# Fix interval %/% and %% errors, NA in setdiff(), century leap years

[sims1253]

Report the operator the user wrote for %/% and %% on an Interval

%/% for (Duration, Interval) and (Period, Interval) in R/ops-integer-division.r dispatches to trunc_divide(), which computes trunc(e1 / e2). / for those signatures is a stop-only method in R/ops-division.r, so the error blames an operator the user never wrote:

int <- interval(ymd("2020-01-01"), ymd("2020-02-15"))
ddays(10) %/% int
#> Error: Incompatible classes: <Duration> / <Interval>

%% for (Period, Interval) in R/ops-modulo.r goes through modulo_period_by_period(), which calls %/% and lands on the same message:

months(3) %% int
#> Error: Incompatible classes: <Period> / <Interval>

Treat NA entries as not discontinuous in setdiff()

setdiff.Interval() in R/intervals.r computes makes2 <- !aligned & inside and then runs if (sum(makes2)). Any NA in x or y turns inside into NA, and the if() fails:

x <- interval(ymd(c(NA, "2020-02-01")), ymd(c(NA, "2020-04-10")))
y <- interval(ymd("2021-01-01"), ymd("2021-02-01"))
setdiff(x, y)
#> Error: argument is not interpretable as logical

Use the calendar year for February lengths in .int_to_period()

.int_to_period() in R/coercion.r converts end to POSIXlt and passes end$year to .days_in_month() at two sites. POSIXlt stores years since 1900, and .days_in_month() (R/accessors-month.r) calls leap_year() on an absolute year, so February gets the wrong length in years where the two disagree. For 2000 it receives 100: leap_year(100) is FALSE, leap_year(2000) is TRUE. For 1900 it receives 0, which the rule counts as a leap year while 1900 is not.

as.period(interval(ymd("2000-01-20"), ymd("2000-03-01")))
#> [1] "1m 9d 0H 0M 0S"    # February 2000 has 29 days; should be 10d
as.period(interval(ymd("1900-01-20"), ymd("1900-03-01")))
#> [1] "1m 10d 0H 0M 0S"   # February 1900 has 28 days; should be 9d

Found during my most recent ry audit.