`increment_deriv!(::Float64, ::Float64)` MethodError with a vector of `Real`s

[penelopeysm]

Hello! I'm passing on this error originally reported at Turing.jl: TuringLang/Turing.jl#2364

Here's a simplified MWE that doesn't involve Turing at all. This example fails but when changing Real to Float64, it gives the correct derivative of 1.

using ReverseDiff: gradient

function f(u)
    x = (Real[1.0, 2.0] * u[], u[])
    return last(x)
end

ReverseDiff.gradient(f, [2.0])

Traceback

ERROR: MethodError: no method matching increment_deriv!(::Float64, ::Float64)

Closest candidates are:
  increment_deriv!(::ReverseDiff.TrackedReal, ::Real)
   @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/propagation.jl:45
  increment_deriv!(::AbstractArray, ::Any)
   @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/propagation.jl:38
  increment_deriv!(::ReverseDiff.TrackedArray, ::Real, ::Any)
   @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/propagation.jl:34
  ...

Stacktrace:
  [1] increment_deriv!
    @ ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/propagation.jl:36 [inlined]
  [2] broadcast_increment_deriv!(input::Vector{…}, x::Vector{…}, partial::Float64, input_bound::CartesianIndex{…}, ::Nothing)
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/propagation.jl:173
  [3] special_reverse_exec!(instruction::ReverseDiff.SpecialInstruction{…})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/derivatives/elementwise.jl:533
  [4] reverse_exec!(instruction::ReverseDiff.SpecialInstruction{Tuple{…}, Tuple{…}, ReverseDiff.TrackedArray{…}, Tuple{…}})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/tape.jl:93
  [5] reverse_pass!(tape::Vector{ReverseDiff.AbstractInstruction})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/tape.jl:87
  [6] reverse_pass!
    @ ~/.julia/packages/ReverseDiff/p1MzG/src/api/tape.jl:36 [inlined]
  [7] seeded_reverse_pass!(result::Vector{…}, output::ReverseDiff.TrackedReal{…}, input::ReverseDiff.TrackedArray{…}, tape::ReverseDiff.GradientTape{…})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/utils.jl:31
  [8] seeded_reverse_pass!(result::Vector{…}, t::ReverseDiff.GradientTape{…})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/tape.jl:47
  [9] gradient(f::Function, input::Vector{Float64}, cfg::ReverseDiff.GradientConfig{ReverseDiff.TrackedArray{…}})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/gradients.jl:24
 [10] gradient(f::Function, input::Vector{Float64})
    @ ReverseDiff ~/.julia/packages/ReverseDiff/p1MzG/src/api/gradients.jl:22
 [11] top-level scope
    @ REPL[4]:1
Some type information was truncated. Use `show(err)` to see complete types.

Version info

The error above occurs with a fresh environment containing only ReverseDiff@v1.15.3.

Julia Version 1.10.5
Commit 6f3fdf7b362 (2024-08-27 14:19 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin22.4.0)
  CPU: 10 × Apple M1 Pro
  WORD_SIZE: 64
  LIBM: libopenlibm
  LLVM: libLLVM-15.0.7 (ORCJIT, apple-m1)
Threads: 1 default, 0 interactive, 1 GC (on 8 virtual cores)

Failing Turing model

The below is the simplest Turing model I could get to yield the same error. I'm including it here just in case I overly simplified the MWE above.

using Turing

@model function f(x)
    u ~ Uniform(0, 1)
    return x * u
end

# works with Float64, as above
sample(f(Real[1.0, 2.0]), NUTS(; adtype=AutoReverseDiff()), 10)