getindex(A::AbstractArray, ::Val{:..})

[ranocha]

See SciML/OrdinaryDiffEq.jl#214

[ranocha]

I'm just curious: Why do you use only Travis and not AppVeyor for this repository?

[ChrisRackauckas]

No reason, just never got it setup.

[ranocha]

Okay, I thought I might have missed something.

[mbauman]

This seems a little strange to me — why extend getindex here instead of view?

[ChrisRackauckas]

SciML/OrdinaryDiffEq.jl#214

I wanted it to be the identity operation A[..] === A. Since it's over the full thing, I'm not sure a view is necessary.

Mostly it's to make that broadcasting free.

[mbauman]

My point is purely semantics. I'd expect A[..] to copy and view(A, ..) to alias. Making the former an identity is far more surprising than the latter.

I'm beginning to think the @. should imply @views. You're probably the heaviest user of @. — what would you think about that?

[ChrisRackauckas]

Anytime I do @. with [idxs] I put @views. For a single broadcast I am not sure there's a case where this is not beneficial (of course if you use the save views/slices over and over, that can be a different story though, or in matrix computations).

I would think it would help a lot of people if @. used views since the whole idea of it is "it's fusing everything so I expect it to not allocate", which isn't true when you are only doing a subset of indices without @views.