Illegal replace ficticious phi for Dicts with different key types

[penelopeysm]

The following errors on both 1.10 and 1.11 with Illegal replace ficticious phi for: %unbox37.unpack77.unpack_replacementA = phi {} addrspace(10)* , !dbg !62 of %unbox37.unpack77.unpack = load {} addrspace(10)*, {} addrspace(10)* addrspace(11)* %unbox37.unpack77.elt, align 8, !dbg !81, !enzyme_type !86, !enzyme_inactive !0, !enzymejl_byref_MUT_REF !0, !enzymejl_source_type_Dict\7BVN\2C\20Int64\7D !0.

Only reverse-mode errors; forward-mode is fine on both versions.

import Enzyme: Enzyme, Forward, Reverse, set_runtime_activity, Const

struct VN end
vnm = VN()
struct MyMD{
    TIdcs<:Dict{<:Any,Int},
    TVal<:AbstractVector{<:Real},
}
    idcs::TIdcs
    ranges::Vector{UnitRange{Int}}
    vals::TVal
end
struct MyVI{S,M,A}
    s::S
    m::M
    a::A
end
function tass(vn, vi)
    md = vi.s
    i = md.idcs[vn]
    r = md.ranges[i]
    y = md.vals[r]
    x = only(y)
    return x, vi
end
function modelf(vi, x=[1.5, 2.0])
    s = Vector{Float64}(undef, length(x))
    for i in eachindex(s)
        s[i], vi = tass(i, vi)
    end
    j = vi.m.idcs[vnm]
    return vi
end
function g(a)
    return iterate([]) === nothing ? a : a
end
function f(x)
    md_s = MyMD(
        Dict(1 => 1, 2 => 2),
        [1:1, 2:2],
        x[1:2],
    )
    md_m = MyMD(
        Dict(vnm => 1),
        [1:2],
        x[3:4],
    )
    vi = MyVI(md_s, md_m, 0.0)
    new_vi = modelf(vi)
    return g(0.0)
end

x = fill(0.5, 4)
f(x)
Enzyme.gradient(set_runtime_activity(Reverse), f, x)

Full error: https://gist.github.com/penelopeysm/0b03d8670bf5f754278784ce72c84268

Versions and stuff:

(ppl) pkg> st
Status `~/ppl/Project.toml`
  [7da242da] Enzyme v0.13.75

julia> versioninfo()
Julia Version 1.11.6
Commit 9615af0f269 (2025-07-09 12:58 UTC)
Build Info:
  Official https://julialang.org/ release
Platform Info:
  OS: macOS (arm64-apple-darwin24.0.0)
  CPU: 10 × Apple M1 Pro
  WORD_SIZE: 64
  LLVM: libLLVM-16.0.6 (ORCJIT, apple-m1)
Threads: 1 default, 0 interactive, 1 GC (on 8 virtual cores)

The Turing model that errors looks like this: (cf. https://turinglang.org/ADTests/ - I reran the table today)

@model function demo_dot_assume_observe_index(
    x = [1.5, 2.0],
    ::Type{TV} = Vector{Float64},
) where {TV}
    s = TV(undef, length(x))
    s .~ InverseGamma(2, 3)
    m = TV(undef, length(x))
    m ~ product_distribution(Normal.(0, sqrt.(s)))
    for i in eachindex(x)
        x[i] ~ Normal(m[i], sqrt(s[i]))
    end
end

[wsmoses]

this will be fixed by the pending jll

[wsmoses]

the jll has landed on main, please reopen if error persists!