fixes for breakage introduced by recent Julia versions #220
Conversation
It's not an AbstractMatrix after a breaking change in LinearAlgebra: JuliaLang/LinearAlgebra.jl#1097. Also, LQ decomposition in Julia is only consistent for square matrices.
|
All tests pass, see at https://github.com/aplavin/NamedDims.jl/actions/runs/16301695343/job/46037637119! |
|
CI had stopped running due to activity. I clicked some buttons... |
| for (T, S) in [ | ||
| (:NamedDimsVecOrMat, :NamedDimsVecOrMat), | ||
| (:NamedDimsVecOrMat, :AbstractVecOrMat), | ||
| (:AbstractVecOrMat, :NamedDimsVecOrMat), | ||
| for (T, S, U) in [ | ||
| (:NamedDimsVecOrMat, :NamedDimsVecOrMat, :AbstractVecOrMat), | ||
| (:NamedDimsVecOrMat, :AbstractVecOrMat, :AbstractVecOrMat), | ||
| (:AbstractVecOrMat, :NamedDimsVecOrMat, :AbstractVecOrMat), | ||
| # needed since Julia 1.10 because of ambiguities introduced by SparseArrays | ||
| (:(NamedDimsVecOrMat{<:Any,<:Number}), :(NamedDimsVecOrMat{<:Any,<:Number}), :(AbstractVecOrMat{<:Number})), | ||
| (:(NamedDimsVecOrMat{<:Any,<:Number}), :(AbstractVecOrMat{<:Number}), :(AbstractVecOrMat{<:Number})), | ||
| (:(AbstractVecOrMat{<:Number}), :(NamedDimsVecOrMat{<:Any,<:Number}), :(AbstractVecOrMat{<:Number})), |
There was a problem hiding this comment.
Sad but probably unavoidable
There was a problem hiding this comment.
I didn't check if all entries are needed for the tests to pass (probably no), just repeated the same things as above with <: Number added.
Idk what the ultimate solution to ambiguities in Julia should be... Note that here in NamedDims ambiguities aren't even related to KeyedArray <-> NamedDims nesting, they are just caused by wanting to overload v/hcat.
Although maybe overloading v/hcat is not ultimately necessary?.. Didn't explore it.
There was a problem hiding this comment.
Do you know if the 3rd loop is required to pass?
I would love there to be a better way than overloading hcat. At some point wanted to make Base use axes everywhere & then perhaps there's a chance you could introduce some pairwise promote-like scheme. But... for now we hack things to keep working?
I thought there was talk of removing the offending SparseArrays methods but it looks like nothing has happened... JuliaSparse/SparseArrays.jl#431 is the relevant thread.
There was a problem hiding this comment.
Which 3rd loop? Do you mean the one with reduce below?
There was a problem hiding this comment.
By 3rd I mean the one over U, instead of cs::AbstractVecOrMat... before.
There were 3^2 == 9 methods, PR currently makes 6^3 == 216 instead. (And 4096 in mcabbott/AxisKeys.jl#169 ) I don't know that this causes any problems but it does seem like a lot!
Numbers before these PRs:
julia> methods(hcat) |> length # Base & LinearAlgebra
23
julia> using SparseArrays
julia> methods(hcat) |> length
29
julia> using AxisKeys # and NamedDims
julia> methods(hcat) |> length
42
There was a problem hiding this comment.
Afaict, it is needed. Note that there's no cartesian product, how do you calculate 216 and 4096?
There was a problem hiding this comment.
Oh I'm sorry, you are right. I thought there were 3 nested loops, but clearly there are not.
CI should hopefully pass now!
[ Edit: closes #216 right? ]