Fix device

[Tokazama]

Fixing device following discussion in #120. Default is now CPUIndex and no default for StridedArray. Instead, array types that work with pointer and define strides should explicitly define this

ArrayInterface.device(::Type{T}) where {T<:NewType} = device(parent_type(T))

This should also makes it so GPUArrays won't need to define device for all the StridedArray types because the proper AbstractDevice will be pulled from the original parent type if strides are defined.

[chriselrod]

Once upon a time I thought that StrideArray's purpose was to say it supports BLAS, which requires both strided memory and pointer (or Base.unsafe_convert) to work.

[codecov]

Codecov Report

Merging #123 (f7efa58) into master (f957da5) will decrease coverage by 0.11%.
The diff coverage is 76.92%.

@@            Coverage Diff             @@
##           master     #123      +/-   ##
==========================================
- Coverage   82.82%   82.71%   -0.12%     
==========================================
  Files          10       10              
  Lines        1520     1527       +7     
==========================================
+ Hits         1259     1263       +4     
- Misses        261      264       +3     

Impacted Files	Coverage Δ
src/ArrayInterface.jl	81.84% <76.92%> (-0.43%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f957da5...f7efa58. Read the comment docs.

[Tokazama]

I'm still trying to figure out the ambiguity on nightly, but does this approach look correct?

[chriselrod]

I'm still trying to figure out the ambiguity on nightly

JuliaLang/julia#39698

It's not an ambiguity error, but an error that happens while checking for ambiguities.

does this approach look correct?

Yes, looks good.