Symbolic values (##NaN, ##Inf) rendered as null when formatted

[aisamu]

It looks like the formatter is changing symbolic values to null during internal processing:

(pr-str ##NaN)
;; "##NaN"

(pr-str [##NaN])
;; "[##NaN]"

(. js/console log #js [##NaN ##Inf ##-Inf])
;; [NaN, Infinity, -Infinity]

(. js/console log [##NaN ##Inf ##-Inf])
;; [null null null]

This initial attempt seems to fix it - at least through inspection of the demo page with symbolic values added.

modified   src/lib/devtools/formatters/helpers.cljs
@@ -62,6 +62,11 @@
 (defn bool? [value]
   (or (true? value) (false? value)))
 
+(defn symbolic-value? [value]
+  (or (js/Number.isNaN value)
+      (= js/Number.POSITIVE_INFINITY value)
+      (= js/Number.NEGATIVE_INFINITY value)))
+
 (defn instance-of-a-well-known-type? [value]
   (let [well-known-types (pref :well-known-types)
         constructor-fn (get-constructor value)
@@ -76,7 +81,8 @@
 (defn directly-printable? [value]
   (or (string? value)
       (number? value)
-      (bool? value)))
+      (bool? value)
+      (symbolic-value? value)))

I had no success getting it to pass the whole test suite. Some issues were just phantomjs/chrome differences[1], but others might indicate that this approach might conflict with other internal mechanisms.

[1] - e.g. Phantom seems to lack the safer Number.isNaN, but has the old isNaN.

We're running devtools with shadow-cljs:

thheller/shadow-cljs: "2.8.68"
binaryage/devtools: "0.9.10"
clojurescript: "1.10.520"

[aisamu]

Welp, that was fast! Thanks!

[darwin]

btw. your code didn't really have effect on directly-printable? because number? returns true for NaN and Inf values. The probable reason why you saw it fixed in your experimentations is because you js/console.log them as unwrapped. Unfortunately custom formatters in devtools do not give me a chance to render simple values like numbers or strings. They are applied for objects only.

[aisamu]

Yes, that makes sense!
Thanks again for tackling it so promptly