@thheller I found a nasty issue with shadow + lein + cljs 1.10.312

I can't use test.check generators with this combined settings

there was some changes in dynamic var checking, not sure what exactly, but it keeps complaining that clojure.test.check.generators is not loaded, but it is

using just shadow without lein works fine

I just created a minimum repro, I'll open an issue with that

hmm lein basically only constructs the classpath. are you sure there aren't any classpath issues?

the repro is really simple, I can't see how that would be

unless I'm blind and missing something, can you take a look?

sure

just preparing the repro to upload, one min

@thheller https://github.com/wilkerlucio/bug-shadow-lein-cljs-312

here I get this with this code:

alpha.cljs:70 Uncaught Error: Var clojure.test.check.generators/return does not exist, clojure.test.check.generators never required
    at alpha.cljs:70
    at cljs.spec.gen.alpha.LazyVar.cljs$core$IDeref$_deref$arity$1 (alpha.cljs:22)
    at Object.cljs$core$_deref [as _deref] (core.cljs:673)
    at Object.cljs$core$deref [as deref] (core.cljs:1449)
    at Function.cljs$core$IFn$_invoke$arity$variadic (alpha.cljs:70)
    at alpha.cljs:91
    at cljs.core.Delay.cljs$core$IDeref$_deref$arity$1 (core.cljs:10369)
    at Object.cljs$core$_deref [as _deref] (core.cljs:673)
    at Object.cljs$core$deref [as deref] (core.cljs:1449)
    at Object.cljs$spec$gen$alpha$gen_for_pred [as gen_for_pred] (alpha.cljs:148)

I do get it as well with just shadow-cljs

(using master and 1.10.312)

strange, when I replace :lein true with:

:source-paths ["src"]

 :dependencies [[org.clojure/clojure "1.9.0"]
                [org.clojure/clojurescript "1.10.312"]
                [org.clojure/test.check "0.9.0"]
                [thheller/shadow-cljs "2.4.5"]]

it works here

clojure.test.check.generators.return$ exists so there is some renaming going on

you can't put org.clojure/clojurescript into shadow-cljs.edn

or rather its filtered out to always use the version shadow-cljs "prefers"

does CLJS not rename the return?

ah, gotcha, and how do you force shadow to use a different version?

I don't. master is already bumped to 1.10.312. just didn't make a new release yet

just did a lein install to use it

it seems like everything is correct to me?

yeah, but compiling with pure cljs main things, it works

so I guess there is something getting off

strange

wonder what cljs generates

this can be a problem to bump shadow, anyone using generators will be affected

when I was trying to debug in #clojurescript yesterday david said this: https://clojurians.slack.com/archives/C03S1L9DN/p1529073112000560

ring any bells?

no, so far it just looks like a renaming issue

return is a reserved word so it should be munged to return$

but for some reason this is not happening for the var generation stuff

need to figure out how to compile cljs without shadow-cljs

not used to this cljs.main business

I can send you a repo ready here

deps.edn

{:paths ["src"]
 :deps {org.clojure/clojurescript {:mvn/version "1.10.312"}
        org.clojure/test.check {:mvn/version "0.9.0"}}}

cljs.main -c my-app.core

got it

I tried here, seems the name is clojure.test.check.generators.return$

as you said

hmm yeah the code is actually looking for that so no idea why it doesn't find it

there is some weird code genration going on

and we though cljs would come with no surprises 😛

if((((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined')) && ((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined') && (typeof clojure.test.check !== 'undefined') && (typeof clojure.test.check.quick_check !== 'undefined')))){

((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined')) is tested twice

shadow-cljs generates this though

if((((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined')) && ((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined') && (typeof clojure.test.check !== 'undefined')) && ((typeof clojure !== 'undefined') && (typeof clojure.test !== 'undefined') && (typeof clojure.test.check !== 'undefined') && (typeof clojure.test.check.generators !== 'undefined') && (typeof clojure.test.check.generators["return"] !== 'undefined')))){

quadruple checked

clojure.test.check.generators["return"] this fails

not yet sure why shadow generates different code

the compiler hacks were gonna get me at some point 😉

just curious now, how do you find the compiled code? there is a way in chrome to go from source mapped to compiled source?

just open the file in the editor. just searched for the never required error and went from there

ok, more a brute force thing 🙂

ok the exists? thing is definitely weird and messes with how shadow-cljs handles js/stuff

ok got an idea to fix it

haha the cljs.spec.gen.alpha/dynaload macro does the work that the new exists? macro does. thats why the check is done twice

(and [email protected] (c/exists? ~s))

seems like it

meh if I fix it the simple way I create another bug again 😞

is that some shadow hack that is not going well with the new exists??

https://github.com/thheller/shadow-cljs/blob/master/src/main/shadow/build/cljs_hacks.cljc#L448-L468

basically the problem is that clojure.test.check.generators/return gets turned into clojure.test.check.generators.return$ but js/clojure.test.check.generators.return is turned into clojure.test.check.generators["return"]

and the new exists? macro turns all vars into js/

no idea why that is

js/ has special treatment in shadow-cljs because of the JS interop it does

JS code in npm doesn't munge .return or so but the CLJS compiler does when you try to access it

I was reading the commits, seems like it was to fix some self-hosting thing

so that might be way they keep touching on JS, since it's all self-hosting have access too

https://github.com/clojure/clojurescript/commit/9f8ad161eda1eb4b0fdb782e593ee9400b4e1f96

this wasn't self host related

(exists? this.ns.does.not.exist) would fail previously

the first impl just had to be fixed for self-host later

I'm just trying to make sense of WHY the js/ is added

(core/defmacro exists?
  "Return true if argument exists, analogous to usage of typeof operator
   in JavaScript."
  [x]
  (if (core/symbol? x)
    (core/let [x     (core/cond-> (:name (cljs.analyzer/resolve-var &env x))
                       (= "js" (namespace x)) name)
               segs  (string/split (core/str (string/replace (core/str x) "/" ".")) #"\.")
               n     (count segs)
               syms  (map
                       #(vary-meta (symbol "js" (string/join "." %))
                          assoc :cljs.analyzer/no-resolve true)
                       (reverse (take n (iterate butlast segs))))
               js    (string/join " && " (repeat n "(typeof ~{} !== 'undefined')"))]
      (bool-expr (concat (core/list 'js* js) syms)))
    `(some? ~x)))

this is the current impl

if I change it to

(core/defmacro exists?
  "Return true if argument exists, analogous to usage of typeof operator
   in JavaScript."
  [x]
  (if (core/symbol? x)
    (core/let [x     (core/cond-> (:name (cljs.analyzer/resolve-var &env x))
                       (= "js" (namespace x)) name)
               segs  (string/split (core/str (string/replace (core/str x) "/" ".")) #"\.")
               n     (count segs)
               syms  (map
                       #(vary-meta (symbol (string/join "." %))
                          assoc :cljs.analyzer/no-resolve true)
                       (reverse (take n (iterate butlast segs))))
               js    (string/join " && " (repeat n "(typeof ~{} !== 'undefined')"))]
      (bool-expr (concat (core/list 'js* js) syms)))
    `(some? ~x)))

it works just the same and fixes the shadow issue

literally just remove the "js" in the symbol call

I don't get why its there

js/ will also cause shadow to generate externs and I think CLJS as well

I don't get it 😛

I really need to learn cljs.main. someone should verify that the exists? calls not actually generate externs via externs inference. I think they might.

maybe we should bring that to #cljs-dev ?

nah I'd rather verify first

ah I get why the "js" is there now

whys that?

(exists? cljs.core/return) "(typeof cljs.user.cljs !== 'undefined') && (typeof cljs.core !== 'undefined') && (typeof cljs.core.return$ !== 'undefined')"

seems like the cljs.analyzer/no-resolve is ignored for some reason and js/ is just an easy way to force it to not resolve

meh this messes with my externs inference. will need to think about it

thanks for checking it out