Ha! After playing with this, I discovered that only the seconds were set and all other fields remained untouched. At least in ghc's implementation. Interestingly, TimeDiff derives Eq and Ord, but I'd better not ask for their implementation...
TimeDiff has a perfectly well-defined meaning for Eq and Ord: one TimeDiff is equal to another iff all the fields are respectively equal. See http://www.haskell.org/pipermail/haskell-cafe/2002-December/003758.html
There is also a more fundamental problem with the TimeDiff data type. While seconds, minutes, hours, and days are clearly specified amounts of time, the duration of a month or a year may vary depending on the reference point where the time difference is applied.
This isn't a problem with the spec, I think. A TimeDiff of "1 month" is precisely a difference, which, when added to a given ClockTime, produces a ClockTime which is one month later (with respect to some timezone, presumably UTC since it isn't specified otherwise). That is, January 12th 12:00 UTC becomes February 12th 12:00 UTC, and so on. Adding one month to certain ClockTimes is meaningless: eg. adding one month to January 31st doesn't work. At least, that's how I believe it is supposed to work. GHC's implementation *is* completely wrong (which is perhaps where some of the confusion comes from), but the TimeExts library can be used instead of TimeDiffs.
My conclusion is that time differences really should be measured in seconds and picoseconds.
type TimeDiff = (Integer, Integer)
We already have this kind of TimeDiff: just don't use anything except the seconds and picoseconds fields. I agree with John Meacham in that it ought to be possible to get a raw time in terms of seconds/picoseconds since the epoch, and indeed GHC lets you do that: importing System.Time lets you get at the representation of ClockTime: data ClockTime = TOD Integer -- Seconds since 00:00:00 on 1 Jan 1970 Integer -- Picoseconds with the specified second This is a perfectly reasonable definition of absolute time, we don't need to talk about TAI at all, except to define what the epoch means. Cheers, Simon