Clojurians Log v2

indeed it is not valid to use ^integer? as a type hint. However, if I re-evaluate the same piece of code, the error is not shown a second time.

Here is the complete macro expansion. rte-case/clauses-to-dfa is a normal function, not a macro.

(def f
  (let [dfa__51157__auto__ (rte-case/clauses-to-dfa
                            '([0
                               (:cat
                                (rte
                                 (:cat
                                  (and :sigma :sigma)
                                  (and :sigma :sigma)))
                                (and :sigma :sigma)
                                (and :sigma :sigma))]
                              [1
                               (:cat
                                (and :sigma :sigma)
                                (rte
                                 (:cat
                                  (and :sigma :sigma)
                                  (and :sigma :sigma)))
                                (and :sigma :sigma))]
                              [2
                               (:cat
                                (and :sigma :sigma)
                                (and :sigma :sigma)
                                (rte
                                 (:cat
                                  (and Ratio :sigma)
                                  (and :sigma :sigma))))]))]
    (fn [& seq__51158__auto__]
      (let [fns__51159__auto__ [(fn conv-1 [[^integer? a ^integer? b] c d] 12)
                                (fn conv-1 [a [b c] ^integer? d] 13)
                                (fn conv-1 [a b [^Ratio c d]] 14)]
            ind__51160__auto__ (rte/match
                                dfa__51157__auto__
                                seq__51158__auto__
                                :promise-disjoint
                                true)]
        (if ind__51160__auto__
          (apply
           (fns__51159__auto__ ind__51160__auto__)
           seq__51158__auto__)
          (throw
           (ex-info
            "No pattern matching given list"
            {:sequence seq__51158__auto__})))))))

uhhhh. now it gives an error every time. something really bizarre is happening.

OK, i’ll keep trying to reproduce the issue. but it’s really strange.

https://chatgpt.com/share/680b3737-c014-8007-8b8a-bf3daeb5219a

@alexyakushev, yes I already understand why the error occurs. The thing I don’t understand is why the same situation fails to trigger an error other times. I know how to fix the error when it gets reported. However, if the error fails to be reported, the user of my library will not realize that there is an error.

This is related https://clojurians.slack.com/archives/C03S1KBA2/p1745523078201339

That’s a bit strange (long (foo 1)) is that really a type hint or a type conversion?

Conversion

indeed, not a type hint at all.

however, the decompiled code is a bit surprising to me. I was under the (apparently false) impression that the type hint instructed the java compiler to create a seperate code path for the case that the argument is of type Long, which would theoretically allow the JVM to do optimizations specific to that type, but there would always be a default code path in case the run-time object is not of type Long

(fn [a ^integer? b [c d]] nil) macroexpands to (fn* ([a ^integer? b p__59939] (let [[c d] p__59939] nil))) but (fn [a [^integer? c d]] nil) macroexpands to (fn* ([a p__59942] (let [[^integer? c d] p__59942] nil))) which further macroexpands to

(fn* ([a p__59942] (let*
                       [vec__59945
                        p__59942
                        ^integer? c
                        (nth vec__59945 0 nil)
                        d
                        (nth vec__59945 1 nil)]
                       nil)))

Ask this in #clojure as this is language level. I’m sure you will get some help there

Type hints that are not used by the compiler are just ignored. I'm not sure why it's used in your first case, maybe to determine if the function can used primitive argument ?