Haskell time library: Weekday and Month

[wren ng thornton]

Henning Thielemann wrote:
> 
> I recently wanted to move from old-time to time in my bibtex package, 
> but I was no longer so enthusiastic about it, when I became aware, that 
> for instance weekdays and months are handled as plain numbers. Wouldn't 
> it be better to make them enumerations or newtypes (if efficiency is 
> important)? For a complex matter like coping with time it would be nice 
> to have precise types in order to know, what can be converted to what 
> given what context. In case of weekdays and months a custom type would 
> also make it irrelevant for the user, whether you count days and months 
> starting from zero or one.

I'm all for using newtypes instead of plain numbers, but there are some 
problems. Namely, what *is* a "day"? Is 31 a day? Well, it depends which 
month we're in. Is 364 a day? Well, it depends whether days include days 
of the year instead of just days of some month. Is 4982 a day? Not if we 
take "day" to be ordinal, naming the days of the month/year; but it 
would be fine if we consider "day" as a unit of duration for cardinals.

Similar problems arise for the definition of a "second" (due to leap 
seconds), and for other units for similar reasons. This may be why they 
didn't use enumerations/newtypes. But I think using newtypes for the 
ordinal concepts (and ignoring the cardinal concepts) is better than 
doing nothing.

-- 
Live well,
~wren