Ambigous definition of `Macro.underscore/1`

[up2jj]

Elixir and Erlang/OTP versions

1.16.2, OTP-26

Operating system

Mac

Current behavior

Macro.underscore/1 takes as argument either an atom or a string:

@spec underscore(module() | atom() | String.t()) :: String.t()
  def underscore(atom_or_string)

Looking at the function definition, atom must be associated with module:

def underscore(atom) when is_atom(atom) do
    "Elixir." <> rest = Atom.to_string(atom)
    underscore(rest)
  end

We cannot therefore pass nil, which is also an atom after all:

iex(1)> Macro.underscore(nil)
** (MatchError) no match of right hand side value: "nil"
    (elixir 1.13.0) lib/macro.ex:1840: Macro.underscore/1

Should atoms that are not associated with modules be supported somehow by Macro.underscore/1?

Expected behavior

To be discussed

[josevalim]

A module is just an atom as well, :foo is a valid after all. So the spec is redundant, we need to fix it as well as the docs.