Bug in Time module

[Simon Marlow]

>  | I'm not sure it's a bug - Haskell 98 doesn't require
>  | that the TimeDiff value returned from diffClockTimes
>  | is normalised in any way, and it can't be done in
>  | general of course because months and years have
>  | different numbers of days, even minutes have different
>  | numbers of seconds if leap seconds are taken into
>  | account.
> 
> You are absolutely right of course. Shouldn't this be
> regarded as a bug in the Haskell'98 spec then? It is not
> sure to me at all what a value of TimeDiff really means.

Well, there are two aspects to TimeDiff:

	1. addToClockTime t1 (diffClockTimes t1 t2) == t2
	   (the report should probably specify this)

	2. Being able to do things like
		   addToClockTime t1 noTimeDiff{tdYear=1}
	   to increase the year by one.  (noTimeDiff doesn't
	   exist in Haskell 98, but it's easy to define).

Note that certain uses of addToClockTime may be undefined - eg. what
does it mean to add one month to 31 January?  I remember having a long
discussion about this a while back, with the result that we now have the
TimeExts library which supposedly implements a saner kind of time
difference.  The previous discussion should be in the archives.

Cheers,
	Simon