Convergence ABI: async Thenable-handle + thenableThen continuation protocol (AffineScript + Ephapax)

[hyperpolymath]

Tracks the async section of the typed-wasm convergence ABI, surfaced by AffineScript issue #225 / ADR-013 and needing Ephapax co-stakeholder review (typed-wasm ADR-004: typed-wasm is the shared aggregate/convergence layer for AffineScript and Ephapax).

What is being fixed as a shared convention

The async-over-wasm boundary protocol both languages must agree on:

• An async host import returns a Thenable handle (i32) synchronously; the host registers the Promise in the handle table.
• The guest registers a continuation closure via thenableThen(handle, closurePtr) using the closure ABI ([fnId@0, envPtr@4] dispatched through __indirect_function_table).
• The host re-enters the guest after settlement; thenableResultJson(handle) (or a typed reader) yields the settled value.
• An async guest function itself returns a Thenable handle, so the protocol composes up the call chain — this is what lets a backend present a transparent value-returning surface (AffineScript ADR-013's CPS transform) without JSPI.

Why it is convergence-layer, not AffineScript-internal

AffineScript ADR-013 (transparent CPS transform of Async functions on WasmGC) is the consumer; the protocol/handle conventions above are the shared contract. Ephapax compiles to the same typed WasmGC and must interoperate on identical conventions. This issue does not couple AffineScript to typed-wasm nor constrain Ephapax — it fixes the shared async convention where both already converge.

Asks

• Ephapax co-stakeholder review of the protocol (closure ABI shape, settlement re-entry, handle lifetime/once-settle guarantee, error/reject shape { __error })
• Record the agreed protocol as a typed-wasm convergence-ABI spec section
• Confirm no divergence vs Ephapax's existing async lowering (or note required reconciliation)

Refs: affinescript#225, affinescript#160, affinescript ADR-013 (docs/specs/async-on-wasm-cps.adoc), affinescript #199 (closure ABI), affinescript #205 (Thenable resolution), typed-wasm ADR-004.

[hyperpolymath]

Ephapax convergence-ABI review — complete (PR #32)

Reviewed the AffineScript-proven async-over-wasm protocol against Ephapax's current wasm + effect lowering.

Conclusion: NO DIVERGENCE. Adoption is purely additive.

Ask	Result
Co-stakeholder review of the protocol	✅ reviewed vs ephapax-wasm + effects (ADR-007..ADR-010)
Record as typed-wasm convergence-ABI spec section	✅ spec/async-convergence-abi.adoc + ADR-005 (PR #32)
Confirm no divergence vs Ephapax async lowering	✅ none — see below

Findings

• Ephapax has no async lowering yet (perform/handle type-checked but emit unreachable) → nothing to conflict with; the protocol is adopt-when-implemented.
• Closure ABI already converges byte-identically: Ephapax [table_idx@0, env_ptr@4] (8-byte cell, call_indirect (env_ptr,param)->i32) == AffineScript #199 [fnId@0, envPtr@4] via __indirect_function_table. The continuation mechanism the protocol needs is a match on both sides with zero reconciliation.
• Once-settle: Ephapax's planned one-shot resume(once) for linear-capturing handlers aligns exactly with the Thenable once-settle / fire-at-most-once guarantee.
• One representational point, reconciled (not a blocker): the reject shape. AffineScript uses a host-JSON { "__error" } envelope; Ephapax models errors as native sum types (inl/inr). Resolution recorded in the spec: {__error} is the host-boundary envelope only; each language maps it to its native error type guest-side. No on-the-wire divergence; neither language's error model is constrained.

ADR-005 is proposed, pending explicit co-stakeholder sign-off. On agreement it becomes accepted and AffineScript #225 PR 2–4 (the CPS transform proper) are unblocked. Leaving this issue open for sign-off.

[hyperpolymath]

Resolved by PR #32 (merged to main). The 3 asks are complete and now canonical: (1) shared async boundary protocol fixed in spec/async-convergence-abi.adoc; (2) Ephapax co-stakeholder review recorded (no divergence — closure cell byte-identical, {__error} boundary-only); (3) ADR-005 accepted, amended 2026-05-19 to ratify the accessor-model continuation signature (env_ptr)->i32 (the (env_ptr, settled) ambiguity removed) + a Conformance — AffineScript marry-up section added. Ephapax's actual async lowering is separate future work and adopts this contract when implemented (not tracked by this issue). Closing per owner sign-off.