Typespecing a map of arbitrary integer keys?

I am wondering if this is correct, this should be a map of arbitrary size (could be 0, could be 1000) of integer keys.

@type quests() :: %{Integer => any()}
defmodule GameState do
    @type obj() :: %{0=> Integer, optional(1)=> Integer, optional(2)=> Integer}
    @type quest() :: %{id: Integer, step: Integer, obj: obj()}
    @type quests() :: %{Integer => quest()}

    @type gamestate() :: %{quests: quests()}

    @spec get_quests(gamestate()) :: quests()
    def get_quests(gs) do
        :erlang.map_get(:quests, gs)
    end

    @spec act(gamestate()) :: any()
    def act(gamestate) do
        quests = get_quests(gamestate)

        cur_quest_id = quests
        |> Map.keys()
        |> List.last()
        cur_quest = Map.fetch!(quests, cur_quest_id)

        cur_objs = Table.QuestStep.by_id_order(cur_quest.id, cur_quest.step)
        done_objs = Enum.filter(cur_objs, fn{idx,obj}->
            cur_quest.obj[idx].amount >= obj.amount
        end)
    end

    def go do
        quest = %{id: 10010, step: 2, obj: %{0=> 14}}
        quests = %{10010=> quest}
        gamestate = %{quests: quests}
        act(gamestate)
    end
end

The idea here is to trigger a error from dialyzer here cur_quest.obj[idx].amount that .amount is not a valid key for obj().

Instead I get this, but if I call this from the REPL it works just fine. The function call succeeds.

lib/game/gamestate.ex:28:call
The function call will not succeed.

Table.QuestStep.by_id_order(Integer, Integer)

will never return since it differs in arguments with
positions 1st and 2nd from the success typing arguments:

(pos_integer(), 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9)

Can anyone advise please?

Integer mean that you are matching against :"Elixir.Integer" atom, you wanted to use integer() type.

1 Like

you probably want:

@type quests :: %{optional(integer) => any}

or maybe

@type quests :: %{optional(non_neg_integer) => any}

I prefer omitting parens on non parametric types, to cut down on line noise, but if you prefer keeping them in (which I believe is idiomatic?):

@type quests() :: %{optional(non_neg_integer()) => any()}
1 Like

Yikes I need to fix that message.

1 Like

Awesome, it doesn’t have any errors now, which is troubling, it should error here cur_quest.obj[idx].amount that amount is not a field.

It seems the Access module messes up so many things, if I replace the Access[] with a Map.fetch!() dialyzer picks it up and errors that amount is invalid and can never match the type. Guessing maybe because the implied nil?

Any ideas how to not lose the flexibly of the access module but keep dialyzer happy?

@vans163 mind checking out the better-error-message branch on Dialyxir and seeing if that reads any better?

Heh dialzyer with typespecs is not what we were looking for. So not sure if I will continue looking into this.

image

Basically because of that lol.

Waiting now to see if Facebook reveals anything interesting on Nov, if they made a totally new static typed language that compiles to beam that seems useless. If they build a new analysis tool or IL to spot type errors in beam files/core erlang that is more interesting now.

If nothing interesting from FB will probably take an attempt at walking compiled beam bytecode.