Axes for transposed OffsetArray vector incorrect due to LinearAlgebra method #1042
Labels
arrays
[a, r, r, a, y, s]
Comments
|
Looks ok to me? What was the result that you were expecting? julia> A = zeros(3); axes(A)
(Base.OneTo(3),)
julia> B = centered(A); axes(B) # shift midpoint to zero
(OffsetArrays.IdOffsetRange(values=-1:1, indices=-1:1),)
julia> C = B'; axes(C) # the new axis added by the transpose is not centered about zero
(Base.OneTo(1), OffsetArrays.IdOffsetRange(values=-1:1, indices=-1:1))
julia> D = A'; axes(D)
(Base.OneTo(1), Base.OneTo(3))
julia> E = centered(D); axes(E) # both axes are now centered about zero
(OffsetArrays.IdOffsetRange(values=0:0, indices=0:0), OffsetArrays.IdOffsetRange(values=-1:1, indices=-1:1)) |
|
With respect, I think my post is clear in what I expected: The new first dimension should inherit the offset of the parent. The error in Base though is assuming that the first axis of a transposed vector should always be 1-indexed. However, I see this is a multi-level issue. Because OffsetArrays assumes all later dimensions start at 1, and would just provide |
|
I apologize for my brusque tone, but this was not what I had intended. I'll make a PR to change the hard-coded axis to use that of the parent instead. |
|
No worries, and thanks! |
N5N3
referenced
this issue
in JuliaLang/julia
Dec 5, 2023
Close #52373, or at least the part that may be addressed here. After
this, the first axis for an `Adjoint(parent::AbstractVector)` will be
the second axis of the `parent`. This will change the type of the axis
where the parent array type specializes `axes(A, d)`. In the short term,
this would improve type-stability in cases such as
```julia
julia> A = OffsetArray([1,2], 2);
julia> @code_typed axes(A')[1]
CodeInfo(
1 ─ %1 = $(Expr(:boundscheck))::Bool
│ %2 = Base.getfield(t, i, %1)::OffsetArrays.IdOffsetRange{Int64, Base.OneTo{Int64}}
└── return %2
) => OffsetArrays.IdOffsetRange{Int64, Base.OneTo{Int64}}
```
where the result is now concretely inferred instead of being a small
`Union`.
In principle, with
JuliaArrays/StaticArrays.jl#916, this would make
the axes of the adjoint of a `StaticVector` statically sized.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Incorrect axes returned for a transposed OffsetArray vector. This causes incorrect image filtering results for kernels passed as vector components.
See: JuliaImages/ImageFiltering.jl#269
(lines based on 1.9 but still present in 1.10)
https://github.com/JuliaLang/julia/blame/8e5136fa2979885081cd502d2210633dff1d2a1a/stdlib/LinearAlgebra/src/adjtrans.jl#L298
The text was updated successfully, but these errors were encountered: