Problem with hilbert() and Makie

New to Julia
,
1

Hi, I’m not sure whether this is a Makie or a DSP problem, but here we go:

(jl_SlnPN9) pkg> st
Status `/private/var/folders/gt/z8fycpfs2z7c_mb18zvhjrf00000gp/T/jl_SlnPN9/Project.toml`
  [717857b8] DSP v0.8.7
  [ee78f7c6] Makie v0.24.17

julia> using Makie, DSP
julia> randn(100) |> hilbert
ERROR: MethodError: plan_rfft(::Vector{Float64}, ::Tuple{Int64}) is ambiguous.

Candidates:
  plan_rfft(X::StridedArray{Float64, N}, region; flags, timelimit, num_threads) where N
    @ FFTW ~/.julia/packages/FFTW/uNgRy/src/fft.jl:846
  plan_rfft(x::AbstractArray{T, N}, region::Union{Int64, NTuple{N, Int64}, AbstractVector{Int64}} where N; BLUESTEIN_CUTOFF, _kwargs...) where {T<:Real, N}
    @ FFTA ~/.julia/packages/FFTA/bnEzc/src/plan.jl:102

Possible fix, define
  plan_rfft(::StridedArray{Float64, N}, ::Union{Int64, NTuple{N, Int64} where N, AbstractVector{Int64}}) where N

Stacktrace:
 [1] plan_rfft(X::Vector{Float64}; kws::@Kwargs{})
   @ FFTW ~/.julia/packages/FFTW/uNgRy/src/fft.jl:887
 [2] hilbert(x::Vector{Float64})
   @ DSP.Util ~/.julia/packages/DSP/XtGMj/src/util.jl:36
 [3] |>(x::Vector{Float64}, f::typeof(hilbert))
   @ Base operators.jl:1001
 [4] top-level scope
   @ REPL[13]:1
 [5] top-level scope
   @ REPL[13]:1

julia> versioninfo()
Julia Version 1.14.0-alpha2
Build Info:
  Official https://julialang.org release
  Commit 22d5d63fa18 (2026-10-26 06:08 UTC)
  GC: Built with stock GC
Platform Info:
  OS: macOS (arm64-apple-darwin25.0.0)
  CPU: 10 × Apple M2 Pro (apple-m2)
  WORD_SIZE: 64
  LLVM: libLLVM-23.1.4 (ORCJIT, apple-m2)
Threads: 10 default, 1 interactive, 10 GC (on 10 virtual cores)
Environment:
  JULIA_DEPOT_PATH = /Users/maarhart/.julia
  JULIA_EDITOR = emacsclient

I have the same issue in the latest 1.13.

2

(post deleted by author)

3

Looks like it is a problem with neither, but rather FFTA.jl and FFTW.jl conflicting (see link).

I would recommend uninstalling one of them if possible, since complications in selecting the right FFT backend is a known issue with AbstractFFTs.jl:

If uninstalling is not an option, you can monkey-patch the shown signature of plan_rfft to use the correct backend.

4

My understanding is that KernelDensity in Makie uses FFTA and FFTW is in DSP.

5

Yeah, my bad. Looks like KernelDensity.jl switched to FFTA in v0.6.11, so downgrading to v0.6.10 can solve the problem.