Issues & PRs in other repos

[gdalle]

As part of our unification efforts, we contributed to quite a lot of issues in other repos, sometimes with PRs to fix them. Let's try to keep track.

AbstractDifferentiation

• Comparison with DifferentiationInterface.jl AbstractDifferentiation.jl#131
• Handling of thunks and tangents AbstractDifferentiation.jl#132

ADTypes

• What is the right way to specify the Enzyme mode? SciML/ADTypes.jl#24
• Add AutoDiffractor SciML/ADTypes.jl#27
• Give AutoSparseFiniteDiff the same type parameters as AutoFiniteDiff SciML/ADTypes.jl#33
• Change handling of sparse backends SciML/ADTypes.jl#38
• Restructure sparse backends and replace subtyping by traits SciML/ADTypes.jl#40

ComponentArrays

• ComponentVectors are not stack-able SciML/ComponentArrays.jl#254

DiffResults

• Dependents who forget to re-alias results DiffResults.jl#27

Diffractor

• frule_via_ad tries to call NoTangent Diffractor.jl#277
• Error trying to differentiate eachindex Diffractor.jl#280
• Is there a Diffractor-native interface? Diffractor.jl#287

Enzyme

• Why does Duplicated(x, dx) assume x and dx have the same type? EnzymeAD/Enzyme.jl#1329
• StackOverflow in forward mode EnzymeAD/Enzyme.jl#1330
• Jacobian has wrong shape for matrix input EnzymeAD/Enzyme.jl#1332
• Why is autodiff_thunk useful in forward mode? EnzymeAD/Enzyme.jl#1335
• Add FAQ section to docs EnzymeAD/Enzyme.jl#1343
• Intimidating docs EnzymeAD/Enzyme.jl#1344
• Matrix exponential fails EnzymeAD/Enzyme.jl#1346
• Type analysis depth limit / task switch not allowed from inside staged nor pure functions EnzymeAD/Enzyme.jl#1354
• Julia 1.11 Unsupported EnzymeAD/Enzyme.jl#1358
• make_zero is type-unstable EnzymeAD/Enzyme.jl#1359
• Why does gradient in forward mode return a tuple? EnzymeAD/Enzyme.jl#1368
• Reverse mode jacobian is wrong for matrix input EnzymeAD/Enzyme.jl#1391
• Reverse mode is type-unstable in v0.12 EnzymeAD/Enzyme.jl#1401
• Reverse mode gradient is up to 100x slower in Enzyme v0.12 EnzymeAD/Enzyme.jl#1409
• Enzyme v0.12.17 fails the DifferentiationInterface test suite EnzymeAD/Enzyme.jl#1562
• Add DifferentiationInterface test suite EnzymeAD/Enzyme.jl#1563

FastDifferentiation

• Support for AbstractDifferentiation.jl / DifferentiationInterface.jl brianguenter/FastDifferentiation.jl#37
• Change make_function to allow for multiple vector input arguments brianguenter/FastDifferentiation.jl#39
• make_function returns wrong result with sparse Jacobian for very simple functions only brianguenter/FastDifferentiation.jl#67

FiniteDiff

• What does "multi point derivatives" mean? FiniteDiff.jl#181
• Add docstrings and update documentation FiniteDiff.jl#182
• Cannot reuse HessianCache for another value of x FiniteDiff.jl#185

FiniteDifferences

• README gives wrong signature for VJP FiniteDifferences.jl#234

ForwardDiff

• Document internals? ForwardDiff.jl#682
• DiffResults objects are not re-aliased properly ForwardDiff.jl#696
• gradient! allocates for matrices but not for vectors ForwardDiff.jl#698

MathOptSymbolicAD

• Use as an AD backend in DifferentiationInterface.jl? lanl-ansi/MathOptSymbolicAD.jl#30

ReverseDiff

• DiffResults objects are not re-aliased properly ReverseDiff.jl#251

SparseDiffTools

• Auto(Sparse)ReverseDiff not supported SparseDiffTools.jl#287

Tapir

• Document how to call rrule!! chalk-lab/Mooncake.jl#94
• Array output seems to be initialized but not filled chalk-lab/Mooncake.jl#97
• Why the dependence on BenchmarkTools? chalk-lab/Mooncake.jl#105
• Tapir.value_and_pullback!! and mutation chalk-lab/Mooncake.jl#113
• Split reverse mode for Tapir chalk-lab/Mooncake.jl#115

Tracker

• withgradient and gradient are incoherent for scalar input FluxML/Tracker.jl#165

Zygote

• withjacobian flattens the output when it is a matrix FluxML/Zygote.jl#1506

[gdalle]

Closing this because I no longer keep track