with-redefs doesn't redef core fns

[cldwalker]

Hi. I'm using with-redefs for a testing library and came across the following bug:

version

Version which comes with bb 0.1.3

platform

Mac binary

problem

with-redefs is unable to redef a clojure.core fn

repro

;; Doesn't return Hi
=> (with-redefs [spit (constantly "Hi")] (spit "file" "yolo"))
nil
;; Define fn with same name as clojure.core fn
=>  (defn spit [& args] (prn args))
nil
;; Behaves correctly now that spit is defined in ns
=> (with-redefs [spit (constantly "Hi")] (spit "file" "yolo"))
"Hi"

expected behavior

I expected to be able to redef a clojure.core fn as clojure does

[borkdude]

I think this may be due to an optimization, but maybe that optimization isn't worth it. I'll take a look.

[borkdude]

Undoing the performance optimization (which is effectively the same as direct linking in Clojure):

(time (sci/eval-string "(dotimes [_ 1000000] (+ (/ 1 2 3)))"))

is about 10s with the optimization and 12s without it on the JVM. In CLJS 17 vs 19s. This is on a Macbook Air which doesn't have a super powerful processor, but it's noticeable.

In Clojure you see the same effect with direct linking:

user=> (alter-var-root #'*compiler-options* (constantly {:direct-linking true}))
{:direct-linking true}
user=> (with-redefs [assoc dissoc] (assoc {:a :b} :a :b))
{:a :b}

Maybe sci should support *compiler-options* so in babashka you can turn it off for tests and with-redefs will work the way you would expect for core vars. Although this isn't exactly the same as direct linking since it's only done for built-in vars.

[borkdude]

Never mind, I found another optimization which gets back the performance up to the level it was, while fixing this issue.

[borkdude]

Should be fixed now in babashka master.

[cldwalker]

Thanks!!