Special case blocks for non-blocked arrays

[mtfishman]

Closes #403.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 94.07%. Comparing base (a4a0b93) to head (c0d077d).
Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #474      +/-   ##
==========================================
+ Coverage   93.81%   94.07%   +0.26%     
==========================================
  Files          19       19              
  Lines        1761     1772      +11     
==========================================
+ Hits         1652     1667      +15     
+ Misses        109      105       -4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dlfivefifty]

The conflicts need to be resolved.

[mtfishman]

I'll add more tests for the missing lines.

[dlfivefifty]

I'm not convinced the added complexity in the code is worth it.

Also, why not just use [A]?

[mtfishman]

I'm not convinced the added complexity in the code is worth it.

I also wish we could avoid it, but unfortunately in some cases SubArray wrappers are not always easy to deal with, for example when sparse arrays or GPU arrays are involved, and a definition like this was helpful in use cases I'm working on (I wish I could share a specific example where I found it was helpful but I can't think of one right now, I defined this in my own code to avoid a certain issue and it has "just worked" ever since).

Also, why not just use [A]?

The issue is the definition of setindex!. The closest Base object would be B = fill(A, ntuple(one, ndims(A))), but the issue with that is B[1, ..., 1] = X would replace the entire block instead of reusing the memory of the block, and wouldn't check the size of the block properly:

julia> A = randn(2, 2)
2×2 Matrix{Float64}:
 -1.02513    0.497601
  0.723849  -0.255301

julia> B = fill(A, ntuple(one, ndims(A)))
1×1 Matrix{Matrix{Float64}}:
 [-1.0251267250499947 0.4976007776562032; 0.7238493487560987 -0.25530073591274616]

julia> B[1, 1]
2×2 Matrix{Float64}:
 -1.02513    0.497601
  0.723849  -0.255301

julia> B[1, 1] = randn(4, 4)
4×4 Matrix{Float64}:
 -2.21523    1.72294     0.348749  -0.793305
  0.762541   0.979422   -0.193424  -0.217467
  1.92247    0.501233   -0.445871   1.352
 -0.977946  -0.0563409  -0.160537  -0.714262

julia> B[1, 1]
4×4 Matrix{Float64}:
 -2.21523    1.72294     0.348749  -0.793305
  0.762541   0.979422   -0.193424  -0.217467
  1.92247    0.501233   -0.445871   1.352
 -0.977946  -0.0563409  -0.160537  -0.714262

i.e. it doesn't really act as a view into the block data in the same way BlocksView would. Of course, the definition of setindex! in this PR would only work for mutable arrays, but that's the case with BlocksView as well.