simplify package extension for post Julia 1.10

[yebai]

No description provided.

[Red-Portal]

This looks fine for me, but the failing tests are a bit annoying. Do you want me to fix the tests first before merging this PR (and the other one as well)?

[yebai]

Do you want me to fix the tests first before merging this PR (and the other one as well)?

Yes, please

[github-actions]

Benchmark Results

Benchmark suite	Current: 3b795fd	Previous: a683e96	Ratio
normal/RepGradELBO + STL/meanfield/Zygote	16473872136 ns	14636863236 ns	1.13
normal/RepGradELBO + STL/meanfield/ForwardDiff	3304125733.5 ns	3350086725 ns	0.99
normal/RepGradELBO + STL/meanfield/ReverseDiff	3094818311.5 ns	3091246711 ns	1.00
normal/RepGradELBO + STL/fullrank/Zygote	16338470403 ns	14488178420 ns	1.13
normal/RepGradELBO + STL/fullrank/ForwardDiff	3688626107 ns	3684779258 ns	1.00
normal/RepGradELBO + STL/fullrank/ReverseDiff	5646882456 ns	5742736189 ns	0.98
normal/RepGradELBO/meanfield/Zygote	7530702671 ns	6910985576 ns	1.09
normal/RepGradELBO/meanfield/ForwardDiff	2353602785 ns	2459214663 ns	0.96
normal/RepGradELBO/meanfield/ReverseDiff	1313395716.5 ns	1370548995.5 ns	0.96
normal/RepGradELBO/fullrank/Zygote	7682697941 ns	6937940281 ns	1.11
normal/RepGradELBO/fullrank/ForwardDiff	2635490263.5 ns	2708551001 ns	0.97
normal/RepGradELBO/fullrank/ReverseDiff	2436281408 ns	2497588532 ns	0.98
normal + bijector/RepGradELBO + STL/meanfield/Zygote	25565717237 ns	22847036335 ns	1.12
normal + bijector/RepGradELBO + STL/meanfield/ForwardDiff	10309343356 ns	10452617328 ns	0.99
normal + bijector/RepGradELBO + STL/meanfield/ReverseDiff	4476042343 ns	4494655977 ns	1.00
normal + bijector/RepGradELBO + STL/fullrank/Zygote	25818366598 ns	23169952741 ns	1.11
normal + bijector/RepGradELBO + STL/fullrank/ForwardDiff	10786917463 ns	10972906220 ns	0.98
normal + bijector/RepGradELBO + STL/fullrank/ReverseDiff	7641622123 ns	7716647322 ns	0.99
normal + bijector/RepGradELBO/meanfield/Zygote	15735398805 ns	14557557757 ns	1.08
normal + bijector/RepGradELBO/meanfield/ForwardDiff	9335421424 ns	9793310961 ns	0.95
normal + bijector/RepGradELBO/meanfield/ReverseDiff	2541208289 ns	2569890147.5 ns	0.99
normal + bijector/RepGradELBO/fullrank/Zygote	16236341205 ns	14601467359 ns	1.11
normal + bijector/RepGradELBO/fullrank/ForwardDiff	10048414269 ns	10039535578 ns	1.00
normal + bijector/RepGradELBO/fullrank/ReverseDiff	3948466684 ns	4028716220.5 ns	0.98

This comment was automatically generated by workflow using github-action-benchmark.

[yebai]

Do you want me to fix the tests first before merging this PR (and the other one as well)?

Yes, please

Removing AbstarctOperator fixes it, but it is not clear why since abstract type constraints on keyword arguments are supported:

julia> f(x::Int;y::Real) = x,y
f (generic function with 1 method)

julia> f(1;y=2)
(1, 2)

julia> f(1;y=2.)
(1, 2.0)

julia> f(1;y="ab")
ERROR: TypeError: in keyword argument y, expected Real, got a value of type String

For future reference, here is the error message. https://github.com/TuringLang/AdvancedVI.jl/actions/runs/13550036596/job/37871165993#step:2:1021

TypeError: in keyword argument operator, expected AdvancedVI.AbstractOperator, got a value of type ClipScale{Float64}
│    Stacktrace:
│     [1] optimize(::Main.__atexample__named__repgradelbo.NormalLogNormal{Float64, Float64, FillArrays.Fill{Float64, 1, Tuple{Base.OneTo{Int64}}}, LinearAlgebra.Diagonal{Float64, FillArrays.Fill{Float64, 1, Tuple{Base.OneTo{Int64}}}}}, ::RepGradELBO{ClosedFormEntropy}, ::Bijectors.MultivariateTransformed{MvLocationScale{LinearAlgebra.Diagonal{Float64, Vector{Float64}}, Distributions.Normal{Float64}, Vector{Float64}}, Bijectors.Stacked{Vector{Function}, Vector{UnitRange{Int64}}}}, ::Int64; kwargs::@Kwargs{show_progress::Bool, adtype::ADTypes.AutoForwardDiff{nothing, Nothing}, optimizer::Optimisers.Adam{Float64, Tuple{Float64, Float64}, Float64}, operator::ClipScale{Float64}, callback::typeof(Main.__atexample__named__repgradelbo.callback)})
│       @ AdvancedVI ~/work/AdvancedVI.jl/AdvancedVI.jl/src/optimize.jl:128