Construction of ComponentArray inside of AD/Zygote

I want to compute the gradient of a loss function with respect to a ComponentArray. In the loss function, I need to reconstruct a ComponentArray.
Based on @jonniedie reply https://github.com/jonniedie/ComponentArrays.jl/issues/126#issuecomment-1141580528, I tried

function my_sum(v)
    ax = getaxes(v)
    @unpack x, y = v
    ca = ComponentArray([x..., y...], ax)
    return sum(ca.x + ca.y)
end

Zygote.gradient(my_sum, ComponentArray(x=[0.0], y=[0.0]))

which fails with

ERROR: ArgumentError: indexed assignment with a single value to possibly many locations is not supported; perhaps use broadcasting `.=` instead?
Stacktrace:
  [1] setindex_shape_check(::ChainRulesCore.Tangent{Any, Tuple{Float64}}, ::Int64)
    @ Base ./indices.jl:261
  [2] _unsafe_setindex!(#unused#::IndexLinear, A::Vector{Float64}, x::ChainRulesCore.Tangent{Any, Tuple{Float64}}, I::UnitRange{Int64})
    @ Base ./multidimensional.jl:939
  [3] _setindex!
    @ ./multidimensional.jl:930 [inlined]
  [4] setindex!
    @ ./abstractarray.jl:1344 [inlined]
  [5] macro expansion
    @ ~/.julia/packages/ComponentArrays/EjZNJ/src/array_interface.jl:0 [inlined]
  [6] _setindex!(x::ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(x = 1:1, y = 2:2)}}}, v::ChainRulesCore.Tangent{Any, Tuple{Float64}}, idx::Val{:y})
    @ ComponentArrays ~/.julia/packages/ComponentArrays/EjZNJ/src/array_interface.jl:129
  [7] setproperty!
    @ ~/.julia/packages/ComponentArrays/EjZNJ/src/namedtuple_interface.jl:17 [inlined]
  [8] (::ComponentArrays.var"#getproperty_adjoint#87"{ComponentVector{Float64, Vector{Float64}, Tuple{Axis{(x = 1:1, y = 2:2)}}}, Symbol})(Δ::ChainRulesCore.Tangent{Any, Tuple{Float64}})
    @ ComponentArrays ~/.julia/packages/ComponentArrays/EjZNJ/src/compat/chainrulescore.jl:4
  [9] ZBack
    @ ~/.julia/packages/Zygote/PD12J/src/compiler/chainrules.jl:206 [inlined]
 [10] Pullback
    @ ~/.julia/packages/UnPack/EkESO/src/UnPack.jl:34 [inlined]
 [11] (::typeof(∂(unpack)))(Δ::Tuple{Float64})
    @ Zygote ~/.julia/packages/Zygote/PD12J/src/compiler/interface2.jl:0
 [12] macro expansion
    @ ~/.julia/packages/UnPack/EkESO/src/UnPack.jl:101 [inlined]
 [13] Pullback

pointing to the @unpack call. @avik-pal noted that it also happens even without the @unpack

function my_sum(v)
    ax = getaxes(v)
    ca = ComponentArray([v.x..., v.y...], ax)
    return sum(ca.x + ca.y)
end

Zygote.gradient(my_sum, ComponentArray(x=[0.0], y=[0.0]))

but is resolved by using vcat

function my_sum(v)
    ax = getaxes(v)
    @unpack x, y = v
    ca = ComponentArray(vcat(x,y), ax)
    return sum(ca.x + ca.y)
end

The issue seems to be that \Delta is a Tuple{Float64} in
https://github.com/jonniedie/ComponentArrays.jl/blob/cbb24ef7156d18f1576ea48d7ae42023cc5bfa70/src/compat/chainrulescore.jl#L4
for splatting.

Start with src/compat/chainrulescore.jl at the referenced line and reproduce the failure using the two my_sum examples with Zygote.gradient. Compare the splatted array construction with the working vcat form and inspect how the Tangent delta is handled. Done means reconstructing a ComponentArray inside the gradient path works without the indexed-assignment error, with a regression test for the example.