Clojurians Log v2

any hints on what may cause this?

ClassCastException: clojure.lang.KeywordLookupSite$1 incompatible with clojure.lang.IFn
        clojure.spec.alpha/spec-impl/reify--1987 (alpha.clj:875)
        clojure.spec.alpha/conform (alpha.clj:150)
        clojure.spec.alpha/conform (alpha.clj:146)
        clojure.spec.alpha/macroexpand-check (alpha.clj:685)
        clojure.spec.alpha/macroexpand-check (alpha.clj:681)
        clojure.lang.Var.invoke (Var.java:385)
        cljs.analyzer/do-macroexpand-check (analyzer.cljc:3312)
        cljs.analyzer/do-macroexpand-check (analyzer.cljc:3306)
        cljs.analyzer/macroexpand-1* (analyzer.cljc:3326)
        cljs.analyzer/macroexpand-1* (analyzer.cljc:3314)
        cljs.analyzer/macroexpand-1 (analyzer.cljc:3375)
        cljs.analyzer/macroexpand-1 (analyzer.cljc:3371)
        cljs.analyzer/analyze-seq (analyzer.cljc:3408)
        cljs.analyzer/analyze-seq (analyzer.cljc:3388)

never seen this problem before and don't know where to start looking https://github.com/thheller/shadow-cljs/issues/443

@thheller Given that it is ClojureScript, you may also want to check in #cljs-dev channel

thx but I'm quite familiar with the CLJS side of things, this seems to be something on the CLJ side

new to me

but smells classloadery to me

like a fn compiled with IFn from a different classloader

nah

it's leaking an inner class of KLS

"it"

https://github.com/clojure/clojure/blob/master/src/jvm/clojure/lang/KeywordLookupSite.java#L42-L48

something is invoking this object as a function

oh, gotcha

a bad implementation of ILookupSite

or whichever of those interfaces you need to implement :)

I can never remember

this is the second time I've seen this in a couple weeks

either that, or there's something broken in the inline keyword lookup cache

I’ve seen an error like this “bad lookup site” before

same "impossible" situation with the innards of a KLS

discussed it inthis channel to

but that would be really surprising to me

with @ghadi contributing to that discussion before too I think - not sure what I had before and think we never found the answer

but don’t think it had anything to do with cljs and wasn’t using shadow I’m sure

@ghadi I mean it's not really an impossible situation, just impossible provided you're not handling KLS manually I think?

foudn it https://clojurians-log.clojureverse.org/clojure-dev/2019-01-10

you can access that thunk using fault

I wasn’t doing anything weird

a kw site was just failing magically

interesting

wasn’t trying to get down and manually call the callsite thunk or anything

^ that's all I need -- user wasn't doing anything weird

need a repro case

well, obviously you do, hah

could tooling be doing weird things?

just saying “hey, I’ve saw this and tried to discuss similar recently”

seriously thank you for chiming in - I couldn't remember the details

ah, was clara-rules and core.async in my case

so there were threads involved

and that’s all I ever knew

common tooling maybe? CIDER etc?

and we thought - perhaps the record type was being redef’ed so hitting the non-`identical?` branch of the callsite check

but that shouldn't leak

the bytecode the compiler emits should handle that case cleanly

it's very strange

not sure on tooling, it was actually a reported problem I was trying to diagnose. so yeah, not adding enough useful info here sadly.

I was surprised at it being able to leak at all though indeed.

I tried for a while at the repl to manipulate things enough to cause it, with like redefining things etc, had no success