|
| 1 | +# October 29, the parity reference moves to published 2.11.7 |
| 2 | + |
| 3 | +- **Before.** The Shared Server fixtures compared native and embedded |
| 4 | + against published Server 2.11.5 (Core 2.15.1). |
| 5 | +- **Reference.** Published Server 2.11.7, Core 2.15.7, as General published |
| 6 | + it and Security accepted it (#861): |
| 7 | + - source `f55f90e54a213872f1c164a1b22c082d859c18aa`; |
| 8 | + - index `sha256:96e3e7de8c58f31bfb1ef4d5bddddc2678f4684f5cb3505e6e9ddc6fa3f7f625`; |
| 9 | + - linux/amd64 `sha256:be7c2288ff935396262d743590530b204fac11b76582e56a525cc81a8da840d3`; |
| 10 | + - linux/arm64 `sha256:7dbc3db050fec05ea5cd10e07ac8399ecf3e285d593bca32979a2320f02a104f`. |
| 11 | + |
| 12 | + `tests/Fixtures/ServerParityBaselines/php-2.11.7.json` records it. The |
| 13 | + SDK, CLI, Waterline and sample-app pins are unchanged. |
| 14 | +- **Why not 2.11.6.** A local survey of published 2.11.6 found that a |
| 15 | + repeated bridge-adapter update answered `500` (#861). 2.11.7 carries the |
| 16 | + fix (#862). |
| 17 | +- **Survey.** Every fixture in `tests/Fixtures/ServerParity` was recorded |
| 18 | + locally against the 2.11.7 amd64 image, SQLite, maintenance loop running, |
| 19 | + except `service-calls` (local SQLite lock pressure) and |
| 20 | + `visibility-current-runs` (it needs the published CLI artifact CI |
| 21 | + provides). Results: |
| 22 | + - 72 of 74 pass unchanged, `bridge-adapters` among them: 2.11.7 gives the |
| 23 | + intended answers at all three of its published-defect steps; |
| 24 | + - `schedule-catalog` fails locally the same way on 2.11.5; |
| 25 | + - `waterline-import` changes as expected (below). |
| 26 | +- **Expectation change.** Server #812 (#770) counts a v1 projection's |
| 27 | + history bytes as Core's `HistoryBudget::sizeOfEvent` does: the event |
| 28 | + type's bytes plus its payload as `json_encode` with |
| 29 | + `JSON_UNESCAPED_SLASHES | JSON_UNESCAPED_UNICODE`. That is the unit |
| 30 | + `DW_V2_LIMIT_HISTORY_TRANSACTION_BYTES` limits everywhere else. |
| 31 | + `waterline-import` records five `history_size_bytes` values: |
| 32 | + |
| 33 | + | Export | 2.11.5 | 2.11.7 | |
| 34 | + | --- | --- | --- | |
| 35 | + | project | 5431 | 4713 | |
| 36 | + | running | 2683 | 2445 | |
| 37 | + | failed | 5425 | 4707 | |
| 38 | + | offsets | 6242 | 5336 | |
| 39 | + | bare | 1856 | 1668 | |
| 40 | + |
| 41 | + No other value in that fixture changed. |
| 42 | +- **Native.** The prepared meter change (`e0ed32a6`, held since October 27 |
| 43 | + for this move) counts each event the same way, for the byte limit and |
| 44 | + for the stored summary. |
| 45 | +- **Local checks before the pull request.** |
| 46 | + - `waterline-import` was recorded against a fresh 2.11.7 harness and |
| 47 | + against a native image built from this branch. Both pass, and |
| 48 | + `compare` accepts the pair. The 2.11.7 recording equals the stored |
| 49 | + contract observation (with the five new sizes) exactly. |
| 50 | + - The Rust library and integration suites pass (634 and 359). |
| 51 | + - The node unit suite passes, except the PHP image-context test, |
| 52 | + which fails the same way on a clean `main` checkout here because of |
| 53 | + local ignored state. |
| 54 | +- **Not changed here.** The `bridge-adapters` contract still accepts the |
| 55 | + published-defect answers of 2.11.1 to 2.11.5. 2.11.7 no longer gives |
| 56 | + them, so a later change can make that contract exact. |
0 commit comments