This spec is a work-in-progress

Standard events with enough information for Indexers to reconstruct the state

When a record or a single field is edited, or when a record is deleted, Store emits a standard event with enough information for off-chain actors to reconstruct the new version of the Store with even-sourcing.

event StoreSetRecord(uint256 tableId, bytes32[] key, bytes data);
event StoreSetField(uint256 tableId, bytes32[] key, uint8 schemaIndex, bytes data);
event StoreDeleteRecord(uint256 tableId, bytes32[] key);

Each event includes the table ID and each key encoded as bytes32.

MUD comes with many libraries and services in order to reconstruct the state of any Store with typed columns and keys:

1. Indexer is a service that mirrors the state of a Store in a database, and keeping it update to that with milliseconds of latency.
2. The MUD Networking stack can reconstruct the state of Store in the browser using either JSON-RPC or an Indexer.

Data encoding

bytes data uses custom encoding, which MUD networking stack decodes. Each event carries only data. To get a table's schema (information on how to decode it), use IStore.getSchema(tableId) or listen to StoreSetRecord updates of the schema table.

Schema is a way to dynamically store solidity types. Use the schema-type package (opens in a new tab) to work with it directly. It's needed because abi.decode requires statically hardcoded types, and ABI encoding is very inefficient.

• Schema itself is a tightly packed list of types, like uint256, bytes32, ... in abi.decode(data, (uint256, bytes32, ...)), but the types aren't hardcoded.
• Record data is multiple Fields together, with PackedCounter to tell where arrays end/begin.
• PackedCounter contains lengths to tell arrays apart, packed tightly in 1 bytes32 word.
• Field data is very similar to solidity's abi.encodePacked, but for arrays it has no element padding (this is how solidity internally packs data for storage, but it doesn't expose this tight packing in any way (opens in a new tab))