Nil lifting for method calls

[jclark]

We need to make nil lifting for method calls work well with the desugaring of method calls for non-objects into functions described in #111.

[jclark]

Also relates to #85, which has to do with nil-lifting generally.

[jclark]

Suppose we have an object type OT that provides a method foo. Given an expression E of type OT|(), and an expression call E.foo(), there may or may not be a method value:foo that accepts an argument of type (). The fundamental problem is that there are two plausible choices for what E.foo() should do when E results in ():

• call value:foo( () ), or
• nil lift i.e. return ()

Every approach has its downside:

1. Never do nil lifting. So it would be an error if value:foo( () ) is not a valid function cal. This would mean we lose the convenience of nil lifting.
2. Always do nil lifting. This would result in inconsistent behaviour for expressions of type OT? and any.
3. Do nil lifting conditionally: i.e. call value:foo( () ) is that is legal, otherwise nil lift. This means that if we add new functions to lang.value which accept a () argument but do not return () with a () argument, we will break code.

[jclark]

Conclusion today was not to do nil-lifting for either method calls or object field access.

[jclark]

This corresponds to how the spec is currently written for method calls.