Dates defines nonsensical conversions

[timholy]

I've posted this kind of thing elsewhere about other unitful quantities, but just noticed that we have the same problem with dates. Succinct example:

julia> d = convert(Dates.Day, 1)
1 day

julia> convert(Int, d)
1

Seems OK, right? But consider:

julia> s = convert(Dates.Second, d)
86400 seconds

julia> convert(Dates.Second, convert(Int, d))
1 second

You can get two wildly different answers when you convert days to seconds this way. The only sensible approach is not to define conversions between unitless and unitful objects.

[nalimilan]

Or to define such conversions only in one direction (probably the one from unitful to unitless?).

[TotalVerb]

See also related issues #16109, #10451, #15394.

[timholy]

@nalimilan, I don't think they can be defined at all. Consider:

julia> d = convert(Dates.Day, 1)
1 day

julia> d == 1
false

Saying that you can convert one object to another type, but then to have those two not be equal, is pretty confusing. One should force people to explicitly use a constructor (Dates.Day(1)), which is a computer science construct, not something with physical meaning.

[martinholters]

Attaching units and removing specific units are reasonable operations, but just removing any units from a quantity is usually dangerous and should not be easily available. "I know the computation lasted for 6.2, but I don't care whether it were seconds or hours."---That doesn't seem right.

And I also agree that attaching units should not be a convert, but maybe a constructor.