Improve validation of intervals?

[SimonGustavsson]

Hello 👋

Creating a RecurrenceRule with an interval that is just large enough causes IllegalArgumentException.

It looks like this happens here where the interval is multiple with 7 which causes it to overflow to a negative number which the calendar rightly rejects.

I do think lib-recur is doing the right thing rejecting this (it's a silly interval after all), but the exception is very unexpected.

Could/should this be validated a little earlier on and surfaced in a more obvious way perhaps?

Repro steps:

    @Test
    public void itOverflows() throws InvalidRecurrenceRuleException {
        RecurrenceRule rule = new RecurrenceRule("FREQ=WEEKLY;INTERVAL=1000000000", RfcMode.RFC5545_STRICT);
        RecurrenceRuleIterator iterator =
             rule.iterator(1578406277000L, TimeZone.getTimeZone("America/New_York"));
    }

Exception:

java.lang.IllegalArgumentException: n must be >=0
	at org.dmfs.rfc5545.calendarmetrics.NoLeapMonthCalendarMetrics.nextDay(NoLeapMonthCalendarMetrics.java:226)
	at org.dmfs.rfc5545.recur.Freq$3.next(Freq.java:74)
	at org.dmfs.rfc5545.recur.FreqIterator.next(FreqIterator.java:98)
	at org.dmfs.rfc5545.recur.SanityFilter.next(SanityFilter.java:92)
	at org.dmfs.rfc5545.recur.RecurrenceRuleIterator.fetchNextInstance(RecurrenceRuleIterator.java:95)
	at org.dmfs.rfc5545.recur.RecurrenceRuleIterator.<init>(RecurrenceRuleIterator.java:89)
	at org.dmfs.rfc5545.recur.RecurrenceRule.iterator(RecurrenceRule.java:2216)
	at org.dmfs.rfc5545.recur.RecurrenceRule.iterator(RecurrenceRule.java:2140)

[dmfs]

I guess the problem goes a bit deeper. The underlying datetime lib stores the year in just 18 bits. There will be an overrun in year 262144. At present I don't consider that an issue. That might change around the year 260000, though. If this lib is still in use by then, people might be worried about this issue. ;-)

I guess it makes sense to validate the interval value and limit it to some reasonable value. Not sure what a reasonable limit would be though. It probably depends on the FREQ value. For starters we could limit the interval to 0<n<10000000. That's probably a reasonable range. For a SECONDLY rule that would be an interval of about 7.6 years.