Repository navigation
seed! zero-seed methods removed in ForwardDiff v1.4.4, breaks threaded_gradient!/threaded_jacobian! #35
Description
ForwardDiff v1.4.4 dropped the zero-seed seed! methods (seed!(duals, x), seed!(duals, x, index)) in favor of seed_zero_partials!. PolyesterForwardDiff still calls the old forms in evaluate_chunks!, evaluate_jacobian_chunks!, and evaluate_f_and_jacobian_chunks!.
using PolyesterForwardDiff
using ForwardDiff
using Random
Random.seed!(20260805)
f(x) = sum(sin.(x) .+ x.^2)
x = rand(1000)
dx = similar(x)
chunk = ForwardDiff.Chunk(8)
PolyesterForwardDiff.threaded_gradient!(f, dx, x, chunk)ERROR: MethodError: no method matching seed!(::StrideArraysCore.StrideArray{ForwardDiff.Dual{Nothing, Float64, 8}, ...}, ::StrideArraysCore.PtrArray{Float64, ...})
Closest candidates are:
seed!(::AbstractArray{ForwardDiff.Dual{T, V, N}}, ::Any, !Matched::NTuple{N, ForwardDiff.Partials{N, V}}) where {T, V, N}
@ ForwardDiff ~/.julia/packages/ForwardDiff/IleGP/src/apiutils.jl:107
Versions: Julia 1.12.6, ForwardDiff 1.4.4, Polyester 0.7.19, PolyesterForwardDiff 0.1.3
Activity
ChrisRackauckas commentedon Aug 5, 2026
Fixed
dr-ko commentedon Aug 5, 2026
Thanks. Releasing the patch as well?
ChrisRackauckas commentedon Aug 5, 2026
Should already be done.
dr-ko commentedon Aug 5, 2026
Is there now, and works like a charm. Thank you for quick actions :-)