onTakeOrders2 fires when the taker received zero tokens, contradicting the IRaindexV6OrderTaker NatSpec

[thedavidmeister]

Summary

takeOrders4 gates the onTakeOrders2 callback only on config.data.length > 0 — not on whether any order actually filled. The IRaindexV6OrderTaker NatSpec documents the callback as firing "if it caused a non-zero input amount. I.e. if the order(s) taker received some tokens." When non-empty data is passed but every order is dead/skipped (so totalTakerInput == 0), the callback still fires (with inputAmountSent == 0).

Observed (verified on current main)

Repro testCallbackDoesNotFireOnZeroInput: a single never-added (dead) order is taken with data = 0x01; totalTakerInput == 0 (asserted as a precondition). The test asserts — per the interface NatSpec — that the callback does not fire. It fails on main: the callback fired with inputAmountSent == 0. pushTokens(.., 0) is a no-op so no tokens move, but the external call to onTakeOrders2 still happens.

RaindexV6.sol gates the call at roughly L587–590; the NatSpec is in IRaindexV6OrderTaker (raindex-interface).

Why it matters / triage

A code-vs-documented-contract divergence — ambiguous which side is canonical:

• If the code is the bug: an unexpected external call to msg.sender mid-takeOrders that an integrator relying on the documented "you received tokens" contract would not anticipate.
• If the NatSpec is stale: the interface doc should be corrected to match.

Either way, no existing test pins this gating in either direction (a coverage gap). Adjacent to I02 (#2625), which implicitly assumes the callback corresponds to a real settlement.

Provenance

Surfaced and verified by an adversarial review pass over current main. Repro on a throwaway scratch branch (uncommitted).

[linear]

RAI-844

[thedavidmeister]

Re-confirmed on 08d547fab (blind sweep, 3 independent repros): onTakeOrders2 still fires on a zero-input / zero-fill batch (gate is only config.data.length > 0, no totalTakerInput > 0 check), contradicting the IRaindexV6OrderTaker NatSpec ("if it caused a non-zero input amount").

[thedavidmeister]

Resolved by #2690 (merged) — commit ff61713 ("track order take-vs-skip with an explicit bool", which references #2648) gates the onTakeOrders2 callback on whether an order was actually taken vs skipped, so it no longer fires when the taker received zero tokens. Fix is on main; closing as completed.