[WIP] Erlang Intellisense

[kxlsx]

This PR implements Intellisense functionality for erlang cells.

Status

The main functionality has been implemented, but the code needs
to be cleaned up & tests need to be added.

At the moment there is a lot of shared code between
Intellisense.Erlang and Intellisense.Elixir. We've opted
for the time being to make the functions we need public,
but we could maybe move them to a separate, shared module.

TODO:

• Function signature preview
• Details preview
• Module member completion
• Atom name completion
• Variable completion
• Bitstring options completion
• Module attribure completion*
• Tests

* Builtin module attributes are currently hard-coded into Intellisense.Erlang.IdentifierMatcher, as we haven't yet found a way
to integrate them cleanly.

[CLAassistant]

All committers have signed the CLA.

[jonatanklosko]

Hey, thanks for the PR! I tried it locally and it is already looking great :)

I found some rough edges, so I will focus on the functionality for now:

1. When you trigger completions in an empty cell (Cmd + Space), currently there are no completions, but we should actually show all completions (variables, modules, atoms, keywords). The same applies to other "neutral" places like math:pow(CURSOR) and 1 + CURSOR.

2. Similarly, when you trigger completion in <<X:8, Y/CURSOR>>, we should show all modifiers, but nothing shows up. Completions are only available once there is at least one letter, for example Y/bCURSOR.

3. Given math:pow(Left, Right)., if you put cursor anywhere within the variable names and trigger signature hint (Cmd + Shift + Space), nothing shows up. The hint only works immediately after ( or ,. Perhaps after tokenization we should remove anything before the cursor until we run into the closest ( or ,.

4. When hovering over math and pow in math:pow, the hover documentation says :math and :math.pow(X, N) respectively, which is Elixir syntax. It should follow Erlang syntax, so math and math:pow(X, N).

5. When hovering over pow in math:pow, the spec also uses Elixir syntax:

   @spec pow(x, n) :: float() when x: number(), n: number()

   It should instead be this:

   -spec pow(X, N) -> float() when X :: number(), N :: number().

   ex_doc already does this, as you can see for math:pow/2.

   Both 4. and 5. are because you reuse details formatting from Elixir. I think it should be separate. Some helpers that are language-agnostic we can later move to a shared module.

I would also suggest adding tests next, to cover various edge cases, including the list above. This way you can more confidently refactor/extend the code.

[jonatanklosko]

• Builtin module attributes are currently hard-coded into Intellisense.Erlang.IdentifierMatcher, as we haven't yet found a way to integrate them cleanly.

That's totally fine. In Elixir there is Module.reserved_attributes/0, but I don't think Erlang exposes this kind of API. We actually added it to Elixir when we were implementing Elixir intellisense in Livebook :)

[josevalim]

When hovering over pow in math:pow, the spec also uses Elixir syntax:

Here is how ExDoc does it for Erlang: https://github.com/elixir-lang/ex_doc/blob/1b1fe51e479b9bcbf72802e7ef700ed3cedf53ac/lib/ex_doc/language/erlang.ex#L511-L523