(Friday + March) / April === 2.0

[LilithHafner]

This is too permissive.

[Seelengrab]

This is because none of the months have a proper type - they are literally numeric constants as magical numbers:

julia/stdlib/Dates/src/query.jl

Lines 259 to 270 in e708c5e

	"""
	January
	The first month of the year.
	# Examples
	```jldoctest
	julia> January
	1
	```
	"""
	const January = 1

julia> January
1

julia> January |> typeof
Int64

[jacobusmmsmit]

I can't think of any languages that have months and weeks be treated as integers by default.
Is there a precedent for this in a language someone knows? Is it useful in any applications?

[timholy]

Presumably they should be @enums

[llbit]

I can't think of any languages that have months and weeks be treated as integers by default. Is there a precedent for this in a language someone knows? Is it useful in any applications?

It could be useful for iterating over a range of months, for example August:December

[Seelengrab]

That can easily be accomodated by another approach as well. Ranges are not exclusive to numbers; 'a':'z' works too, for example, which has an analogous situation.

[KristofferC]

I don't think fixing this has to be breaking.

Although looking at it a bit more, it seems that returning integers for months (e.g. in the month function) is used in quite a lot of places.

[jacobusmmsmit]

How about making Dates an upgradeable stdlib?

I mean, this kinda misses the point a little as users have to opt in when the problematic constants are already defined.

[JeffBezanson]

I don't think this can be considered a bug --- Dates exports these constants and the design, clearly, was to make them just integers. You can certainly disagree with that design but changing it now would cause bugs.

[Seelengrab]

Right - hence the requirement to do this in a major version bump of the stdlib (which need not coincide with a major version bump of Julia, if I'm not mistaken).

[simonbyrne]

See #19210 (and #19221, which was folded into it)

[Seelengrab]

A bit depressing that #19210 has been open for numerous breaking releases now 😬

[KristofferC]

Dates could be moved to an upgradable stdlib and there could be a breaking version made there maybe?

[jacobusmmsmit]

The only potential problem I see with upgradeable stdlibs is backporting bugfixes and maintaining the non-breaking Dates package. Would everyone who just uses Julia 1.X without touching Dates be left bugfixless?

[Seelengrab]

Has seen Dates that much movement/fixes in the past that this is a concern? Most Dates stuff seems recent and about the core design of the package, not just related to easily fixable bugs.