Fix pointer to no longer assume contiguity

[mbauman]

Fixes #36402. Turns out this affected both pointer(::SubArray) as well as pointer(::AbstractArray, i).

[mbauman]

As I went through this, I realized that there's a question mark in how pointer(A, i) is documented: does the i mean an offset of elements or does it mean an index?

help?> pointer
search: pointer pointer_from_objref unsafe_pointer_to_objref codepoint

  pointer(array [, index])

  Get the native address of an array or string, optionally at a given location index.

To me, location index reads like a getindexy index (and not a memory offset), but this was intentionally disabled by PermutedDimsArray because it wasn't clear. This would definitively make it an index-location.

julia/base/permuteddimsarray.jl

Lines 56 to 61 in 52c55d7

	# It's OK to return a pointer to the first element, and indeed quite
	# useful for wrapping C routines that require a different storage
	# order than used by Julia. But for an array with unconventional
	# storage order, a linear offset is ambiguous---is it a memory offset
	# or a linear index?
	Base.pointer(A::PermutedDimsArray, i::Integer) = throw(ArgumentError("pointer(A, i) is deliberately unsupported for PermutedDimsArray"))

I'm really just trying to fix #36402 in this PR (and intentionally structured this to allow for further improvements to pointer, and we could re-enable PermutedDimsArray later, too), but I just wanted to call this ambiguity out. Cc @timholy and #20385.

[vtjnash]

It's intended to be the index of the ith element (getindexy)

[codecov-commenter]

Codecov Report

Merging #36405 into master will increase coverage by 0.00%.
The diff coverage is 83.90%.

@@           Coverage Diff           @@
##           master   #36405   +/-   ##
=======================================
  Coverage   86.13%   86.14%           
=======================================
  Files         349      349           
  Lines       65323    65401   +78     
=======================================
+ Hits        56265    56337   +72     
- Misses       9058     9064    +6     

Impacted Files	Coverage Δ
base/abstractarraymath.jl	94.81% <ø> (ø)
base/file.jl	79.00% <ø> (ø)
base/math.jl	90.57% <ø> (ø)
base/special/trig.jl	94.36% <79.10%> (-1.99%)	⬇️
base/abstractarray.jl	86.36% <100.00%> (+0.05%)	⬆️
base/array.jl	95.55% <100.00%> (ø)
base/methodshow.jl	82.25% <100.00%> (ø)
base/reflection.jl	86.92% <100.00%> (+0.36%)	⬆️
stdlib/LinearAlgebra/src/givens.jl	56.73% <0.00%> (-4.81%)	⬇️
stdlib/SparseArrays/src/linalg.jl	95.24% <0.00%> (+0.19%)	⬆️
... and 4 more

---

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 8db9495...65f4046. Read the comment docs.

[KristofferC]

Good to go?

[mbauman]

Yes, this is good to go on my side once tests pass.