array and diamond

[Maurizio Cimadamore]

On 11/12/11 15:21, Rémi Forax wrote:
> On 12/10/2011 12:38 PM, maurizio cimadamore wrote:
>> On 09-Dec-11 10:56 PM, Rémi Forax wrote:
>>> Is there a reason why the diamond syntax can't be used
>>> with an array ?
>>>
>>> List<?>[] list = new List<>[12];
>>>
>> Because the current inference rules would end up inferring:
>>
>> List<?>[] list = new List<Object>[12];
>>
>> If you special-cased diamond on arrays so that it is inferred as:
>>
>> List<?>[] list = new List<?>[12];
>>
>> Then it would be safe.
>
> but in that case
>   List<?>[] list = new List<?>[12];
> is not equivalent to
>   static <E> List<E>[] foo() { ... }
>   ...
>   List<?>[] list = foo();
> because as you said, Object will be inferred.
>
> Is there another case where diamond inference behave differently from 
> method inference ?
No - that's why the array case is banned; after going back and forth we 
decided to adopt the principle of least surprise, which means diamond 
always behave as a static generic method (the diamond spec is also 
design to map the diamond case back to the generic factory method). What 
I was saying is that in order for diamond to be safely used in array 
context we'd need to revisit both the spec and the compiler, and to 
rethink the diamond vs. generic method inference strategy. I think it 
will be much more productive if we could remove the ban about creating 
an array of List<Object> (reification) - which then would mean that the 
current inference strategy will apply to arrays as well.

Maurizio
>
>>
>> Maurizio
>
> Rémi
>