error when saving fully sparse array

[tbenst]

julia> h5_path = tempname()*".h5"
julia> data = zeros(100,1)
julia> H5SparseMatrixCSC(h5_path, "test", sparse(data))
HDF5-DIAG: Error detected in HDF5 (1.12.0) thread 0:
  #000: H5Pdcpl.c line 2004 in H5Pset_chunk(): chunk dimensionality must be positive
    major: Invalid arguments to routine
    minor: Out of range
ERROR: LoadError: Error setting chunk size
Stacktrace:
  [1] error(s::String)
    @ Base ./error.jl:33
  [2] h5p_set_chunk
    @ ~/.julia/packages/HDF5/0iEnL/src/api.jl:1211 [inlined]
  [3] set_chunk(::HDF5.Properties)
    @ HDF5 ~/.julia/packages/HDF5/0iEnL/src/HDF5.jl:1915
  [4] _prop_set!(p::HDF5.Properties, name::Symbol, val::Vector{Int64}, check::Bool)
    @ HDF5 ~/.julia/packages/HDF5/0iEnL/src/HDF5.jl:837
  [5] create_property(class::Int64; pv::Base.Iterators.Pairs{Symbol, Any, Tuple{Symbol, Symbol}, NamedTuple{(:chunk, :blosc), Tuple{Vector{Int64}, Int64}}})
    @ HDF5 ~/.julia/packages/HDF5/0iEnL/src/HDF5.jl:865
  [6] create_dataset(parent::HDF5.Group, path::String, dtype::HDF5.Datatype, dspace::HDF5.Dataspace; pv::Base.Iterators.Pairs{Symbol, Any, Tuple{Symbol, Symbol}, NamedTuple{(:chunk, :blosc), Tuple{Vector{Int64}, Int64}}})
    @ HDF5 ~/.julia/packages/HDF5/0iEnL/src/HDF5.jl:729
  [7] #create_dataset#32
    @ ~/.julia/packages/HDF5/0iEnL/src/HDF5.jl:737 [inlined]
  [8] h5writecsc(fid::HDF5.File, name::String, m::Int64, n::Int64, colptr::Vector{Int64}, rowval::Vector{Int64}, nzval::Vector{Float64}; overwrite::Bool, chunk::Nothing, blosc::Int64, kwargs::Base.Iterators.Pairs{Union{}, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ H5Sparse ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:250
  [9] h5writecsc
    @ ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:231 [inlined]
 [10] #h5writecsc#3
    @ ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:228 [inlined]
 [11] h5writecsc
    @ ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:227 [inlined]
 [12] H5SparseMatrixCSC(fid::HDF5.File, name::String, B::SparseMatrixCSC{Float64, Int64}; kwargs::Base.Iterators.Pairs{Union{}, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ H5Sparse ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:86
 [13] H5SparseMatrixCSC(fid::HDF5.File, name::String, B::SparseMatrixCSC{Float64, Int64})
    @ H5Sparse ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:86
 [14] H5SparseMatrixCSC(::String, ::String, ::Vararg{Any, N} where N; kwargs::Base.Iterators.Pairs{Union{}, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ H5Sparse ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:84
 [15] H5SparseMatrixCSC(::String, ::String, ::Vararg{Any, N} where N)
    @ H5Sparse ~/.julia/packages/H5Sparse/oVw2O/src/H5Sparse.jl:84
 [16] top-level scope
    @ ~/code/lensman/notebooks/debug2.jl:9
in expression starting at /home/tyler/code/lensman/notebooks/debug2.jl:9

[severinson]

Good catch. Seems like HDF5.heuristic_chunk returns Int64[] when called with a zero-length vector. We could replace this with Int64[1] for zero-length vectors, which would solve the issue. Performance would be terrible if more data is appended to this matrix later, but it's not clear what would be a better default than Int64[1].

[tbenst]

I think [] is / is intended to be a valid chunk: https://github.com/JuliaIO/HDF5.jl/blob/56c795cf876bcaeb37c9143fcccec01cbf0b1f30/src/HDF5.jl#L357

[severinson]

It looks like it, but create_group fails when chunk=[]. I've made a series of commits that address this issue, and allows for writing matrices with no stored entries to disk, by adding some code to handle empty vectors. Saving matrices with zero dimension still doesn't work though (e.g., sprand(0, 0)).
541781c acefb07 501335e