deprecate randn((1,2))

[cossio]

Since rand((1,2)) was deprecated, it seems that randn((1,2)) should be deprecated too.

See https://discourse.julialang.org/t/rand-1-2-vs-rand-1-2/16254

[ExpandingMan]

I was thinking about this a little bit and I realized that it's not completely clear that this should be deprecated since randn only supports samples in continuous (or pseudo-continuous) spaces. For example, randn(Float64, (1,2)) makes sense but randn([1,2]) does not. I think the possibility that rand((1,2)) could be interpreted (by a user) as rand([1,2]) was the primary reason that rand((m,n)) was deprecated.

Still, the inconsistency is confusing, and after all randn does indeed support type arguments such as randn(Float32, (1,2)). My vote would be for deprecating rand((m,n)), but in light of the above, it might be a long time before we're actually allowed to do that under the constraints of semantic versioning.

[rfourquet]

I think the possibility that rand((1,2)) could be interpreted (by a user) as rand([1,2]) was the primary reason that rand((m,n)) was deprecated.

Correct, see #25278.

[ViralBShah]

Now, we have:

julia> rand((1,2))
2

julia> randn((1,2))
1×2 Matrix{Float64}:
 -0.937521  -0.590051