https://clojurians.slack.com/archives/C03S1KBA2/p1516725912000001

@dnolen would you be up for a patch that unified ^ ?

making (defn foo ^boolean [] ..) work in cljs and soft deprecating (defn ^boolean foo [] ..)

well in Clojure that’s required for primitive type hints

I didn’t think needed in general for fn return

so I don’t know about deprecating, I guess the former thing should at least be made to correctly propagate type info

these have different meanings(whether you’re type-hinting the var or the return type of the function) and different behavior (var meta are eval’ed, the signature meta is not). Generally type-hinting the return type (before the args vector) is the recommended and preferred place in Clojure.

In the AST, for (defn ^boolean foo [] ..) we end up with both :ret-tag boolean and :tag boolean. Perhaps these could somehow be detangled so that (defn foo ^boolean [] ..) affects :ret-tag…

in cljs type hints on vars are not evaluated I believe

@alexmiller one difference here is that var meta is never eval’ed

it’s not possible

but point taken about the preferred way

we should reconcile

note that in clojure argvec type hints do not propagate to the var

(defn foo ^bar [] you won't get (:tag (meta #'foo

and what about multiple arities

also (defn foo ^boolean [] ...) in clj will throw anyway. we'd be matching style but not substance

@favila Interesting point about multiple arities. It reminds me that the patch in https://dev.clojure.org/jira/browse/CLJS-1997 doesn't handle that case.

does the truth_ elider understand type hints on specific arities?

if not, that's the real thing to support

and it can fall back to the var's tag metadata if missing

actually I guess there are in theory six possible things to check: arity tag, arity ret-tag, fn tag and ret-tag, and var tag and ret-tag.

not all of those exist presently though

Yeah, there is some interesting fallback logic here already: https://github.com/clojure/clojurescript/blob/master/src/main/clojure/cljs/analyzer.cljc#L1543-L1545