Generic log0-log4 emitter in Solidity assembly fails from confusing calldata .offset with .length; use calldataload(topics.offset)

Goal: a privileged emitGenericEvent(bytes32[] calldata topics, bytes calldata logData) so that a satellite controller contract can make the canonical contract emit arbitrary events (e.g. an ERC-2981 metadata update) without hardcoding each event.

Symptom: hand-written (or LLM-written) assembly that does let logDataLength := calldataload(logData.offset) and then calldatacopy(topicsPtr, add(topics.offset, 32), mul(topicsLength, 32)) produces garbage or reverts.

Root cause: confusion between the two members Solidity exposes for a calldata slice. topics.offset is already a pointer to the *start of the element data*, and topics.length is the length (element count for arrays, byte count for bytes). The ABI's length word lives before that offset, not at it. So calldataload(logData.offset) reads the first 32 bytes of the payload as if it were a length, and add(topics.offset, 32) skips the first real topic.

Working shape — use .length directly, copy from .offset with no adjustment, and switch on the topic count because logN opcodes take a fixed number of topics:

let dataPtr := mload(0x40) let logDataLength := logData.length calldatacopy(dataPtr, logData.offset, logDataLength) mstore(0x40, add(dataPtr, logDataLength)) switch topics.length case 0 { log0(dataPtr, logDataLength) } case 1 { log1(dataPtr, logDataLength, calldataload(topics.offset)) } case 2 { log2(dataPtr, logDataLength, calldataload(topics.offset), calldataload(add(topics.offset, 32))) } // ... through case 4 with log4

Two secondary points: topics never need to be copied to memory at all — read them with calldataload — and there is no variadic log, so the switch over 0..4 is unavoidable. Keep a require(topics.length <= 4) guard.

Source: adapted from a Stack Exchange answer by 0xSanson (question by okwme), https://ethereum.stackexchange.com/questions/164188/is-it-possible-to-emit-generic-events-using-assembly — CC BY-SA 4.0