Clojurians Log v2

oh gosh i didn't realize that the default namespace is created by listing out all variables that are copied https://github.com/babashka/sci/blob/56104c81cb7a3804f8ee84aa64b313dc7143f30e/src/sci/impl/namespaces.cljc#L1652

i suspect update-keys is an oversight? i can make a PR if you're interested

if it helps, here's a list of all the things that are bound in clojure.core but not sci's clojure.core:

user=> (clojure.set/difference (set (keys (ns-publics 'clojure.core))) (set (keys sci.impl.namespaces/clojure-core)))
#{primitives-classnames restart-agent ensure proxy-mappings ref-min-history gen-class println-str slurp import splitv-at pvalues bases read+string dosync stream-transduce! send-via random-uuid PrintWriter-on future-call resultset-seq struct *source-path* with-loading-context NaN? compile set-error-mode! set-agent-send-executor! remove-tap gen-interface *command-line-args* -cache-protocol-fn await spit future-done? -reset-methods partitionv load stream-into! *verbose-defrecords* stream-reduce! pmap error-mode StackTraceElement->vec alter with-precision shutdown-agents *use-context-classloader* await1 ref-set defstruct update-vals definline find-protocol-impl method-sig unquote-splicing future definterface add-tap pcalls *allow-unresolved-vars* vector-of set-error-handler! inst-ms* default-data-readers ->VecSeq Inst ref construct-proxy agent-errors *compile-files* *math-context* set-agent-send-off-executor! ref-history-count cast parse-double EMPTY-NODE future? agent-error future-cancelled? struct-map stream-seq! mix-collection-hash ->Vec file-seq agent set-validator! tap> sync *compile-path* release-pending-sends print-ctor clear-agent-errors proxy-super add-classpath iteration accessor infinite? partitionv-all error-handler proxy-name find-protocol-method find-keyword send-off load-file extenders ->ArrayChunk ->VecNode abs ref-max-history Throwable->map parse-uuid io! update-keys *agent* *repl* *fn-loader* parse-boolean send proxy-call-with-super *compiler-options* get-proxy-class update-proxy create-struct get-validator await-for parse-long init-proxy future-cancel commute}

hm, i tried to add these to src/sci/impl/namespaces.cljc but i kept getting errors like "Exception in thread "main" java.lang.RuntimeException: Unable to resolve symbol: update-keys in this context, compiling:(sci/impl/namespaces.cljc:1667:19)". i'm just going to post a list of the bindings i tried to add: • println-str • read+string (wasn't sure how to implement this so i just left a comment) • abs • infinite? • iteration • NaN? • parse-boolean • parse-long • parse-double • partitionv • partitionv-all • splitv-at • update-keys • update-vals • with-precision (wasn't sure how to implement this since it's a macro) i tried to be very conservative; i didn't include anything related to parallelism, transactions, agents, IO, UUIDs, or things that looks related to the state of the clojure interpreter; nor anything that wasn't documented.

hm also (/ 0.0 0.0) throws an exception in an SCI context instead of returning NaN, not sure if that's intentional

Please one thing at a time. The reason update-keys isn't there is that SCI also works with older versions of clojure

you can add it yourself in the 'clojure.core namespace per regular configuration. Also see babashka

The difference you are seeing with / is because of this differnce in Clojure:

user=> (/ 0.0 0.0)
##NaN
user=> (#'/ 0.0 0.0)
Execution error (ArithmeticException) at user/eval3 (REPL:1).
Divide by zero

here is a configuration for clojure 1.11: https://github.com/babashka/sci.configs/blob/main/src/sci/configs/clojure_1_11.cljc

🧵

If it's fixed, there might be a bit of pain that eventually goes away. If it's never fixed, there's already a different kind of pain that will never go away. If it's put behind a potentially obligatory feature flag, it becomes a "choose your own adventure" kind of pain. E.g. :dashes :to-underscore would convert a-b into a_b (as a bonus you can use :-2_ instead of :to-underscore if you feel like the latter is a bit too long 🙂), :dashes :as-is would preserve the current behavior, and :dashes :warn would be the new default option that just complains when it encounters (.-a-b x).

So what would be your favorite pain?

Definitely the last. Doesn't break any existing code, does bring attention to the matter when there's something to pay attention to. Warnings could be annoying to some, but the cost-benefit with the other options still checks out.

The :warn aspect of this I could include in clj-kondo, but it's still an optional linter since in CLJS only code it still causes false positives like

(deftype Foo []
  Object
  (foo-bar [_]))

(.foo-bar (->Foo))

(deftype Foo []
  Object
  (foo_bar [_]))

(.foo_bar (->Foo))

Hmm. deftype or defrecord are not something I have thought about. With that self-contained example, no warning is necessary at all - both versions are equally valid. But it seems that the first version could be problematic in some scenarios where JS interop is involved...

Yep and it’s not clear from context what thing you are calling interop on

Here is my response just as a data point. I don't have a strong opinion on this. My main usage is Scittle (not nbb) and I feel everything is still quite greenfield. So I could tolerate a breaking change like this. Not sure if I would run into it myself. I like the idea of having Scittle as close to cljs as possible for potential migrations (for performance) or code sharing for libraries.