Sized AbstractArray

[mateuszbaran]

Allocation-free statically sized views in Julia 1.5 🎉

The code should be more-or-less complete but I need to add more tests.

One thing that is potentially controversial (and subject to change) is allowing one way of reshaping. Sometimes I represent arrays of different sizes as parts of a long 1-dimensional array, I could of course call reshape before wrapping in SizedArray but that's easy enough to do directly in SizedArray, and currently SizedArray allows for even more reshaping. Relevant PR: #666.

This PR supersedes #341 .

[c42f]

The CI failures here are weird - it's a git error at checkout where supposedly d2adcd0 doesn't exist. But it exists in my local repo.

[mateuszbaran]

I've pushed a new commit and the tests seem to be running OK.

I have one question here: do you think view(::SizedArray, something...) should (where possible) return a SizedArray wrapping a view of the wrapped AbstractArray?

[andyferris]

I have one question here: do you think view(::SizedArray, something...) should (where possible) return a SizedArray wrapping a view of the wrapped AbstractArray?

To me - the ideal behavior has axes(getindex(a, b)) === axes(view(a, b)).

If this involves effort I would wait until we support axes which are "partially static" (a mixture of OneTo and SOneTo) and we can clean it all up then.

[mateuszbaran]

To me - the ideal behavior has axes(getindex(a, b)) === axes(view(a, b)).

As far as I can tell, this is already the case 🙂 . I'm primarily concerned that without swapping of SubArray and SizedArray when taking a view lots of methods from StaticArrays.jl couldn't be used.

[mateuszbaran]

I think this is ready for review now. I have finished views of SizedArray and MArray.

[andyferris]

I think this is good-to-go (thanks!) but I'd prefer the MArray view to go into the other file. Julia will still find the new_out_size function.

[c42f]

Based on Andy's review I think this should be good to merge.

I went ahead and released 0.12.4 this morning to get that in before this breaking change.

[mateuszbaran]

I've fixed one remaining problem with views. Failure on nightly is related to a change in printing of types.

[chriselrod]

Could you define parent now that the parent array's properties determine those of the SizedArray?

[mateuszbaran]

Sure, I've just added parent (and vec).

[mateuszbaran]

Are there any plans to release StaticArrays 0.13 sometime soonish? If not, I think I'll put sized abstract arrays in HybridArrays.jl for now. It would solve many problems with SSubArray.

[c42f]

@mateuszbaran I approved this, so I was kind of waiting for you to go ahead and merge it. Go on :)

Regarding release, I was imagining we should just release 1.0 next, following the most conservative release plan (ie, just release it more or less as-is not trying to get any more major breaking changes in). IIRC there was one other minor breaking change we should do.

[mateuszbaran]

Great! I will merge it.

Releasing 1.0 would be quite a big thing and I'm happy with your plan about it 👍 .