While fixing something in clj-kondo, I discovered that in .cljs code Exception and String are not found as an unresolved symbol (due to a bug in kondo). I fixed that, but I'm left with some examples like:

(defn foo ^String [x] x)

(defn wrap-baz [^java.io.File x] (.getName x))

java.io.File

This happens mostly in .cljc files , so maybe since CLJS is forgiving, the linter should not complain about unknown types in type hints either as far as CLJS is concerned?

IIRC if the resolve fails its just treated as unknown and ignored

And there is a deliberate choice of not reporting, like e.g. in

cljs.user=> (def x String)
                   ^
WARNING: Use of undeclared Var cljs.user/String at line 1

resolving tags is different than resolving actual code, since JS is so dynamic knowing the actual class is not really important since there is no such thing as reflection

but users making typos could result in wrong inference?

so in CLJ it is required and used. in CLJS it is optional. any none js tag has no impact on inference

I see :)

there is also a ^clj tag right?

not sure if this is an official tag. don't think so, meaning it doesn't have special meaning besides making externs inference warnings shut up

right. I thought I'd seen it in the wild but then that may be the de facto way of disabling those warnings - thanks

I might be wrong on this though. there is clj-nil for sure with meaning in the analyzer, just not sure about clj. I just use it to shut up externs inference

I thought I'd also seen ^number somewhere

numeric IIRC

https://github.com/clojure/clojurescript/blob/9359f40e7537307b9272fd322ae5cfff88e079f2/src/main/cljs/cljs/math.cljs#L80

^not-native also exists and has special meaning

ok its number then

The not-native type hint works in CLJS in clj-kondo because it's also something in CLJS core

cljs.user=> (doc not-native)
-------------------------
cljs.core/not-native
  nil

number and js are specially supported

I'll just make clj-kondo more relaxed about these type hints in CLJS so at least you don't have to branch out in .cljc files to "undo" JVM type hints

which was already the case, but now as a feature instead of a bug ;)

but hmm, adding JVM type hints in a .cljc file may have the undesirable effect of accidentally adding externs by inference in .cljs right?

not that anyone will probably have ever noticed this

I'd say thats unlikely given that only js hints add externs

ah right

summary of the above sounds right to me