Changes for version 0.10 - 2026-11-07
- NormaliseEvent (and so CompareEvents) keeps what it does not know instead of deleting it. It used to delete every key missing from its table, so every recurrenceOverrides patch normalised to {} ({"excluded":true} compared equal to an RDATE), every alert trigger to {} (any two alert offsets compared equal), and every @type was stripped (an Event compared equal to a Task). Now:
- unknown keys are kept, at every level
- override patches keep their contents; a patch value at a known object path (participants/p1, alerts/a1/trigger, locations, ...) is normalised as that object, and a patched value is never dropped for being the default, since in a patch it overrides the master
- an OffsetTrigger drops relativeTo "start" and an implied @type; an AbsoluteTrigger or unknown trigger is kept whole
- @type is dropped only where the position implies it (Location, Participant, Alert, NDay, ...); the Event or Task keeps it
- defaults follow jscalendarbis-22: status "confirmed", duration "PT0S", descriptionContentType "text/plain", interval 1, and scheduleAgent "server" (RFC 8984); participant kind has no default, so "individual" no longer equals an unset kind
- a JSON false is now dropped where false is the default (showWithoutTime, expectReply); it used to be kept because JSON booleans are references
- a null map or array is dropped as before, but a malformed value (a map that is not an object) is kept rather than dropped
- t/icalfiles.t marks two round trips TODO that the old NormaliseEvent hid (override patches lost on the way back)
- add t/normalise.t
- add t/roundtrip.t: JSCalendar -> iCalendar -> JSCalendar is the identity for hand-written Events and Tasks using every converted property (testdata/roundtrip/*.json), and iCalendar -> JSCalendar -> iCalendar -> JSCalendar is stable for every .ics under testdata, including the jscalendar-icalendar-29 examples (tools/extract-icalendar-examples.pl). Known conversion bugs are TODO tests, each with its cause
- MANIFEST lists the tests and test data added since 0.08
- one bad component no longer costs the whole calendar. The reader converts each VEVENT/VTODO on its own, and each VALARM and RRULE in it: a component that cannot be converted is dropped (an RRULE with FREQ=FORTNIGHTLY used to confess, a VALARM without ACTION to die on lc undef, each taking every event with it). The writer likewise writes each object, and each alert, on its own. What was dropped is reported by the new errors() method (uid, component, id, message), reset by each vcalendarToEvents, eventsToVCalendar or _argsToVCalendar call; their return values are unchanged
- add t/isolation.t
- no process-wide side effects:
- loading the module no longer replaces $SIG{__WARN__} for the whole process; the Data::ICal "Unknown property" carps are caught only while the module itself is writing
- the module uses plain warnings, not FATAL => 'all' (a warning in it, such as an alert of an untitled event, killed the caller); the warnings that then showed are fixed: an untitled event's alarm has a DESCRIPTION again, a SEQUENCE, PRIORITY, PERCENT-COMPLETE or SIZE that is not an integer is ignored
- the MIME::Types cache is private, not $::MimeTypes
- $ENV{JMAP_ALWAYS_FULL} (write whole objects in override patches) is gone; nothing set it
- nothing goes to STDERR: a second component of one UID without RECURRENCE-ID (which used to Dumper both events), DTEND with DURATION, an all-day DTSTART with a time zone, and a mandatory property Data::ICal finds missing are reported by the new warnings() method, the same shape as errors(); an unknown VALARM ACTION drops the alarm and is an errors() entry
- add t/side-effects.t
- the master the reader makes up for stand-alone recurrence instances is marked with the vendor-specific property "jscalendar.fastmail.com:syntheticMaster": true (the constant SYNTHETIC_MASTER; isSyntheticMaster($Event) tests it), and the writer goes by that alone. It used to guess, taking any object with organizerCalendarAddress and no participants for a synthetic master and silently not writing it, so a real event of that shape was lost. A marked master writes each of its instances as a RECURRENCE-ID component (an empty patch too, not an RDATE; an excluded one not at all) and no component of its own; every instance of it is an override, even one that patches nothing; and it starts at its earliest instance, not at whichever came first in the file, so that it reads back the same
- add t/synthetic-master.t
- eventsToVCalendar (and _argsToVCalendar) no longer changes its input: matching recurrenceRule's interval, count and byX numbers against a pattern stringified them in the caller's objects, which then encoded as JSON strings ("interval": "2")
- add t/no-mutation.t
- new public API for what consumers had to rebuild themselves:
- occurrences($Event, after => $from, before => $to, max => $n): the recurrence ids in [after, before), in order: the start (always the first occurrence), every date-time of the recurrenceRule (jscalendarbis-22 3.3.3.1, with every rule property the module reads, firstDayOfWeek, byYearDay, byWeekNo, bySecond and skip included; rscale must be gregorian), plus added recurrenceOverrides keys, minus excluded ones. A rule of its own, not DateTime::Event::ICal (which, among other things, loses most occurrences of a weekly rule with an interval and byMonth); bounded by max (default 5000) and by 400 years without a match
- instance($Event, $recurrenceId): that occurrence as an object of its own, the override patch applied with the module's own PatchObject code (participants/p1/participationStatus and the like), recurrenceId and recurrenceIdTimeZone set; undef if excluded
- add t/occurrences.t
- declare Data::ICal as a prerequisite, and stop loading XML::Spice and URI::Escape, which nothing used (a CPAN install without XML::Spice failed)
Documentation
autogenerated timezone data for ME::CalDAVTalk
Modules
autogenerated timezone data for Text::JSCalendar