Salvage code actions from experimental server

[sheldak]

@lukaszsamson As we have discussed in #891, I recovered code actions with some helper functions that I found useful from the experimental server.

This PR introduces two code actions.

Replacing an unused variable with an underscore

e.g.

def foo(x), do: 42

x is a variable that is not used, so would get an action:

Rename to _x

that will replace the code with

def foo(_x), do: 42

Replacing unknown remote function

e.g.

def foo, do: Enum.counts([2, 3])

contains an unknown function: counts/1. After this change, we would get 2 possible actions:

Rename to concat
Rename to count

that rewrite our code to:

def foo, do: Enum.concat([2, 3])

and

def foo, do: Enum.count([2, 3])

respectively.

To find the best functions for replacement, I use String.jaro_distance/2 with threshold 0.77 and show at most 5 suggestions (similar to did you mean functionality).

[sheldak]

The following test of unused variables would not pass:

test "applied to a branch in a case" do
  {:ok, result} =
	~q[
	  case my_thing do
		{:ok, unused} -> :ok
		_ -> :error
	  end
	]
	|> modify(line: 1)

  expected = ~q[
	case my_thing do
	  {:ok, _unused} -> :ok
	  _ -> :error
	end
  ]

  assert result == expected
end

because ElixirSense.string_to_quoted(" {:ok, unused} -> :ok", {1, 1}) returns

{:ok, {:__atom_elixir_marker_1__, [closing: [line: 1, column: 27], line: 1, column: 2], []}}

that is not an AST of that line we can use to add underscore. It seems like ElixirSense issue. Now, in that case, we are just not producing any code action.

[lukaszsamson]

You cannot use ElixirSense here. Better stick with Code.string_to_quoted_with_comments as it preserves more info and do not attempt to fix errors

[sheldak]

I see. But it seems that Code.string_to_quoted_with_comments was added in Elixir 1.13 and ElixirLS supports 1.12. Will that be a problem?

[lukaszsamson]

Not really. For now we can limit this feature to 1.13+ and 1.12 goes out of support in few months anyway

[sheldak]

If the original line has one-line do, after conversion to AST and conversion back to string, the output line has do end. Here I made a workaround that preserves one-line do. It is possible because the ASTs of the cases differ - just Macro.to_string/1 creates the same line out of both.

[lukaszsamson]

Awesome @sheldak. I plan to play around with it after I'm done with v0.20 release

[lukaszsamson]

I rebased to master and merged it in 43b656d. Thanks

[sheldak]

I rebased to master and merged it in 43b656d. Thanks

@lukaszsamson That's great! But didn't you prefer to use Code.string_to_quoted_with_comments instead of ElixirSense? I'm actually working on that change

[lukaszsamson]

Yes, let’s bring that in following PRs. I didn’t want this one to be open too long