An interesting aspect of type checking a Clojure dialect is dealing with host interop.
In Typed Clojure, Java interop is handled by utilising Java type information. For the collected Java types to be sound in Typed Clojure, we then transform them to deal with corner cases in the Java type system like covariant arrays and null.
In checking Clojurescript, we need to type check interactions with Javascript. Clearly Javascript does not provide its own type annotations like Java.
Checking DOM interactions
The nature of gradual typing means we only need to annotate the interface between typed and untyped code. In practice, this means all Javascript calls found via type checking with Typed Clojurescript must be annotated.
Take this helper function by-id
, a thin wrapper around a Javascript method call.
(ann by-id [string -> (U nil js/HTMLElement)])
(defn by-id [id]
(.getElementById js/document id))
Typed Clojurescript can statically verify the expected annotation of by-id
if it knows
js/document
is of typejs/Document
- an instance of
js/Document
has a method of type[string -> (U nil js/HTMLElement)]
.
This information needs to be provided explicitly, and the current implementation of Typed Clojurescript provides a small base environment.
Future work
Clearly the amount of work to manually annotate the entirety of the DOM is not worth the effort. However, there are several sources of inspiration for automating this effort.
- TypeScript annotates the DOM and provides type checking on its usage.
- The DOM4 spec includes detailed type signatures for the DOM.
This work is an important deliverable for the Typed Clojure crowdfunding campaign. The longer I can work on Typed Clojure, the more viable Typed Clojurescript will become.
In Part 2, we look at why type checking Clojurescript is more satisfying than type checking Javascript