1.6 KiB
Local edit journal v1
Scope: ordered fresh excavation requests against one already-loaded snapshot. Cache load/evict/pin operations require a new snapshot and are not silently omitted replay. All fields are explicit little endian; identity and digest trust limits match snapshot v1.
| Offset | Field | Width |
|---|---|---|
| 0 | Magic 0x314c4e4a57564541 | u64 |
| 8 | Version 1 | u32 |
| 12 | Total length including checksum | u32 |
| 16 | World, frame | 2*u64 |
| 32 | Predecessor snapshot checksum | u64 |
| 40 | Predecessor covered command high-water | u64 |
| 48 | Request count (0..64) | u32 |
| 52 | Requests | variable |
| before checksum | Resulting canonical snapshot checksum | u64 |
| final 8 | FNV-1a over all preceding bytes | u64 |
Request: command u64, request digest u64, destination u32, count u32, then 48-byte cuts defined in world-snapshot.md. No receipt is trusted from the journal: ordinary excavation computes it. IDs strictly increase above the predecessor high-water; gaps are allowed, retries are not journal events. Empty journal length is 68. Max length is 256 KiB. Exact consumption is required; partial/trailing/oversized data is rejected.
Restore predecessor into a detached store under an explicit host budget; verify identity/checksum/high-water; apply every command; compare final canonical snapshot checksum before returning. On any failure, discard the entire candidate. Input and any existing live store remain unchanged. The caller must select a complete committed journal chain via M5.7 manifests; codec validation cannot know about omitted external segments. No files, fsync, crash-atomic publication or durable ACK are provided here.