feat: namespace directive

[mitchellwrosen]

Overview

This PR implements the "namespace directive" feature: a Unison file can contain an optional namespace foo line that affects parsing as follows:

1. All bound names (type names, constructor names, generated accessor names, term names, and watch expression names) are prefixed with the namespace.
2. Variables that refer to locally-bound names exactly are prefixed with the namespace.

That is,

namespace a.b.c

type Foo = Bar
type Qux = { quaf : Foo }

baz = ...

honk = ... baz ...

is equivalent to

type a.b.c.Foo = Bar
type a.b.c.Qux = { quaf : a.b.c.Foo }

a.b.c.baz = ...

a.b.c.honk = ... a.b.c.baz ...

Note that the existence of a namespace directive therefore prevents one from referring to a name outside of the file whose suffix matches something in the file.

For example, without a namespace directive, I could write

foo.factorial = ... foo.factorial ... factorial ...

and refer to both my locally-bound foo.factorial as well as a term literally named factorial (no other prefix) in the namespace. (N.B that's not true in trunk today, but it will be fixed soon: #5268).

However, with a namespace directive...

namespace foo

factorial = ???

there's no way to write the same function (because factorial resolves to the locally-bound factorial, which will be expanded to foo.factorial during parsing).

For this reason, it would be unsafe to implement edit.namespace foo naively by putting a namespace foo directive at the top of the block.

Test coverage

I've added a transcript to demonstrate the feature.

[pchiusano]

Awesome!! That was fast.

I'm not sure why the transcript error message regression is happening. Though it might not matter too much if the only behavior is a worse error message in the event you call a definition namespace. Still, would be nice to keep the good message.

[hojberg]

Very cool! Can you have multiple namespace directives in a file?

[mitchellwrosen]

@hojberg You cannot – zero or one

[mitchellwrosen]

@pchiusano I looked into this briefly, it really isn't more complicated than we're now parsing namespace as a Reserved lexeme rather than an Identifier lexeme, and go down different error paths from there. This could/should be cleaned up for sure, but I'm inclined to leave it alone for now as it also affects other reserved keywords in the same way.

[pchiusano]

@mitchellwrosen cool. Yeah I wouldn’t sweat it for this PR then.