Tuesday 28 April 2015

Strongly-Typed Time. Part 1: Rationale

I've quite recently become involved in an after-hours project that has a strong temporal component to it. Basically every interaction with the system will need to be labelled with a time, and they will constantly need to be compared and converted. Add to this the fact that the first beta customers are located on opposite sides of the Pacific, and that events can occur in a further 3 European countries, and a way to safely and unambiguously represent the time of something happening in a time and a place seems paramount.

While Joda Time has undoubtedly made date/calendar/timezone manipulation a happier task for the JVM developer, I'm looking for something stronger. I can pass around a DateTime all day long (no pun intended) but until I inspect its TimeZone I can't be sure where it originated from, or if it is in fact the canonical UTC.

As a result, there is nothing at compile time to stop me doing something like:
  def displayAllEventsBefore(allEvents:Seq[Event], threshold:DateTime) = {
 
    // allEvents have been normalized to UTC. But there's no way of knowing this:
    allEvents.filter(_.isBefore(threshold)).display    
  } 

  // ... much later, miles away
  val myTime = new DateTime() // Happens to be in "Europe/Paris"

  displayAllEventsBefore(events, myTime)

Which will work just fine most of the time, except when there's been an event in the last hour, when we won't see it. Or is it the other way around? Tricky, isn't it?

There's nothing in the type system to prevent these kinds of runtime problems. It comes down to developer diligence in naming/commenting/testing all the things - literally, every thing that uses a representation of time - to ensure correctness.

But hang on, aren't compilers really, really good at ensuring correctness?