Final variables without explicit type: Intersection types issue.

[Reinier Zwitserloot]

Intersection types are annoying, to say the least. Possibly hand-wave  
it away by stating that you can't use those expressions in this new  
language construct.

Or, a slight adaptation to that: If all but 1 of the intersection  
types are subtypes of the other one, and this can be said for only 1  
type, then that type wins. Thus:

final foo = someBoolean ? Arrays.asList("foo", "bar") : new  
ArrayList<String>();

foo's type would be List<String>, because that's the type of one of  
the intersections (List<String>), and all other types of that  
intersection (ArrayList<String>) are a subtype of this, and there's no  
other type for which this can be said. However, writing up the  
specifics of this sounds difficult to say the least, and it can always  
be added in java8 if it causes an inordinate amount of whining.

Either way, this:

final foo = someBoolean ? new LinkedList<String>() : new  
ArrayList<String>();

would be a compiler error. What should foo be? List<String>?  
AbstractList<String>? Serializable? Cloneable? Object? They're all  
common supertypes, and there's no clear winner in the set; attempting  
to use the nearest common parent still gives you 2 winning options:  
Serializable, and Cloneable, while in actual fact you were probably  
shooting for List<String>, which isn't even close to the winner here,  
what with AbstractList in the way, and AbstractList itself being 2  
removed from LinkedList, which extends AbstractSequentialList, which  
extends AbstractList.


I think it'll be okay to just compiler-error on those, because I  
expect the majority usage would be something akin to:

final foo = methodCall();

  --Reinier Zwitserloot



On Mar 21, 2009, at 22:46, Florian Weimer wrote:

> * John Rose:
>
>> On Mar 21, 2009, at 12:52 PM, Marek Kozieł wrote:
>>
>>> Allow final variables and final Fields (except blank final), to not
>>> having explicit Type.
>>
>> Yes.  Someone should work exactly this (and no more) into a separate
>> proposal, if it hasn't been done already.
>
> What should be the inferred type of an expression with an intersection
> type?  Is there an answer which is acceptable in the COIN context?
>