I think the idea is that there’s a lot of mistakes that can be caught/prevented by the type checker without any code changes or type annotations, but at a certain point, it’s going to be on us to clarify our intentions. There’s plenty of places where it makes total sense to compare numbers to atoms or atoms to tuples.
matt-savvy
matt-savvy
Also, I think this is a place where the type checking is still limited to your project and not its dependencies.
A basic example gets you a warning like you would expect:
def run(x) do
res = result(x)
if res > 5 do
IO.puts("hello")
end
x
end
defp result(x) when x < 10, do: {:ok, x + 1}
defp result(_), do: :error
Compiling 1 file (.ex)
warning: comparison between distinct types found:
res > 5
given types:
dynamic(:error or {:ok, float() or integer()}) > integer()
where "res" was given the type:
# type: dynamic(:error or {:ok, float() or integer()})
# from: lib/some_app.ex:3:9
res = result(x)
While Elixir can compare across all types, you are comparing across types which are always disjoint, and the result is either always true or always false
type warning found at:
│
5 │ if res > 5 do
│ ~
│
└─ lib/some_app.ex:5:12: SomeApp.run/1
With your dependency :
def run(x) do
Cachex.start_link(:my_cache)
{:ok, true} = Cachex.put(:my_cache, "my_key", 5)
res = Cachex.incr(:my_cache, "my_key")
if res > 5 do
IO.puts("larger than three")
end
x
end
No warnings
0
dimitarvp
matt-savvy
Being able to use normal sort operations on lists of elements of different types, things like.
iex(6)> results = [{:ok, 2}, :error, {:ok, 1}, :error, {:ok, 3}]
[{:ok, 2}, :error, {:ok, 1}, :error, {:ok, 3}]
iex(7)> Enum.sort(results)
[:error, :error, {:ok, 1}, {:ok, 2}, {:ok, 3}]
1
tjchambers
@matt-savvy thanks for your time and comments - much appreciated.
Based on what I have seen so far it seems as though this may be a hole in the current type checking - especially given the “disjoint” comment in one of the warnings in your example. I say that with the highest regard for the awesome work on the type checking to date. Kudos to everyone on this effort.
0
dimitarvp
Ah, that’s what you mean. Because if not for that use-case then I’d be genuinely puzzled why would people want to compare tuples with integers.
0