width of integer & similar intervals

[tpapp]

Currently

julia> width(Date(1980,1,1)..Date(1980,1,10))
9 days

julia> width(1..10)
9

which makes sense in some contexts, but in others 10 would be a better choice (eg duration of a spell represented by the interval).

I would like to initiate discussion on how to handle this. The following come to mind:

1. Encourage the user to redefine width with +1 or equivalent whenever that makes sense, but don't do this automatically. Problem: width(1.0..10.0) ≠ width(1..10) if one redefines it. For Date and similar this would not be a problem.

2. To implement 1. more elegantly, one can define a trait, eg Date would have the trait RepresentsDuration(), width would test for that, and add 1 accordingly. The default would be not having this trait. Problem: same as above, advantage: user could inspect this behavior for generic types.

3. Add another generic function, eg span, which adds 1 whenever applicable. Again, with or without trait.

Thoughts?

[timholy]

How about length and define it only for Integer intervals?

[tpapp]

Thanks for the suggestion, see PR #22.

[dlfivefifty]

We now have duration. But probably the current behaviour is wrong and duration of dates should return a date?

julia> duration(Date(1980,1,1)..Date(1980,1,10))
10

julia> duration(1..10)
10

[timholy]

Yes, definitely the units need to be preserved. And just to check (sorry, short on time), I think we got rid of length, right?

[dlfivefifty]

Yes we got rid of length: length is a notion of cardinality, not measure, so it was an abuse of terminology, and also doesn't play well with Base's assumptions on length.

If anything, the following is the most reasonable definition of length:

length(d::AbstractInterval) = (@assert !isempty(d); ℵ₁)

[hyrodium]

length was removed #31, so I'll close this issue. Note that duration will also be removed in the future release. (#102)