Weird error when loading DifferentiableFrankWolfe extension

[BatyLeo]

There is a weird error when loading the DifferentiableFrankWolfeExt, that does not seem to affect InferOpt other than displaying an error message.

To reproduce:

using InferOpt
using DifferentiableFrankWolfe

outputs the following error:

Precompiling ImplicitDifferentiationForwardDiffExt...
Info Given ImplicitDifferentiationForwardDiffExt was explicitly requested, output will be shown live
ERROR: LoadError: MethodError: no method matching AbstractDifferentiation.ForwardDiffBackend()
The type `AbstractDifferentiation.ForwardDiffBackend` exists, but no method is defined for this combination of argument types when trying to construct it.
Stacktrace:
  [1] forward_conditions_backend(::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:62
  [2] (::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})(::Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}; kwargs::@Kwargs{})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:37
  [3] (::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})(::Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:27
  [4] vector_mode_dual_eval!(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}}, x::Vector{Float64})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/apiutils.jl:24
  [5] vector_mode_jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:129
  [6] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}}, ::Val{true})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:22
  [7] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:19
  [8] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:19
  [9] macro expansion
    @ ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:78 [inlined]
 [10] macro expansion
    @ ~/.julia/packages/PrecompileTools/L8A3n/src/workloads.jl:78 [inlined]
 [11] top-level scope
    @ ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:71
 [12] include
    @ ./Base.jl:557 [inlined]
 [13] include_package_for_output(pkg::Base.PkgId, input::String, depot_path::Vector{String}, dl_load_path::Vector{String}, load_path::Vector{String}, concrete_deps::Vector{Pair{Base.PkgId, UInt128}}, source::Nothing)
    @ Base ./loading.jl:2790
 [14] top-level scope
    @ stdin:5
in expression starting at /Users/taleboy/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:1
in expression starting at stdin:5
  ✗ ImplicitDifferentiation → ImplicitDifferentiationForwardDiffExt
  0 dependencies successfully precompiled in 3 seconds. 65 already precompiled.
[ Info: Precompiling ImplicitDifferentiationForwardDiffExt [5d5c0af3-0a85-5199-b335-e133c52d08ae] (cache misses: wrong dep version loaded (4), dep missing source (2), mismatched flags (4))
ERROR: LoadError: MethodError: no method matching AbstractDifferentiation.ForwardDiffBackend()
The type `AbstractDifferentiation.ForwardDiffBackend` exists, but no method is defined for this combination of argument types when trying to construct it.
Stacktrace:
  [1] forward_conditions_backend(::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:62
  [2] (::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})(::Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}; kwargs::@Kwargs{})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:37
  [3] (::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing})(::Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}})
    @ ImplicitDifferentiationForwardDiffExt ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:27
  [4] vector_mode_dual_eval!(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}}, x::Vector{Float64})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/apiutils.jl:24
  [5] vector_mode_jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:129
  [6] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}}, ::Val{true})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:22
  [7] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64}, cfg::ForwardDiff.JacobianConfig{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2, Vector{ForwardDiff.Dual{ForwardDiff.Tag{ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, Float64}, Float64, 2}}})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:19
  [8] jacobian(f::ImplicitDifferentiation.ImplicitFunction{ImplicitDifferentiationForwardDiffExt.var"#forward#8", ImplicitDifferentiationForwardDiffExt.var"#conditions#9", ImplicitDifferentiation.DirectLinearSolver, Nothing}, x::Vector{Float64})
    @ ForwardDiff ~/.julia/packages/ForwardDiff/UBbGT/src/jacobian.jl:19
  [9] macro expansion
    @ ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:78 [inlined]
 [10] macro expansion
    @ ~/.julia/packages/PrecompileTools/L8A3n/src/workloads.jl:78 [inlined]
 [11] top-level scope
    @ ~/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:71
 [12] include
    @ ./Base.jl:557 [inlined]
 [13] include_package_for_output(pkg::Base.PkgId, input::String, depot_path::Vector{String}, dl_load_path::Vector{String}, load_path::Vector{String}, concrete_deps::Vector{Pair{Base.PkgId, UInt128}}, source::Nothing)
    @ Base ./loading.jl:2790
 [14] top-level scope
    @ stdin:5
in expression starting at /Users/taleboy/.julia/packages/ImplicitDifferentiation/gOQIS/ext/ImplicitDifferentiationForwardDiffExt.jl:1
in expression starting at stdin:5
┌ Error: Error during loading of extension ImplicitDifferentiationForwardDiffExt of ImplicitDifferentiation, use `Base.retry_load_extensions()` to retry.
│   exception =
│    1-element ExceptionStack:
│    Failed to precompile ImplicitDifferentiationForwardDiffExt [5d5c0af3-0a85-5199-b335-e133c52d08ae] to "/Users/taleboy/.julia/compiled/v1.11/ImplicitDifferentiationForwardDiffExt/jl_hlfsLf".
│    Stacktrace:
│      [1] error(s::String)
│        @ Base ./error.jl:35
│      [2] compilecache(pkg::Base.PkgId, path::String, internal_stderr::IO, internal_stdout::IO, keep_loaded_modules::Bool; flags::Cmd, cacheflags::Base.CacheFlags, reasons::Dict{String, Int64}, isext::Bool)
│        @ Base ./loading.jl:3085
│      [3] (::Base.var"#1082#1083"{Base.PkgId})()
│        @ Base ./loading.jl:2492
│      [4] mkpidlock(f::Base.var"#1082#1083"{Base.PkgId}, at::String, pid::Int32; kwopts::@Kwargs{stale_age::Int64, wait::Bool})
│        @ FileWatching.Pidfile /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/FileWatching/src/pidfile.jl:95
│      [5] #mkpidlock#6
│        @ /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/FileWatching/src/pidfile.jl:90 [inlined]
│      [6] trymkpidlock(::Function, ::Vararg{Any}; kwargs::@Kwargs{stale_age::Int64})
│        @ FileWatching.Pidfile /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/FileWatching/src/pidfile.jl:116
│      [7] #invokelatest#2
│        @ ./essentials.jl:1057 [inlined]
│      [8] invokelatest
│        @ ./essentials.jl:1052 [inlined]
│      [9] maybe_cachefile_lock(f::Base.var"#1082#1083"{Base.PkgId}, pkg::Base.PkgId, srcpath::String; stale_age::Int64)
│        @ Base ./loading.jl:3609
│     [10] maybe_cachefile_lock
│        @ ./loading.jl:3606 [inlined]
│     [11] _require(pkg::Base.PkgId, env::Nothing)
│        @ Base ./loading.jl:2488
│     [12] __require_prelocked(uuidkey::Base.PkgId, env::Nothing)
│        @ Base ./loading.jl:2315
│     [13] #invoke_in_world#3
│        @ ./essentials.jl:1089 [inlined]
│     [14] invoke_in_world
│        @ ./essentials.jl:1086 [inlined]
│     [15] _require_prelocked
│        @ ./loading.jl:2302 [inlined]
│     [16] _require_prelocked
│        @ ./loading.jl:2301 [inlined]
│     [17] run_extension_callbacks(extid::Base.ExtensionId)
│        @ Base ./loading.jl:1500
│     [18] #1068
│        @ ./loading.jl:1557 [inlined]
│     [19] filter!(f::Base.var"#1068#1069", a::Vector{Base.ExtensionId})
│        @ Base ./array.jl:2910
│     [20] macro expansion
│        @ ./loading.jl:1556 [inlined]
│     [21] macro expansion
│        @ ./lock.jl:273 [inlined]
│     [22] retry_load_extensions()
│        @ Base ./loading.jl:1551
│     [23] top-level scope
│        @ REPL[16]:1
│     [24] eval
│        @ ./boot.jl:430 [inlined]
│     [25] eval_user_input(ast::Any, backend::REPL.REPLBackend, mod::Module)
│        @ REPL /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/REPL/src/REPL.jl:245
│     [26] repl_backend_loop(backend::REPL.REPLBackend, get_module::Function)
│        @ REPL /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/REPL/src/REPL.jl:342
│     [27] start_repl_backend(backend::REPL.REPLBackend, consumer::Any; get_module::Function)
│        @ REPL /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/REPL/src/REPL.jl:327
│     [28] run_repl(repl::REPL.AbstractREPL, consumer::Any; backend_on_current_task::Bool, backend::Any)
│        @ REPL /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/REPL/src/REPL.jl:483
│     [29] run_repl(repl::REPL.AbstractREPL, consumer::Any)
│        @ REPL /Applications/Julia-1.11.app/Contents/Resources/julia/share/julia/stdlib/v1.11/REPL/src/REPL.jl:469
│     [30] (::Base.var"#1139#1141"{Bool, Symbol, Bool})(REPL::Module)
│        @ Base ./client.jl:446
│     [31] #invokelatest#2
│        @ ./essentials.jl:1055 [inlined]
│     [32] invokelatest
│        @ ./essentials.jl:1052 [inlined]
│     [33] run_main_repl(interactive::Bool, quiet::Bool, banner::Symbol, history_file::Bool, color_set::Bool)
│        @ Base ./client.jl:430
│     [34] repl_main
│        @ ./client.jl:567 [inlined]
│     [35] _start()
│        @ Base ./client.jl:541
└ @ Base loading.jl:1506

then the extension seems to still work, at least tests still succeed.
It seems the extension of a downstream package (ImplicitDifferentiationForwardDiffExt) fails but does not directly affect InferOpt, although it displays a terrifying error message.
Seems related to the interaction between ImplicitDifferentiation and the new AbstractDifferentiation.
There is also a similar error (from ImplicitDifferentiationZygoteExt) when loading Zygote afterwards.

@gdalle do you have any insights ?

[gdalle]

That's because DifferentiableFrankWolfe v0.3, which uses a new version of ImplicitDifferentiation (switching from AbstractDifferentiation to DifferentiationInterface) was ready but not released. I triggered registration, once it's done InferOpt is due for a compat bump

[BatyLeo]

I tried updating to DifferentiableFrankWolfe v0.3.
It seems to solve the issue, however IterativeLinearSolver is no longer reexported from ImplicitDifferentiation.jl, and currently needed in InferOpt here:

InferOpt.jl/ext/InferOptFrankWolfeExt.jl

Line 47 in 9419111

implicit_kwargs = (; linear_solver=IterativeLinearSolver(; accept_inconsistent=true))

Same for LinearMinimizationOracle from FrankWolfe.jl.

[gdalle]

The default solver for ImplicitDifferentiation.jl is now called KrylovLinearSolver. As for the LMO, it was bad practice re-exporting it, we should get it directly from FrankWolfe.jl

[BatyLeo]

Should we then add FrankWolfe and ImplicitDifferentiation as (weak)dependencies, how does this work?

[gdalle]

Just make a list of packages in the extensions section of the Project instead of a single name.

InferOptFrankWolfeExt = ["DifferentiableFrankWolfe", "FrankWolfe", "ImplicitDifferentiation"]

A bit more cumbersome for users, but hey, there are no users

[BatyLeo]

Ok, I tried to update all this, I have an issue in the paths tests.
It seems that something fails when $\theta$ is a matrix (no method matching (::ImplicitDifferentiation.ImplicitFunction)(Matrix{Float32}, ...), closest candidates are (::ImplicitDifferentiation.ImplicitFunction)(!Matched::AbstractVector, Any...; kwargs...) and (::ImplicitDifferentiation.ImplicitFunction)(!Matched::AbstractArray{ForwardDiff.Dual{T, R, N}, 1}, Any...; kwargs...) where {T, R, N}).
The same tests on ranking with $\theta$ a vector seem to work.
Not sure if something is missing from ImplicitDifferentiation or if this can be fixed on our side.

[gdalle]

Yes, the new version of ImplicitDifferentiation only accepts vectors, since higher-order tensors are flattened into vectors anyway. So we need a manual resize on the InferOpt side

[BatyLeo]

I made a temporary fix that reshapes \theta arrays when computing stuff in RegularizedFrankWolfe, I'll revert it when ImplicitDifferentiation supports matrices.

However, in the Pushforward tests, there are a lot of Warning: Failed to solve the linear system in the implicit function theorem with Krylov.gmres warnings with a found approximate least-squares solution status.
Do you know what could cause these warnings and if we can do something to adress them?

[gdalle]

Unfortunately that's the hard part about implicit differentiation:

1. Maybe the Jacobian of the optimality conditions is not invertible and thus the implicit function theorem becomes invalid, so least-squares solutions are the best we can do (they can be considered a heuristic). In this case you can silence the warnings with

DiffFW(...; implicit_kwargs=(;
    linear_solver=ImplicitDifferentiation.KrylovLinearSolver(; verbose=false))
)

3. Maybe the linear solver needs different parameters, like a higher precision because the Jacobian is ill-conditioned. You can silence the warnings or change the linear solver by using

DiffFW(...; implicit_kwargs=(;
    linear_solver=(A, b) -> Krylov.gmres(A, b, anyparamsyouwant)  # to increase precision
)
DiffFW(...; implicit_kwargs=(;
    linear_solver=\  # to switch to a dense solver with dense matrices, useful in small dimension to debug
)

[BatyLeo]

Ok, I put ImplicitDifferentiation.KrylovLinearSolver(; verbose=false)) as the default linear solver and added an option to RegularizedFrankWolfe so that the user can change the linear solver if they want to.

[gdalle]

The latest version of DiffFW accepts arbitrary arrays @BatyLeo