[CS2] Allow chained access of a property of a function literal

[GeoffreyBooth]

Implements the proposed syntax discussed in #4576:

->
  a
.call(…)

This doesn’t “fix” #4576 per se, because the output desired by that issue’s OP was simply incorrect and was fixed in 1.12.6; but this allows an alternative way to chain after function literals without needing to indent (“fixing” #4576 in the sense that the issue was about avoiding Christmas trees while chaining).

@helixbass is this what you had in mind?

[helixbass]

@GeoffreyBooth looks good. The description is not quite right, this will change the compilation of the example from #4576 from breaking (current behavior) or compiling to something unintuitive (pre-1.12.6 behavior) to compiling in a way that's more consistent with other chaining and with the chaining-after-top-level-function-literal example you give

[GeoffreyBooth]

Fortunately the description doesn't need to be merged. 😄 @lydell, any notes?

[lydell]

Well, if it makes

Promise
.resolve()
.then ->
  yes
.then ok

compile with such a small change then it looks good to me.

[GeoffreyBooth]

Hmm, @lydell your example compiles now. I guess I should change the second test.

The first test is definitely good,

->
  eq @a, 3
.call a: 3

doesn’t currently compile. @helixbass is there another example you can think of that doesn’t compile now, that should after this?

[lydell]

@GeoffreyBooth I copied my example from the tests of this PR!

[GeoffreyBooth]

I just meant, that test is bad because it compiles fine without this PR. So it’s not testing the change that this PR makes. It might be a good test to keep anyway, as a general test of chaining, but I feel like I should have at least one more test specific to this PR’s change.

[GeoffreyBooth]

Thinking about this more, it occurs to me that the only properties or methods you could access off a function literal are the ones attached to Function itself—.call, .apply, .bind etc. Which is useful functionality to add, but maybe we don’t need separate tests for more than just .call.

I think I’ll merge this in, and if anyone suggests additional tests I’ll add them in a future PR.