Fix no_offset_view for Slice

[jishnub]

Currently, the implementation of no_offset_view for a Slice doesn't actually ensure that there is no offset. This is because Base.Slice(UnitRange(a)) has UnitRange(a) as its axes, which are usually offset. This PR changes the behavior to return a UnitRange(a) instead.

Fixes #375

The flip side of the current implementation is that the Slice information is lost, so IndexStyle of 2D views might become IndexCartesian once they are routed through no_offset_view.

julia> arr = OffsetArray(reshape(1:15, 3, 5), 2, 3)
3×5 OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8) with eltype Int64 with indices 3:5×4:8:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> V = view(arr, :, :)
3×5 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), :, :) with eltype Int64 with indices 3:5×4:8:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> IndexStyle(V)
IndexLinear()

julia> V_nooff = OffsetArrays.no_offset_view(V)
3×5 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), 3:5, 4:8) with eltype Int64:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> IndexStyle(V_nooff)
IndexCartesian()

An alternate approach might be to shift the parent to one-based indices, and use Base.Slice{Base.OneTo{Int}} axes, which preserve this information. This, however, departs from the current approach of preserving the parentindices of the SubArray.

Edit: the alternate approach is implemented now.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 95.69%. Comparing base (497deea) to head (284625d).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #376      +/-   ##
==========================================
+ Coverage   95.53%   95.69%   +0.16%     
==========================================
  Files           6        6              
  Lines         448      465      +17     
==========================================
+ Hits          428      445      +17     
  Misses         20       20              

☔ 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.

[jishnub]

In 96d0f15, I have explored the alternate idea, where we shift the indices of the parent as well as the parentindices of the view:

julia> arr = OffsetArray(reshape(1:15, 3, 5), 2, 3)
3×5 OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8) with eltype Int64 with indices 3:5×4:8:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> V = view(arr, :, :)
3×5 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), :, :) with eltype Int64 with indices 3:5×4:8:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> V_nooff = no_offset_view(V) # unwraps the `view` as well as the `OffsetArray`
3×5 reshape(::UnitRange{Int64}, 3, 5) with eltype Int64:
 1  4  7  10  13
 2  5  8  11  14
 3  6  9  12  15

julia> IndexStyle(V_nooff) # retains the `IndexStyle`
IndexLinear()

julia> V = view(arr, :, 4:4)
3×1 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), :, 4:4) with eltype Int64 with indices OffsetArrays.IdOffsetRange(values=3:5, indices=3:5)×Base.OneTo(1):
 1
 2
 3

julia> no_offset_view(V) # changes the `parentindices` as well as the indices of the `parent`, but the axis of the view is still a `Slice`
3×1 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 1:3, 4:8), :, 4:4) with eltype Int64:
 1
 2
 3

julia> V = @view arr[begin:end, 4:4]
3×1 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), 3:5, 4:4) with eltype Int64:
 1
 2
 3

julia> no_offset_view(V) # no change if there isn't a `Slice` as an axis
3×1 view(OffsetArray(reshape(::UnitRange{Int64}, 3, 5), 3:5, 4:8), 3:5, 4:4) with eltype Int64:
 1
 2
 3