Make isstored public

[MasonProtter]

This is a useful function when working with sparse arrays, I think we should make it public but give a weak guarantee, letting people know that a result of true could be wrong (though a value of false should be trustworthy).

[MasonProtter]

@LilithHafner could this get an assigned reviewer? I'm not sure how the bot system for it works or if it can be done retroactively.

[LilithHafner]

This PR proposes expanding the public API of Julia so it should get discussed at Triage before merging. It seems reasonable to me, but I don't have the expertise to know for sure. @stevengj, are you willing to weigh in here?

[halleysfifthinc]

What about using a default of missing? That would allow a strong interpretation of the true cases, and would match the standard definition/use-case for missingness.

[stevengj]

@MasonProtter, do you know of anyone that is using this function? I can't find anything on JuliaHub.

As far as I can tell, it is used only in stdlib/SparseArrays, and even there only in a single place: in the printing code in order to decide when to print a dot instead of a zero.

Without a better sense of its practical utility, I'm a little cautious about exporting it.

[MasonProtter]

I'd like to use it in GraphDynamics.jl and Neuroblox.jl . We have sparse connection matrix structures that describe the interaction between subsystems, and there are somecases where I care if something is "stored" vs just a zero connection.

Maybe this is a sign I need to stop using SparseArrays though and go build a new type that better exposes this difference. It'd be too bad though if I had to do that though because there's a lot of features of SparseArrays I really like for what I'm doing here.

[stevengj]

Using it for graphs is a good example. In that case, I'm in favor of just making this public.

[LilithHafner]

From triage, why can't this live in LinearAlgebra (or possibly even SparseArrays with a weak dependency on LinearAlgebra)? isstored is a fairly SparseArrays-specific concept. For example a rank-1 matrix which stores two vectors and computes their product on demand... what is stored?

[MasonProtter]

I don't see why it should be moved to another package. Base is the owner of AbstractArray, and isstored is basically a utility for AbstractArray subtypes to opt into. Base seems like the right place to me. I guess if Base really doesn't want to own it we could move it to LinearAlgebra, but I would say absolutely not SparseArrays, because SparseArrays depends on LinearAlgebra and not vice versa. LinearAlgebra has a lot of isstored overloads:

julia> methods(Base.isstored)
# 6 methods for generic function "isstored" from Base:
 [1] isstored(A::Bidiagonal, i::Int64, j::Int64)
     @ ~/.julia/juliaup/julia-1.12.0-rc1+0.x64.linux.gnu/share/julia/stdlib/v1.12/LinearAlgebra/src/bidiag.jl:131
 [2] isstored(D::Diagonal, i::Int64, j::Int64)
     @ ~/.julia/juliaup/julia-1.12.0-rc1+0.x64.linux.gnu/share/julia/stdlib/v1.12/LinearAlgebra/src/diagonal.jl:157
 [3] isstored(A::Union{LowerTriangular{T, S}, UnitLowerTriangular{T, S}, UnitUpperTriangular{T, S}, UpperTriangular{T, S}} where {T, S}, i::Int64, j::Int64)
     @ ~/.julia/juliaup/julia-1.12.0-rc1+0.x64.linux.gnu/share/julia/stdlib/v1.12/LinearAlgebra/src/triangular.jl:230
 [4] isstored(A::Tridiagonal, i::Int64, j::Int64)
     @ ~/.julia/juliaup/julia-1.12.0-rc1+0.x64.linux.gnu/share/julia/stdlib/v1.12/LinearAlgebra/src/tridiag.jl:693
 [5] isstored(A::SymTridiagonal, i::Int64, j::Int64)
     @ ~/.julia/juliaup/julia-1.12.0-rc1+0.x64.linux.gnu/share/julia/stdlib/v1.12/LinearAlgebra/src/tridiag.jl:450
 [6] isstored(A::AbstractArray{<:Any, N}, I::Vararg{Integer, N}) where N
     @ abstractarray.jl:610
``