Sometime DATE_DIFF_MILLIS calculate wrong result as follows
DATE_DIFF_MILLIS(1465634581881,1465634581880,‘millisecond’)
This result was 0. but obviously, it should be 1.
I found something wrong from source code.
https://github.com/couchbase/query/blob/master/expression/func_date.go,
343: da1 := date1.Actual().(float64)
344: da2 := date2.Actual().(float64)
345: pa := part.Actual().(string)
346: diff, err := dateDiff(millisToTime(da1), millisToTime(da2), pa)
In this case date1 is 1465634581881 and date2 is 1465634581880.
from source code, da1 is changed to float64 format
da1 => 1.465634581881e+12
da2 => 1.46563458188e+12
And then set to millisToTime function
1440: func millisToTime(millis float64) time.Time {
1441: return time.Unix(0, int64(millis*1000000.0))
1442: }
… change float64 to int64 and multiply 1000000.0 to change nano_sec from milli_sec.
int64(da11000000.0) => 1465634581880999936
int64(da21000000.0) => 1465634581880000000
dateDiff function is mainly calling diffDates function
diffDates main part is…
1758: setDate(&d1, t1)
1759: setDate(&d2, t2)
1760:
1761: diff.millisecond = d1.millisecond - d2.millisecond
setDate is…
1791: func setDate(d *date, t time.Time) {
1792: d.year = t.Year()
1793: d.doy = t.YearDay()
1794: d.hour, d.minute, d.second = t.Clock()
1795: d.millisecond = t.Nanosecond() / 1000000
1796: }
It’s divide Nanosecond by 1000000
time.Unix(0,int64(da11000000.0) ).Nanosecond()/1000000 => 880
time.Unix(0,int64(da21000000.0) ).Nanosecond()/1000000 => 880
It’s typical round-off error!!
Fix it!
