Fix batch-forward typeassert: Enzyme returns AnonymousStruct, not NTuple

[ChrisRackauckas-Claude]

⚠️ Please ignore this PR until reviewed by @ChrisRackauckas.

Summary

Fixes the long-failing Enzyme batch forward mode (width > 1) testset on main. The failure existed before issue #48 and is unrelated to it — surfaced while running the test suite during #48 investigation.

The forward rules for batch width > 1 asserted:

shadow_result = Enzyme.autodiff(mode, Const(f_orig), BatchDuplicated{T, W}, args...)
return shadow_result[1]::NTuple{W, T}

But Enzyme's batch shadow comes back wrapped in Enzyme.Compiler.AnonymousStruct, which is defined (Enzyme/src/compiler/utils.jl:480) as:

@inline AnonymousStruct(::Type{U}) where {U<:Tuple} =
    NamedTuple{ntuple(Symbol, Val(length(U.parameters))), U}

So the runtime type is @NamedTuple{1::Float64, 2::Float64}, not Tuple{Float64, Float64} — the typeassert fires with:

TypeError: in typeassert, expected Tuple{Float64, Float64},
got a value of type @NamedTuple{1::Float64, 2::Float64}

Fix

Wrap with Tuple(...) before the typeassert in both batch-mode rule branches:

return Tuple(shadow_result[1])::NTuple{W, T}      # shadow-only
shadows = Tuple(shadow_result[1])::NTuple{W, T}   # ForwardWithPrimal

Tuple(::NamedTuple) strips the field-name wrapper and yields a plain NTuple{W, T} (the underlying storage), which is what the BatchDuplicated shadow contract expects.

Bisect

Not an Enzyme regression. Tested Enzyme 0.13.100 → 0.13.148 — every version returns the AnonymousStruct/NamedTuple shape. The FWW rule's assertion was wrong from the moment it was written (introduced in PR #45 / v1.8.0).

Smallest reproducer

using FunctionWrappersWrappers, Enzyme, EnzymeCore
f(x) = x^2
fww = FunctionWrappersWrapper(f, (Tuple{Float64},), (Float64,))
Enzyme.autodiff(Forward, Const(fww), BatchDuplicated, BatchDuplicated(3.0, (1.0, 2.0)))
# Before: TypeError in typeassert
# After:  ((var"1" = 6.0, var"2" = 12.0),)

Tests

The existing Enzyme batch forward mode (width > 1) testset (test/enzyme_tests.jl:55) — which was the failing one — now passes. Full local suite on Julia 1.11 + Enzyme v0.13.147:

Testset	Pass	Total
FunctionWrappersWrappers.jl	48	48
BigFloat support	5	5
UnionAll return types	2	2
Enzyme extension	44	44
Mooncake extension	13	13

Enzyme extension was previously 39 pass / 1 error.

Relationship to #49

Independent of PR #49 (the issue #48 fix). This PR branches off main; #49 should rebase cleanly on top of (or under) this one. Could be merged in either order.

Test plan

• Failing Enzyme batch forward mode (width > 1) testset passes locally
• No other testsets regressed (44 / 44 in Enzyme extension)
• Bisected against Enzyme 0.13.100 → 0.13.148 — same NamedTuple return shape, so fix is forward-compatible
• CI green on all jobs

🤖 Generated with Claude Code