Add elixir 1.19.x compatibility

[Geekfish]

• Introduce elixir v1.19 in CI
• Deal with failing test (both default and required are now returned)
• Deal with type violation warnings

[edgurgel]

Thank you!

[edgurgel]

Given that the if checks for module.__info__(:struct) != nil how could this || [] be reached? 🤔

[Geekfish]

Hey @edgurgel !
That's a great question.

So, if we change nothing, the compiler gives us:

     warning: incompatible value given to for-comprehension:

         %{field: field} <- struct_info

     it has type:

         nil or list(%{default: if_set(term()), field: atom()})

     but expected a type that implements the Enumerable protocol.
     You either passed the wrong value or you must:

     1. convert the given value to an Enumerable explicitly
     2. implement the Enumerable protocol

     where "struct_info" was given the type:

         # type: nil or list(%{default: if_set(term()), field: atom()})
         # from: lib/mimic/module.ex:132:21
         struct_info = module.__info__(:struct)

     hint: the Enumerable protocol is implemented for the following types:

         dynamic(
           %Date.Range{} or %File.Stream{} or %GenEvent.Stream{} or %HashDict{} or %HashSet{} or
             %IO.Stream{} or %MapSet{} or %Range{} or %Stream{}
         ) or empty_list() or fun() or non_empty_list(term(), term()) or non_struct_map()

     typing violation found at:
     │
 137 │           for %{field: field} <- struct_info,
     │                               ~
     │
     └─ lib/mimic/module.ex:137:31: Mimic.Module.generate_mimic_struct/1

I assumed that the compiler can't know if module.__info__(:struct) is a pure function. Perhaps calling it the second time, could theoretically bear a different result of a different type.

After you commented here I thought "I'll just refactor it so that we only call module.__info__(:struct) once".
However this had some very weird results of its own, I ended up in a bit of a rabbit hole. I came up with a couple more implementations, that I would have expected to work, but they still raised warnings. As well as one that I wouldn't have expected to work, that didn't raise warnings. If you're interested, here's a list of commits with different versions and their outcome (in the commit message).

TL;DR: there is a mega-issue in the Elixir repo, relating to false-positive warnings. The work pending includes a couple of unsuccessful rewrites that I've tried (with case and guards).

👉 I have added a new commit with a version that calls module.__info__(:struct).
It's not the nicest way, but it perhaps a bit less confusing than others (that don't raise the warning). Let me know what you think!

[Geekfish]

Actually, I found an even simpler way:
Replaced the list comprehension with an Enum.map. No warnings, no hacky stuff.

[edgurgel]

Oh I see! Thanks for taking the time to explain 🙇

[edgurgel]

2.2.0 is out 🎉