Fix coloring with empty matrix as input

[blegat]

The change to _coloring fixes the following error:

julia> problem = ColoringProblem(; structure = :symmetric, partition = :column)
ColoringProblem{:symmetric, :column}()

julia> algo = GreedyColoringAlgorithm(; decompression = :substitution)
GreedyColoringAlgorithm{:substitution, 1, Tuple{NaturalOrder}}((NaturalOrder(),), false)

julia> S = spzeros(Int, 0, 0)
0×0 SparseMatrixCSC{Int64, Int64} with 0 stored entries

julia> coloring(S, problem, algo)
ERROR: ArgumentError: reducing over an empty collection is not allowed; consider supplying `init` to the reducer
Stacktrace:
  [1] _empty_reduce_error()
    @ Base ./reduce.jl:311
  [2] reduce_empty(f::Function, T::Type)
    @ Base ./reduce.jl:312
  [3] mapreduce_empty(::typeof(identity), op::Function, T::Type)
    @ Base ./reduce.jl:361
  [4] reduce_empty(op::Base.MappingRF{typeof(identity), typeof(max)}, ::Type{Int64})
    @ Base ./reduce.jl:350
  [5] reduce_empty_iter
    @ ./reduce.jl:373 [inlined]
  [6] mapreduce_empty_iter(f::Function, op::Function, itr::Vector{Int64}, ItrEltype::Base.HasEltype)
    @ Base ./reduce.jl:369
  [7] _mapreduce
    @ ./reduce.jl:421 [inlined]
  [8] _mapreduce_dim
    @ ./reducedim.jl:334 [inlined]
  [9] mapreduce
    @ ./reducedim.jl:326 [inlined]
 [10] _maximum
    @ ./reducedim.jl:984 [inlined]
 [11] _maximum
    @ ./reducedim.jl:983 [inlined]
 [12] maximum
    @ ./reducedim.jl:979 [inlined]
 [13] call_composed
    @ ./operators.jl:1099 [inlined]
 [14] ComposedFunction
    @ ./operators.jl:1096 [inlined]
 [15] #argmin##0
    @ ./reduce.jl:1020 [inlined]
 [16] MappingRF
    @ ./reduce.jl:92 [inlined]
 [17] afoldl
    @ ./operators.jl:599 [inlined]
 [18] _foldl_impl
    @ ./reduce.jl:60 [inlined]
 [19] foldl_impl
    @ ./reduce.jl:40 [inlined]
 [20] mapfoldl_impl
    @ ./reduce.jl:36 [inlined]
 [21] mapfoldl
    @ ./reduce.jl:167 [inlined]
 [22] argmin
    @ ./reduce.jl:1020 [inlined]
 [23] _coloring(speed_setting::SparseMatrixColorings.WithResult, A::SparseMatrixCSC{…}, ::ColoringProblem{…}, algo::GreedyColoringAlgorithm{…}, decompression_eltype::Type{…}, symmetric_pattern::Bool)
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/interface.jl:311
 [24] #coloring#28
    @ ~/.julia/dev/SparseMatrixColorings/src/interface.jl:196 [inlined]
 [25] coloring(A::SparseMatrixCSC{…}, problem::ColoringProblem{…}, algo::GreedyColoringAlgorithm{…})
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/interface.jl:189
 [26] top-level scope
    @ REPL[22]:1
Some type information was truncated. Use `show(err)` to see complete types.

After this, we have the following error that is fixed by the change in group_by_color:

julia> coloring(S, problem, algo)
ERROR: ArgumentError: reducing over an empty collection is not allowed; consider supplying `init` to the reducer
Stacktrace:
  [1] _empty_reduce_error()
    @ Base ./reduce.jl:311
  [2] mapreduce_empty(f::Function, op::Function, T::Type)
    @ Base ./reduce.jl:313
  [3] reduce_empty(op::Base.MappingRF{Base.ExtremaMap{typeof(identity)}, typeof(Base._extrema_rf)}, ::Type{Int64})
    @ Base ./reduce.jl:350
  [4] reduce_empty_iter
    @ ./reduce.jl:373 [inlined]
  [5] mapreduce_empty_iter(f::Function, op::Function, itr::Vector{Int64}, ItrEltype::Base.HasEltype)
    @ Base ./reduce.jl:369
  [6] _mapreduce
    @ ./reduce.jl:421 [inlined]
  [7] _mapreduce_dim
    @ ./reducedim.jl:334 [inlined]
  [8] mapreduce
    @ ./reducedim.jl:326 [inlined]
  [9] _extrema
    @ ./reducedim.jl:984 [inlined]
 [10] _extrema
    @ ./reducedim.jl:983 [inlined]
 [11] extrema
    @ ./reducedim.jl:979 [inlined]
 [12] group_by_color(::Type{Int64}, color::Vector{Int64})
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/result.jl:88
 [13] SparseMatrixColorings.TreeSetColoringResult(A::SparseMatrixCSC{…}, ag::SparseMatrixColorings.AdjacencyGraph{…}, color::Vector{…}, tree_set::SparseMatrixColorings.TreeSet{…}, decompression_eltype::Type{…})
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/result.jl:409
 [14] _coloring(speed_setting::SparseMatrixColorings.WithResult, A::SparseMatrixCSC{…}, ::ColoringProblem{…}, algo::GreedyColoringAlgorithm{…}, decompression_eltype::Type{…}, symmetric_pattern::Bool)
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/interface.jl:318
 [15] #coloring#28
    @ ~/.julia/dev/SparseMatrixColorings/src/interface.jl:196 [inlined]
 [16] coloring(A::SparseMatrixCSC{…}, problem::ColoringProblem{…}, algo::GreedyColoringAlgorithm{…})
    @ SparseMatrixColorings ~/.julia/dev/SparseMatrixColorings/src/interface.jl:189
 [17] top-level scope
    @ REPL[22]:1
Some type information was truncated. Use `show(err)` to see complete types.

Will add tests if this changes looks reasonable.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 100.00%. Comparing base (921c7a0) to head (9caca0f).
⚠️ Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff            @@
##              main      #300   +/-   ##
=========================================
  Coverage   100.00%   100.00%           
=========================================
  Files           20        20           
  Lines         2139      2143    +4     
=========================================
+ Hits          2139      2143    +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.

[gdalle]

Is acyclic coloring the only kind where empty matrices are problematic? What happens if only one dimension of the matrix is zero?
The fix is welcome (even though its usefulness isn't obvious to me), but then we may want to fix everything at once

[gdalle]

Thanks! Let's merge if the tests pass