate slower than us (I mean OL and IGNF) ...
>>>
>>> I do think it is worth fixing it ... at least to prove there is no
>>> regression !-D
>>>
>>
>> I've got somethinf working, hence ugly :-)
>>
>>    OpenLayers.overload= function(P,F) {
>>        var pProtoInitialize= typeof(F.initialize)=="function"?
>>            P.prototype.initialize
>>        :   null;
>>        OpenLayers.Util.extend(P.prototype, F);
>>        if (pProtoInitialize!==null) {
>>            // override sub-class having same constructor:
>>            for (var pn in P) {
>>                if (typeof(P[pn])=='function' &&
>> P[pn].prototype.initialize===pProtoInitialize) {
>>                    var f= {};
>>                    eval('f= {"initialize":'+F.initialize.toString()+'}');
>>                    P[pn]= OpenLayers.overload(P[pn],f);
>>                }
>>            }
>>        }
>>
>>        return P;
>>    };
>
> Hi Didier. Just to say that I personally wouldn't rely on an overload
> function that assumes subclasses use their parent constructors as
> their namespaces.

I can't imagine a clean/correct implementation of your "overload"
function that would make test_overload_5 pass with 2.10. If we apply
the OL patch that I've proposed we can address this test case in 2.11
(or 2.12). I still don't understand if that would be ok for you.


   function test_overload_5(t) {
       // overload constructor of parent class
       t.plan(1);
       var A = OpenLayers.Class({
           initialize: function() {
               this.a = "foo";
           }
       });
       var B = OpenLayers.Class(A, {});
       A = overload(A, {
           initialize: function() {
               this.a = "bar";
           }
       });
       var b = new B;
       t.eq(b.a, "bar", "ctor overloaded");
   }


-- 
Eric Lemoine

Camptocamp France SAS
Savoie Technolac, BP 352
73377 Le Bourget du Lac, Cedex

Tel : 00 33 4 79 44 44 96
Mail : eric.lemoine at camptocamp.com
http://www.camptocamp.com