Question about elixirc's static code analysis

Questions / Help
,
1

Hi all,

I’m trying to get up to speed with elixir and have just stumbled across a compiler warning that I find a bit puzzling. I was wondering if someone here could explain to me why the compiler behaves like this. Consider the following code:

defmodule Foo do
  def f(), do: :foo
end

defmodule Bar do
  def f(), do: :bar
end

defmodule Test do
  def return_module_tuple_directly(name) do
    case name do
      "foo" -> {:ok, Foo}
      "bar" -> {:ok, Bar}
      _ -> {:error, :unknown_module}
    end
  end

  def return_module_tuple_with_module_variable(name) do
    module = case name do
      "foo" -> Foo
      "bar" -> Bar
      _ -> nil
    end
    if module != nil, do: {:ok, module}, else: {:error, :unknown_module}
  end

  def this_works_fine(name) do
    {:ok, module} = return_module_tuple_directly(name)
    module.f()
  end

  def this_fails(name) do
    {:ok, module} = return_module_tuple_with_module_variable(name)
    module.f()
  end

  def this_also_fails(name) do
    case return_module_tuple_with_module_variable(name) do
      {:ok, module} -> module.f()
      _ -> :error
    end
  end

  def but_this_works(name) do
    with {:ok, module} <- return_module_tuple_with_module_variable(name) do
      module.f()
    end
  end
end

If I compile that, I’m getting the following warnings:

    warning: nil.f/0 is undefined (module nil is not available or is yet to be defined)
    │
 34 │     module.f()
    │            ~
    │
    └─ modules.ex:34:12: Test.this_fails/1
    └─ modules.ex:39:31: Test.this_also_fails/1

What I don’t understand is:

  1. In this_fails/1, why does the compiler think the module could be nil? We’re explicitly testing if module != nil. So it should know that the module can never be nil if the first tuple element is :ok. I’m assuming it’s smart enough to “see” that the value could be nil from the case statement but then it’s not smart enough to also introspect the condition of if (module != nil)?

  2. Why does wrapping the matching of {:ok, module} in a with statement prevent this warning, but the similar variant with the case statement does not? This is what confuses me the most.

I’d really appreciate an explanation! Thanks!

2

This looks like a bug in the typesystem, though it’s a bit surprising that using with would make things work.

1 Like
3

FWIW,

$ elixirc --version
Erlang/OTP 28 [erts-16.0.1] [source] [64-bit] [smp:10:10] [ds:10:10:10] [async-threads:1] [jit] [dtrace]

Elixir 1.18.4 (compiled with Erlang/OTP 27)