Extend Expression can leak underscore and other issues with underscore

[sorawee]

> {_: 1}
Well-formedness: Cannot use underscore as a field name at...
> {a: 1}.{_: 1}
{_: 1, a: 1}
> {a: 1}.{_: 1}._
1 # oh wow

Is this the way you intend it to be?

> (_.a(123))({a: _ + 456})
579

This shouldn't be right (please note the whitespace after dot)

> (_.    a     )({a: 1})
1

[jpolitz]

The first one is a missing well-formedness check, so that's a bug, thanks!

The second and third work the way I'd expect. The second is

(lam(some-name): some-name.a(123) end)({a: lam(some-other-name): some-other-name + 456 end})

The third is

(lam(some-name): some-name.a)({a: 1})

As for why spaces are allowed around the dot, a common syntactic layout for method chaining is:

o
  .method1(with, some, args)
  .method2(with, some, other, args)
  .method3(with, some, final, args)

and that requires spaces around the dot.

Whether or not the _ syntax is helping or not here is certainly a good question. But these examples (2 and 3) do what I expect based on the meaning of _.

[sorawee]

We currently allow _ as a column name of tables. This ought to be a well-formedness error, too.

[blerner]

Handling _ as a column name was fixed in 5de3dd8. The others aren't fixed yet. I think {a: 1}.{_: 1} and {a: 1}._ should be well-formedness errors on their own ("Underscore not allowed as a field name"), rather than attempting to curry over a field name -- names aren't first-class values. I'll have a commit for this in a few minutes.