Compiler reports deadlock for %__MODULE__{} instead of struct undefined error

[ericentin]

# test.ex
defmodule Deadlock do
  def deadlock do
    %__MODULE__{}
  end
end

╰─$ elixirc test.ex

== Compilation error on file test.ex ==
** (CompileError)  deadlocked waiting on module Deadlock

Compilation failed because of a deadlock between files.
The following files depended on the following modules:

  test.ex => Deadlock

in IEx, where the parallel compiler is unused:

iex(1)> defmodule Deadlock do
...(1)>   def deadlock do
...(1)>     %__MODULE__{}
...(1)>   end
...(1)> end
** (CompileError) iex:3: Deadlock.__struct__/1 is undefined, cannot expand struct Deadlock

So it seems like, in order to provide an accurate error, this is maybe a special case that needs to/should be handled by the parallel compiler?

[josevalim]

Could this have been caused by b98b4bd?

[josevalim]

Actually, I misunderstood the issue. It is mostly a matter of knowing the struct won't be defined if I am trying to access the module itself. :) I don't think it was necessarily caused by above (and if it was, it doesn't matter much).

[ericentin]

Working on this, let me know if anyone has an objection.

[michalmuskala]

Is this related?

defmodule Test do
  require Test
end

== Compilation error on file test.exs ==
** (CompileError)  deadlocked waiting on module Test
    (elixir) src/elixir_aliases.erl:80: :elixir_aliases.ensure_loaded/3

Compilation failed because of a deadlock between files.
The following files depended on the following modules:

  test.exs => Test

[ericentin]

I would argue that message is at least accurate, though it of course could be improved to say what the problem is more directly. I'm not sure it would be worth it, though.

[josevalim]

@antipax one option is to discard the current patch and instead change the parallel compiler to raise a better error when I am waiting on myself. This would solve both issues.

[whatyouhide]

Actually I am 👍 on @josevalim's suggestion, it's a bit like when we say "function undefined or private", so that we don't have to do too many stuff an can just give an error message that explains possible causes.

[ericentin]

Doesn't the current message already say that it is waiting on itself? I don't think that makes the error more clear in the struct situation.

[ericentin]

To be clear, I think we should do both. The error message for deadlocked on self could be improved. The error message for an undefined local struct should say undefined struct, not deadlocked on self, IMO, because the fact that it produces a deadlock at all instead of a undefined error is due to an implementation detail (since structs can be waited on in the parallel compiler).

[ericentin]

So, has anyone come to any conclusions on this? Happy to close this out whatever way you all want to.