Fix respectful_similar with SparsityPatternCSC

[blegat]

Otherwise, decompress returns a dense matrix when a SparsityPatternCSC is given as argument to coloring, see below:

julia> S = sparse([
           0 0 1 1 0 1
           1 0 0 0 1 0
           0 1 0 0 1 0
           0 1 1 0 0 0
       ])
4×6 SparseMatrixCSC{Int64, Int64} with 9 stored entries:
 ⋅  ⋅  1  1  ⋅  1
 1  ⋅  ⋅  ⋅  1  ⋅
 ⋅  1  ⋅  ⋅  1  ⋅
 ⋅  1  1  ⋅  ⋅  ⋅

julia> P = SparseMatrixColorings.SparsityPatternCSC(S)
4×6 SparseMatrixColorings.SparsityPatternCSC{Int64}:
 false  false   true   true  false   true
  true  false  false  false   true  false
 false   true  false  false   true  false
 false   true   true  false  false  false

julia> problem = ColoringProblem()
ColoringProblem{:nonsymmetric, :column}()

julia> algo = GreedyColoringAlgorithm()
GreedyColoringAlgorithm{:direct, 1, Tuple{NaturalOrder}}((NaturalOrder(),), false)

julia> result = coloring(P, problem, algo); # We give P, not S

julia> B = compress(S, result)
4×3 Matrix{Int64}:
 1  1  1
 1  1  0
 1  1  0
 1  1  0

julia> C = decompress(B, result)
4×6 Matrix{Int64}:
 0  0  1  1  0  1
 1  0  0  0  1  0
 0  1  0  0  1  0
 0  1  1  0  0  0

This affects blegat/ArrayDiff.jl#23 because I just create a SparsityPatternCSC with the detected sparsity pattern and use coloring with this as input.

I can write tests if the change sounds reasonable

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (921c7a0) to head (77d32d5).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #299   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           20        20           
  Lines         2139      2141    +2     
=========================================
+ Hits          2139      2141    +2     

☔ 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.

[gdalle]

Can we just overload similar instead?

[blegat]

The issue is that SparsityPatternCSC does not have any field to store the values so it cannot be used for the decompression.
In a sense, coloring only needs the sparsity pattern but decompression needs values so we need some kind of similar_but_with_values ^^ I thought respectful_similar would serve this purpose

[gdalle]

It does have values since SparsityPatternCSC <: AbstractMatrix{Bool}. And since similar preserves the sparsity pattern for SparseMatrixCSC, I think it would make sense to define

function similar(A::SparsityPatternCSC, ::Type{T}) where {T}
    return SparseArrays.SparseMatrixCSC(
        A.m,
        A.n,
        A.colptr,
        A.rowval,
        similar(A.rowval, T),
    )
end

[blegat]

Good point, what about similar with Bool ?

[gdalle]

It should probably also return a SparseMatrixCSC, that way values are modifiable with a surface API

[gdalle]

Lgtm!