UndefVarError: FinchProtocolError not defined

[ericphanson]

I've been trying to figure out how to transfer the result of a finch computation into a "usual" SparseMatrixCSC, which seemed to uncover a small bug:

julia> using Finch

julia> using Finch: AbstractFiber

julia> function transfer!(x, y::AbstractFiber)
           @finch begin
               x .= y
           end
       end
transfer! (generic function with 1 method)

julia> x = spzeros(2,2)
2×2 SparseMatrixCSC{Float64, Int64} with 0 stored entries:
  ⋅    ⋅ 
  ⋅    ⋅ 

julia> transfer!(x, f)
ERROR: UndefVarError: `FinchProtocolError` not defined
Stacktrace:
  [1] declare!(arr::SparseArraysExt.VirtualSparseMatrixCSC, ctx::Finch.LowerJulia, init::Finch.FinchNotation.FinchNode)
    @ SparseArraysExt ~/.julia/packages/Finch/R62UH/ext/SparseArraysExt.jl:60
  [2] declare!(tns::Finch.FinchNotation.FinchNode, ctx::Finch.LowerJulia, init::Finch.FinchNotation.FinchNode)
    @ Finch ~/.julia/packages/Finch/R62UH/src/tensors/combinators/roots.jl:20
  [3] lower(root::Finch.FinchNotation.FinchNode, ctx::Finch.LowerJulia, #unused#::Finch.DefaultStyle)
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:112
  [4] (::Finch.LowerJulia)(root::Finch.FinchNotation.FinchNode, style::Finch.DefaultStyle)
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:34
  [5] (::Finch.LowerJulia)(root::Finch.FinchNotation.FinchNode)
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:32
  [6] (::Finch.var"#324#328")(ctx_3::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/execute.jl:105
  [7] #130
    @ ~/.julia/packages/Finch/R62UH/src/lower.jl:15 [inlined]
  [8] contain(f::Finch.var"#130#131"{Dict{Any, Any}, Finch.var"#324#328", Finch.LowerJulia}, ctx::Finch.JuliaContext, task::Nothing)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:69
  [9] contain(f::Function, ctx::Finch.JuliaContext)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:63
 [10] contain(f::Finch.var"#324#328", ctx::Finch.LowerJulia; bindings::Dict{Any, Any}, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:14
 [11] contain(f::Function, ctx::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:13
 [12] (::Finch.var"#323#327")(ctx_2::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/execute.jl:104
 [13] #130
    @ ~/.julia/packages/Finch/R62UH/src/lower.jl:15 [inlined]
 [14] contain(f::Finch.var"#130#131"{Dict{Any, Any}, Finch.var"#323#327", Finch.LowerJulia}, ctx::Finch.JuliaContext, task::Nothing)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:69
 [15] contain(f::Function, ctx::Finch.JuliaContext)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:63
 [16] contain(f::Finch.var"#323#327", ctx::Finch.LowerJulia; bindings::Dict{Any, Any}, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:14
 [17] contain(f::Function, ctx::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:13
 [18] lower_global(prgm::Finch.FinchNotation.FinchNode, ctx::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/execute.jl:94
 [19] (::Finch.var"#321#322"{Symbol, DataType})(ctx_2::Finch.LowerJulia)
    @ Finch ~/.julia/packages/Finch/R62UH/src/execute.jl:82
 [20] #130
    @ ~/.julia/packages/Finch/R62UH/src/lower.jl:15 [inlined]
 [21] contain(f::Finch.var"#130#131"{Dict{Any, Any}, Finch.var"#321#322"{Symbol, DataType}, Finch.LowerJulia}, ctx::Finch.JuliaContext, task::Nothing)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:69
 [22] contain(f::Function, ctx::Finch.JuliaContext)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:63
 [23] contain(f::Finch.var"#321#322"{Symbol, DataType}, ctx::Finch.LowerJulia; bindings::Dict{Any, Any}, kwargs::Base.Pairs{Symbol, Union{}, Tuple{}, NamedTuple{(), Tuple{}}})
    @ Finch ~/.julia/packages/Finch/R62UH/src/lower.jl:14
 [24] contain
    @ ~/.julia/packages/Finch/R62UH/src/lower.jl:13 [inlined]
 [25] #execute_code#320
    @ ~/.julia/packages/Finch/R62UH/src/execute.jl:79 [inlined]
 [26] execute_code
    @ ~/.julia/packages/Finch/R62UH/src/execute.jl:78 [inlined]
 [27] (::Finch.var"#316#319"{DataType, DataType})(ctx::Finch.JuliaContext)
    @ Finch ~/.julia/packages/Finch/R62UH/src/execute.jl:63
 [28] contain(f::Finch.var"#316#319"{DataType, DataType}, ctx::Finch.JuliaContext, task::Nothing)
    @ Finch ~/.julia/packages/Finch/R62UH/src/environment.jl:69
 [29] contain
    @ ~/.julia/packages/Finch/R62UH/src/environment.jl:63 [inlined]
 [30] macro expansion
    @ ~/.julia/packages/Finch/R62UH/src/execute.jl:62 [inlined]
 [31] (::Finch.var"#315#318"{DataType, DataType})()
    @ Finch ~/.julia/packages/Finch/R62UH/src/util.jl:71
 [32] #s355#314
    @ ~/.julia/packages/Finch/R62UH/src/util.jl:80 [inlined]
 [33] var"#s355#314"(::Any, ex::Any, opts::Any)
    @ Finch ./none:0
 [34] (::Core.GeneratedFunctionStub)(::Any, ::Vararg{Any})
    @ Core ./boot.jl:602
 [35] macro expansion
    @ ~/.julia/packages/Finch/R62UH/src/execute.jl:173 [inlined]
 [36] transfer!(x::SparseMatrixCSC{Float64, Int64}, y::Fiber{DenseLevel{Int64, SparseListLevel{Int64, Vector{Int64}, Vector{Int64}, ElementLevel{0, Int64, Int64, Vector{Int64}}}}})
    @ Main ./REPL[27]:2
 [37] top-level scope
    @ REPL[29]:1

I see the code in question is

function Finch.declare!(arr::VirtualSparseMatrixCSC, ctx::AbstractCompiler, init)
    throw(FinchProtocolError("Finch does not support writes to SparseMatrixCSC"))
end

so presumably I'm not allowed to do that, but the error object isn't defined in that scope.

[willow-ahrens]

Thanks for filing the issue!
I see I have a few TODOs:

• The protocol error needs to be imported from Finch in the sparsearrays extension package
• We should create a function to convert to SparseMatrixCSC (Do we just pirate SparseMatrixCSC(::Fiber) and convert(:Type{SparseMatrixCSC}, ::Fiber)? Should we also do Array(::Fiber) or sparse(::Fiber))
• We need docs for this too

[ericphanson]

You probably know all this already, but FWIW I don't think it's actually piracy to add a method to the constructor or sparse since you own the type in question (Fiber). I would also probably skip convert to start since that will happen silently (e.g. push! into a Vector{SparseMatrixCSC{Int,Int}}() would convert) and require explicit conversion instead (xref: https://docs.julialang.org/en/v1/manual/conversion-and-promotion/#Conversion-vs.-Construction).

But adding methods to SparseMatrixCSC and Array and sparse sounds very helpful to me! I had tried all three of those when playing in the REPL 😅. Also it would be helpful to document if there is any aliasing in the docstrings (e.g. if SparseMatrixCSC always allocates fresh arrays or if for objects like Dense(SparseList(Element(0.0))) it could possibly re-use data).

[willow-ahrens]

Related: #288 #321 #320

[willow-ahrens]

@ericphanson I think the pr above fixes this issue, let me know if there's more to be done on this!

[willow-ahrens]

side note:
Finch currently uses the broadcast syntax as an initializer syntax for array variables, so the Finch code to copy into a sparse array would be

@finch begin
    x .= 0
    for j=_,i=_
        x[i, j] = y[i, j]
    end
end

For simple pointwise expressions, Finch also supports regular broadcast (outside of a @finch region)

#no @finch, just calling broadcast
x .= y

You could also call copyto!(x, y).

However, the above examples won't work when x is SparseMatrixCSC until we implement #320