X-WR-TIMEZONE considered harmful?

In a pair of recent entries, Semantic web 101: Say what you mean and The long tail of the iCalendar ecosystem, I’ve begun to report on what I’m learning about the state of the iCalendar ecosystem as I work in parallel on the elmcity project and on the iCalendar Validator. Today I’ll focus on just one of a number of issues I’ve run into. Consider these two screenshots:

google calendar hotmail calendar

On the left you see Google Calendar displaying two calendars, each representing a single event — the Brower Youth Awards on October 18 at the Herbst Theater in San Francisco — in a different way. On the right you see Hotmail Calendar displaying the same two calendars. The event will happen at 5:30 Pacific time on the 18th. I found it on the Berkeleyside site whose events page offers a companion iCalendar feed.

If you load that feed into both Google Calendar and Hotmail Calendar, and if your calendars are set to Eastern time, you’ll see what’s shown above. If your calendars are set to another timezone the times will be shifted but the pink ones still won’t match.

The green ones should always match and should always be what you’d expect. For me, looking at a 5:30 Pacific event through the lens of calendars set to Eastern, I’d expect both calendars to display 8:30.

What’s the difference between the pink calendar and the green calendar? Here’s the pink one. It’s just the original calendar reduced to a single event. Like the original it declares its timezone using the nonstandard X-WR-TIMEZONE property:

PRODID:-//Refresh Web Development//Helios Calendar//EN
SUMMARY:Brower Youth Awards 2011
LOCATION:Herbst Theater - 401 Van Ness \, San Francisco\, CA US 94102

And here’s the green one. Again it’s a derivation of the original calendar that reduces to a single event. But it also declares its timezone in the standard way, using a VTIMEZONE component and then referring to it using the TZID parameter of the DTSTART and DTEND properties:

PRODID:-//Refresh Web Development//Helios Calendar//EN
SUMMARY:Brower Youth Awards 2011
LOCATION:Herbst Theater - 401 Van Ness \, San Francisco\, CA US 94102

As we see in the picture above, the event time on the green calendar shows up the same way in both Google Calendar and Hotmail Calendar. It should also be the same in any calendar program that supports the iCalendar standard.

The event time on the pink calendar, though, is a up for grabs. Calendar programs that strictly follow the iCalendar standard should ignore X-WR-TIMEZONE and always display the local time, 5:30PM, which will be right for people in the Pacific timezone and wrong for everybody else. Hotmail Calendar does this. Programs that use X-WR-TIMEZONE, on the other hand, can render this calendar just as they would render a standard calendar. Google Calendar does this.

Why do I care? I have to decide whether the elmcity service will or won’t consider X-WR-TIMEZONE to be meaningful. The service is based on DDay.iCal, the same standards-based parser that powers the iCalendar Validator. So when the the service reads the pink calendar, and renders it for users in Berkeley, it will do the wrong thing from their point of view.

To do the right thing for Berkeley it would need to do the wrong thing by iCalendar: transform the nonstandard X-WR-TIMEZONE property into a standard VTIMEZONE component, and then transform all the dates so that they refer to the VTIMEZONE’s TZID. In order to create that VTIMEZONE component, it would interpret X-WR-TIMEZONE value as a TZID (timezone ID) from the Olson database. A Unix-based service would look up the TZID in Olson, find the rule for the timezone — i.e., offsets from GMT for standard time and daylight savings time, and when to appy them — and express the rule using VTIMEZONE syntax. A service running on Windows Azure, like mine, would instead need to map the Olson name to a Windows timezone name, look up the rule using a Windows API, and then express the rule in VTIMEZONE syntax.

Of course this is a slippery slope because, in the end, I’m only guessing what X-WR-TIMEZONE is supposed to mean. Here’s Rick DeNatale engaging in the same kind of guesswork:

Someone pointed me to this icalendar file of Australian holidays for a test case:


This contains NO VTIMEZONE components, but does have the calendar property: X-WR-TIMEZONE:Australia/Sydney

Googling indicates that this is a non-standardized property, but it seems to be used by several calendar apps including Apple’s ical.app
and Google calendar.

I know that it’s non-standard, but it seems to be somewhat important for interoperability. I’m looking for some kind of information about
what it means in general.

It seems to indicate a default tzid for the whole calendar. In the absence of timezone components I’m not sure how to interpret the tzid, though.

Australia/Sydney IS a time zone identifier in the Olsen database, is it standard practice to use olsen tzids in X-WR-TIMEZONE calendar attributes?

Fortunately I can bring some data to bear on this question. Thanks to the iCalendar Validator I can analyze public calendars produced by a variety of iCalendar producers. In The long tail of the iCalendar ecosystem I listed the names of about 600 producers seen recently by the Validator. Of those, about 100 use X-WR-TIMEZONE instead of VTIMEZONE, and 70 of those 100 use local rather that UTC date syntax which implies they are depending on X-WR-TIMEZONE for correct interpretation of those dates.

Note that Google Calendar, the 800-pound gorilla in this space, is not one of those 70 producers. When it writes iCalendar format it uses both X-WR-TIMEZONE and VTIMEZONE; the latter ensures that Google Calendars can be understood properly by standard parsers that don’t support X-WR-TIMEZONE. The 100 producers I’m talking about, though, are using only X-WR-CALENDAR in a way that suggests they expect a nonstandard transformation. The fact that Google Calendar performs that transformation is, of course, a major reason why producers would expect it to happen everywhere.

Should X-WR-TIMEZONE be standard? That’s debatable. It would certainly make life easier for iCalendar producers. They could just mention a timezone rather than having to extract its rule from their operating systems and express the rule in VTIMEZONE syntax. One of the reasons for the success of RSS and Atom, after all, is that it’s always been easy to whip up an RSS or Atom feed which you can then check with the Feed Validator. An analogous simplicity for iCalendar producers would help grow the iCalendar ecosystem.

On the other hand if an iCalendar feed were to only mention a timezone without fully defining it, then the consumer would have to do the work that the producer didn’t. That’s problematic as Doug Day, author of the iCalendar Validator, notes in a recent email exchange:

Unfortunately, without including the actual time zone information in the calendar (i.e. via VTIMEZONE), you can’t be sure that the date/times you’re representing are accurate, even when using X-WR-TIMEZONE. For example, if you’re on a Windows XP machine that hasn’t been updated in 5 or 6 years, your system time zone information will be inaccurate. However, if the VTIMEZONE were included in the calendar, it would remain accurate, even on an older machine with out-of-date time zone definitions. Also, in order to interpret X-WR-TIMEZONE, you’d need to be in an environment where interpreting Olson time zone is realistic (easy on Linux, harder on Windows). I know a global, online time zone registry is in the works, but I don’t think it’s to a point where it’s useful, yet.

You might wonder why all this timezone stuff is even necessary. After all, an iCalendar feed can simply omit VTIMEZONE (and/or X-WR-TIMEZONE), express dates and times in UTC (Coordinated Universal Time), and use UTC syntax for all dates and times. Why not just do that? I asked Doug Day about this a while ago, and here was his reply:

The biggest problem is with recurring events and daylight/standard time transitions. For example, consider the following (all hypothetical):

1. I live in Salt Lake City, Utah

2. I want to schedule a meeting, starting on September 7, 2009 at 9:00 A.M, which recurs every month on the first Monday.

3. Some of the people attending this meeting live outside my current time zone.

So, here are the occurrences you’re ultimately after:

– September 7, 2009 – 9:00 A.M. (3:00 PM UTC)

– October 5, 2009 – 9:00 A.M. (3:00 PM UTC)

– November 2, 2009 – 9:00 A.M. (4:00 PM UTC)

As you can see, once the time changes from daylight back to standard time, so does the UTC representation of that time. So, if you had scheduled your event in UTC time, when the time zone changes, your event time will actually have changed (to 10:00 A.M., rather than 9:00 A.M.)

For this reason, among others, it’s always best to include time zone information whenever available. Traditionally, it’s been pretty difficult to include that information, and it’s more often left out than included.

So what shall I do with X-WR-TIMEZONE? I’ve decided to support it experimentally. If you start a new hub on the elmcity service, the default is to ignore X-WR-TIMEZONE. But if your hub has important sources that depend on it, as Berkeley does, then you can override the default so the times will be as you expect.

Meanwhile we’re going to update the iCalendar Validator to warn producers about this issue. There’s nothing technically invalid about a calendar that uses X-WR-TIMEZONE without VTIMEZONE. To a parser that strictly interprets the RFC5545 standard, that property is just a name that’s “reserved for experimental use.” But as has always been true of the RSS/ATOM Validator, the iCalendar Validator aims to deliver useful real-world guidance. Producers that use X-WR-TIMEZONE alone to declare a timezone should know that while it may often yield expected results, it’s not guaranteed to do so. It would be better to use a standard VTIMEZONE.

One thought on “X-WR-TIMEZONE considered harmful?

  1. The point is that X-WR-TIMEZONE is sooo much easier to implement for those poor implementors. I stumbled across this because I wanted to implement VTIMEZONE on IOS to get functionality which is not given to the dev crowd by Apple (internally they must have this API!). I have no clue how one can get those fancy DTSTART values + RRULE values for each possible timezone, this is neither published in RFC (oh yes the olson DB is there…) nor has the IOS API methods to figure out the daylight rules for infinite recurring events.
    However putting X-WR-TIMEZONE in the generated ics is a one liner and hence all those nice specs are undermined by the reality.

Leave a Reply