Fixing ReinterpretArray breakage - first PR

[pablosanjose]

The julialang PR #23750 merged last month changed the output of reinterpret into a lazy ReinterpretArray <: AbstractArray type. This has broken NearestNeighbors in a few places, where the output of reinterpret was expected to be of type Vector instead. In particular the example in the README.md

using NearestNeighbors
data = rand(3, 10^4)
kdtree = KDTree(data; leafsize = 10)

fails with a MethodError.

I have made a branch locally with a fix for this, but this being my very first PR I am posting here what I did in case I messed up.

Since ReinterpretArray obeys the AbstractArray interface I just changed the signature of a number of functions expecting a ::Vector{T} into ::AbstractVector{T}. I also modified the BruteTree constructor to ensure conversion of generic AbstractArrays into a Vector. These changes make the README.md example work again, and they also makes it possible to once more pass tests up to a point. Tests eventually fail in relation to changes in broadcast, I think, but I think those failures are unrelated to ReinterpretArray. I will look into those later.

I would like some tips on how to proceed to make a PR. I assume I have to push my branch and then start the PR, but I have no permissions for this. What is the standard procedure?

[pablosanjose]

Maarten Pronk kindly pointed me to the 'fork' button in Slack. PR submitted!

[PaulMatlashewski]

Is there anything holding up this PR? I'm getting hit by this issue in a few places. Thanks!

[KristofferC]

The problem was if reinterpret was going to be made fast for 0.7 or if we copy the data. I guess för now we can copy and swap back to reinterpret if it gets optimized. I'll try fix up if needed and merge the PR today.

[KristofferC]

Ok, so StaticArrays doesn't work on 0.7 so we can't really update here until that is fixed.

[pablosanjose]

Ok.
I'm working with the allmerged branch of the pablosanjose/StaticArrays fork, which includes broadcast, random and some other fixes, and it works great on Julia master, in case someone cannot wait for merge decisions.
[EDIT: Not quite, I just encountered some new problems with StaticArrays broadcast in Julia master]

[KristofferC]

PR merged