fix zero-dim similar

[ogauthe]

similar(::BlockArray behaves in a surprising way for 0-dimensional block arrays. This PR is a follow-up to #410 and aims for Base.similar(::T{<:AbstractBlockArray} to still return a BlockArray/BlockedArray in the zero-dimensional limit.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.64%. Comparing base (a80fcf7) to head (634a7af).
Report is 3 commits behind head on master.

Additional details and impacted files

@@             Coverage Diff             @@
##           master     #457       +/-   ##
===========================================
+ Coverage    0.00%   93.64%   +93.64%     
===========================================
  Files          18       19        +1     
  Lines        1642     1667       +25     
===========================================
+ Hits            0     1561     +1561     
+ Misses       1642      106     -1536     

☔ 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]

Could you elaborate on what the issue is?

[ogauthe]

Currently, it is not possible to create a zero dimensional BlockArray using Base.similar. Some methods fail, some other return an Array.

b0 = BlockArray{Float64}(undef)
b1 = BlockArray{Float64}(undef, [1])

@test_broken similar(b0, Float32, ()) isa BlockArray{Float32, 0}  # 0 axis output returns Array
@test_broken similar(b1, Float32, ()) isa BlockArray{Float32, 0}  # 0 axis output returns Array
@test similar(b1, Float32, ()) isa Array{Float32, 0}

similar(b0)  # 0 axis input errors
ERROR: MethodError: no method matching _BlockArray(::Array{Float64, 0}, ::Tuple{})
The function `_BlockArray` exists, but no method is defined for this combination of argument types.

Closest candidates are:
  _BlockArray(::Type{R}, ::NTuple{N, AbstractUnitRange{<:Integer}}) where {T, N, R<:(AbstractArray{<:AbstractArray{T, N}, N})}
   @ BlockArrays ~/.julia/packages/BlockArrays/v4qys/src/blockarray.jl:93
  _BlockArray(::R, ::BS) where {T, N, R<:(AbstractArray{<:AbstractArray{T, N}, N}), BS<:NTuple{N, AbstractUnitRange{<:Integer}}}
   @ BlockArrays ~/.julia/packages/BlockArrays/v4qys/src/blockarray.jl:65
  _BlockArray(::R, ::BS) where {N, R<:(AbstractArray{<:AbstractArray{V, N} where V, N}), BS<:NTuple{N, AbstractUnitRange{<:Integer}}}
   @ BlockArrays ~/.julia/packages/BlockArrays/v4qys/src/blockarray.jl:77
  ...

Stacktrace:
 [1] similar
   @ ~/.julia/packages/BlockArrays/v4qys/src/blockarray.jl:454 [inlined]
 [2] similar(a::BlockArray{Float64, 0, Array{Array{Float64, 0}, 0}, Tuple{}})
   @ BlockArrays ~/.julia/packages/BlockArrays/v4qys/src/abstractblockarray.jl:35
 [3] top-level scope
   @ REPL[4]:1

BlockedArray has similar issues.

This PR fixes similar with zero axes (either in the input or in the output) such that it always return a BlockArray/BlockedArray,

[jishnub]

Oh, I see, the issue is with broadcasting dropping the container for zero-dim cases.