__noSuchProperty__ poll

[Hannes Wallnoefer]

I'm with Marcus: It would be preferable to have consistent behaviour, 
but not if there's a noticeable performance penalty.

Hannes

Am 2013-04-08 20:27, schrieb Marcus Lagergren:
> I think consistency makes sense, but there might be potential performance implications for modifying the index access. These should probably be investigated.
>
> On Apr 8, 2013, at 7:30 PM, Jim Laskey <jlaskey at me.com> wrote:
>
>> In Nashorn, we introduced __noSuchProperty__ as a means of overriding property search when a property is not declared. Example;
>>
>> jjs> var other = { a: 10, b: 20, c: 30 };
>> jjs> var obj = { delegate: other, __noSuchProperty__: function(name) { return this.delegate[name]; } }
>> jjs> obj.b;
>> 20
>>
>> In the current version of Nashorn, we don't call the __noSuchProperty__ handler when accessed using indexing.  However, the question has come up as whether this support should extend to indexing as well; providing symmetry between property access and indexing.  Thus continuing the example;
>>
>> jjs> obj["b"];
>> 20
>>
>> Does this make sense to everyone?
>>
>> The only argument I could come up against was using indexing for property testing, as in;
>>
>> jjs> if (obj["b"]) print(obj.b);
>> 20
>>
>> But that argument doesn't hold since there are alternatives;
>>
>> jjs> obj.hasProperty("b");
>> false
>>
>> or;
>>
>> jjs> "b" in obj;
>> false
>>
>> The reason I ask is that we have a bug where megamorphic call sites switch to indexed access (to reduce the number of cases) and thus __noSuchProperty__ is not invoked when it should.
>>
>> If we change the behaviour, then we will provide symmetry and address the bug.
>>
>> Comments?
>>
>> -- Jim
>>