Possible regression for bitstring completion

[katafrakt]

When I checkout to da065ae and run suggestion on bitstring, I get:

iex(1) > ElixirSense.suggestions("<<x::in", 1, 8) |> Enum.filter(& &1.type == :bitstring_option)
[%{name: "integer", type: :bitstring_option}]

However, the same on later commit from master does not return it:

iex(1) > ElixirSense.suggestions("<<x::in", 1, 8) |> Enum.filter(& &1.type == :bitstring_option)
[]

Possibly related, but not sure if bug or just change in behaviour:

Old:

iex(2) > ElixirSense.suggestions("<<x::ut", 1, 8) |> Enum.filter(& &1.type == :bitstring_option)
[
  %{name: "utf8", type: :bitstring_option},
  %{name: "utf16", type: :bitstring_option},
  %{name: "utf32", type: :bitstring_option}
]

New:

iex(2) > ElixirSense.suggestions("<<x::ut", 1, 8) |> Enum.filter(& &1.type == :bitstring_option)
[
  %{arity: 1, name: "unit", type: :bitstring_option},
  %{name: "utf8", type: :bitstring_option},
  %{name: "utf16", type: :bitstring_option},
  %{name: "utf32", type: :bitstring_option}
]

(additional entry for "unit")

This was found while updating ElixirSense in Expert.

[hiSandog]

A compact regression matrix could cover completion immediately after <<, after an existing segment, and inside modifiers such as ::binary-size(...). Comparing incomplete editor input with a syntactically complete bitstring would exercise parser recovery as well as normal parsing. Nested bitstrings would add a useful boundary for locating the active segment.

[lukaszsamson]

Thanks for reporting. The regression is likely a result of backporting elixir 1.15+ improvements to completion engine. It replaced a custom bitstring reducer with code relying on Code.Fragment API. i'll take a look