`t/micros` returns wrong result for negative durations

[yuhan0]

Observed behaviour

(require '[tick.core :as t])

(t/of-seconds -1)            ;=> #time/duration "PT-1S"
(t/nanos  (t/of-seconds -1)) ;=> -1000000000   ✓
(t/millis (t/of-seconds -1)) ;=> -1000          ✓
(t/micros (t/of-seconds -1)) ;=> 18446744072709551  ✗

(t/micros (t/of-micros -42)) ;=> 18446744073709509  ✗

Note: The result 18446744072709551 is -1e9 reinterpreted as an unsigned 64-bit integer and then divided.

Root cause

micros is implemented as:

(micros [d] (#?(:clj Long/divideUnsigned :cljs cljs.core//) (p/nanos d) 1000))

This bug has been present from its initial implementation in a5f1703

It's unclear why Long/divideUnsigned was used in the first place - presumably to handle positive durations whose nanosecond count exceeds Long/MAX_VALUE (~ 292 years)? But this silently corrupts any negative duration(which are legitimate and supported by the spec) , producing a large positive number with no indication of error.

Suggested fix

(micros [d]
  (clojure.core/+ (* (cljc.java-time.duration/get-seconds d) 1000000)
                  (quot (cljc.java-time.duration/get-nano d) 1000)))

Incidentally this also fixes another bug in the current impl, where the cljs branch returns a non-truncated value:

(t/micros (t/of-nanos 1234)) ;=> 1     (JVM Clojure) 
                             ;=> 1.234 (ClojureScript) 

[henryw374]

Thanks for this. It is clearly a bug. PR welcome