mention getindex() shouldn't return a "view"

[Moelf]

fix #47078

[ViralBShah]

While the added statement is correct, I don't think it addresses the issue.

[Moelf]

isn't the issue we agree it should never be a view but just didn't mention it? what DataFrames.jl does is not our issue (technically)

[ViralBShah]

Never mind, I missed the second file in the PR. Yes, the issue is basically whether getindex is allowed to return a view or not.

[Moelf]

can we merge this?

[ViralBShah]

I'd like a few more eyes on it. @KristofferC Is this reasonable to merge?

[KristofferC]

Shouldn't this be in the documentation for getindex or something? It somehow doesn't feel like this is the right place for a side note like this.

[jishnub]

I'd prefer the phrasing to suggest what to do instead of warning against what not to do. A view is one of the many wrappers that getindex shouldn't return. It shouldn't return a reshaped array either, for example getindex(a::AbstractArray, ::Colon) = reshape(a, Val(1)) isn't a view technically, but it shares the underlying memory with the parent array.

[Moelf]

@KristofferC this is for developers.

for base vectors, getindex don't return view as outlined in multi-dimensional array manual, and we already tell user to use @view if they don't want a copy.

[nalimilan]

Can we revive this? I hope my suggestions can help addressing the comments.

[nalimilan]

How about specifying A::AbstractArray on the line above while we're at it? The description below isn't guaranteed to match non-array objects.

[Moelf]

that page is specific to array interface, i think thats implied

[adienes]

+1

[timholy]

Thanks!