Binary patterns and Elixir.String.t

[mbuhot]

Hello, thanks for creating Gradualizer!
I tried it out on a simple module and ran into this error:

# foo.ex
defmodule Foo do
  @spec is_comment?(String.t()) :: boolean
  def is_comment?("#" <> _rest), do: true
  def is_comment?(_), do: false
end

iex(4)> :gradualizer.type_check_module(Foo)
The pattern <<"#",_rest@1/binary>> on line 3 doesn't have the type 'Elixir.String':t()
:nok

Looks like it hasn't resolved the type Elixir.String.t as a synonym for binary ?

[josefs]

The gradualizer currently doesn't know anything about Elixir. But it would be really nice to support it! Patches are welcome.
@tim2CF do you have any thoughts on this?

[tim2CF]

Hello!
I'll take a look on this week :)

[gomoripeti]

hello, I think this warning is caused by Gradualizer not yet supporting binaries (and binary patterns). If you replace String.t() with binary() you will still get a similar warning.

[zuiderkwast]

Hi! I have added two commits for binaries/bitstrings (70a42c2 subtype relation and 37399e4 type checking bit expressions and patterns). Perhaps this helps a bit? It is not really well-tested yet...

[josefs]

@mbuhot, I believe we've added the support needed for your program to go through Gradualizer. Can you confirm?

[mbuhot]

@josefs I still getting a type error, but now it is consistent whether the function is typed as String.t or binary.

iex(7)> :gradualizer.type_check_module(Foo)
The pattern "#" on line 0 doesn't have the type integer()
:nok

Looks like there is a comment in the tests (https://github.com/josefs/Gradualizer/blob/master/test/should_pass/bitstring.erl#L5-L9) indicating that binary patterns don't type check against binary type specs.

Feel free to close this issue if you'd rather track binary pattern support (independent of Elixir) separately.

[josefs]

@mbuhot, thanks for your quick feedback! The error message that you get now is rather puzzling to me but we don't have anyone working on Elixir support so I'm afraid it won't be investigated any time soon. So I'll close this ticket then.

The comment that you pointed to turned out to be outdated and I've updated the test case to reflect that.