inconsistent behaviour of `hasdim` on `X` vs `Dim{:X}`

[tiemvanderdeure]

Stumbled across this behaviour while making some tests for materializing tables into dimarrays.

using DimensionalData: Dim, X, hasdim
ds1 = (X(1:10),)
ds2 = (Dim{:X}(1:10),)
hasdim(ds1, X) # true
hasdim(ds1, :X) # true
hasdim(ds2, X) # false
hasdim(ds2, :X) # true
ds1 == ds2 # false

Is this intended behaviour? I thought everything just worked through symbols.

In any case this is confusing because it's not obvious at all that something has Dim{:X} dimension and not X, since these behave the same way otherwise.

[rafaqz]

Yeah, maybe some checks still consider Dim{:X} and X to be different. We should maybe fix that

[felixcremer]

According to the Dim docstring Dim types and the Dimension types are different:

Dimension types take precedence over same named Dim types when indexing
with symbols, or e.g. creating Tables.jl keys.

I think it is fine that Dim{:X} and X are different because one could simply use X if one wants the behaviour of X.

And also which Dimension types would we do special cases for? X, Y and Ti or also the dimension types that have been defined either in packages or by the user themself?

Even if we special case it here, they will always behave slightly differently, because Dim{:X} is not a XDim

julia> X<:XDim
true

julia> Dim{:X}<:XDim
false

[rafaqz]

Yeah, we do need to resolve that inconsistency. It's just annoying to do and breaking.

Theres an old PR #697 to swap to using traits rather than inheritance for XDim etc