Infinite types are probably the trickiest kind of bugs to track down. Usually it means you are working on a relatively sophisticated recursive function and something is very subtly off.
I am just going to show a couple examples of infinite types that have been reported and how they were resolved to illustrate the kinds of things you may be running into.
(From issue #987)
Someone was working with records, and wanted to do something kind of tricky with type aliases. In this case, they created a Persisted
type alias that would say "whatever record you have, well I'm talking about one where there is an additional id
field."
type alias Persisted record =
{ record | id : Int }
type alias Update record =
record -> Int -> record
type Updatable record =
Updatable (Update record) (Persisted record)
newRecord (Updatable myUpdate myRecord) =
let
updateThisRecord =
myUpdate myRecord
in
"foo"
Now just looking at the code, it seems fine. You can create Persisted
records, you can have a record Update
, and you can bundle them together with Updatable
. In our newRecord
function we are partially applying the myUpdate
function. Everything seems fine, but something is causing an infinite type.
The best thing to do is start writing down as many types for yourself as possible:
myUpdate : Update record -- record -> Int -> record
myRecord : Persisted record -- { record | id : Int }
The trouble is that it's the same record
in both of these types! The myUpdate
function wants t take a record
and produce a record
, but it is being given { record | id : Int }
. Type checking is going to say, "well, if record
is actually an infinite sequence of { { ... | id : Int } | id : Int }
then they can be equal!" While that is true and very clever, it is not the intent of this code.
So the fix for this person was to add one thing into the Updatable
type:
type Updatable record =
Updatable (Update (Persisted record)) (Persisted record)
Notice the new Persisted
in the Update
part. That was the fix!
(From issue #707)
There is a relatively obscure package called evancz/automaton
that captures state and functionality using callbacks. So far, it does not seem to be extremely useful idea given its downsides, but it did lead to quite a tough infinite type.
Here is a snippet of code that combines two bacic automatons into one. The resulting automaton takes in either a Left
or Right
value. If it is a Left
we will use the left automaton, if it is a Right
we will use the right one.
type Either a b = Left a | Right b
type Automaton a b =
Step (a -> (Automaton a b, b))
(+++) : Automaton a b -> Automaton c d -> Automaton (Either a c) (Either b d)
(+++) (Step leftStep) (Step rightStep) =
Step <| \either ->
case either of
Left a ->
let (newLeft, b) = leftStep a
in (newLeft, Left b)
Right c ->
let (newRight, d) = rightStep c
in (newRight, Right d)
This triggers a relatively confusing error message. The right thing to do in these scenarios is try to figure out the type of literally every piece of the definition. If you can, break parts out into other top-level declarations with type annotations so the compiler can check your work. So you might write down some notes to yourself like this:
left : Automaton a b
leftStep : a -> (Automaton a b, b)
right : Automaton c d
rightStep : c -> (Automaton c d, d)
either : Either a c
newLeft : Automaton a b
newRight : Automaton c d
Knowing that both branches of a case
must end with the same type, it looks like newLeft
and newRight
are causing the issue here. So in the revised code we make it so both branches return the same type.
type Either a b = Left a | Right b
type Automaton a b =
Step (a -> (Automaton a b, b))
(+++) : Automaton a b -> Automaton c d -> Automaton (Either a c) (Either b d)
(+++) (Step leftStep as left) (Step rightStep as right) =
Step <| \either ->
case either of
Left a ->
let (newLeft, b) = leftStep a
in (newLeft +++ right, Left b)
Right c ->
let (newRight, d) = rightStep c
in (left +++ newRight, Right d)
So yeah, if you are skimming this, you probably did not see the change immediately. My experience debugging these kinds of things is that my brain thinks what I wrote is correct so much that it becomes kind of blind to the issue. The best thing to do is slow down and break the problem apart. Or ask someone.