Round 1 feedback

[Brian Goetz]

> Going back to this one, this doesn't work as computeIfAbsent requires
> a Function whose first argument is the key :-(

Ah, right; you want

   k -> new ArrayList<>()

instead.

> Is there any chance such conversions with method references are
> possible, i.e., when their arity is less than the functional interface
> method arity?

This seems harmless, but would hurt you worse in the end.  The more 
automatic adaptation we try and do here, the harder it is to do overload 
resolution correctly.

For example, if you have a function:

   foo(Supplier<List<T>>)

you can call it with

   foo(TreeSet::new)

and the compiler will eliminate all the other constructor candidates 
because it knows the target type is () -> List<T>.  (Remember, it 
doens't know which TreeSet constructor you are asking for, and there are 
more than one, so it has to do overload selection.)  If we allowed fuzzy 
matching, then its ability to prune the "obviously wrong" choices 
necessarily goes down.

So rather than try to make method ref matching magic, we just use a 
lambda, which allows you more control.