14 lines
2.1 KiB
Markdown
14 lines
2.1 KiB
Markdown
# Phase 14 deferred items
|
|
|
|
Out-of-scope discoveries logged by executors. They are not fixed in the plan that found them.
|
|
|
|
## From 14-03
|
|
|
|
- **Phase 12 album cover imports dial the real cover hosts during the parity replay.** `POST albums` (`six-covers`), `PUT albums/{id}` (`cover-urls`), `POST albums/bulk` (`cover-failures`) and `POST albums/{id}/photos` import `cover_urls` through `classes.CoverImporterFromConfig`, whose default fetch is the one-shot `fetchguard.Fetch`. That fetch does not honour `fetchguard.WithTransport`, so the replay reaches `i.discogs.com` instead of an upstream fake, and PHP's recordings of those cases were made against the live hosts too. 14-03 added `classes.NewGuardedCoverImporter` (a `fetchguard.Client` that honours the seam, with PHP's `GuzzleHttp/7` User-Agent) and uses it only on the Discogs routes. Moving the Phase 12 routes onto it needs their cases re-recorded with cover sidecars. Candidate owner: 14-06 or a quick task.
|
|
|
|
## From 14-06
|
|
|
|
- **The Phase 12 cover-import replay (from 14-03) is still open.** 14-06 kept it out of scope: moving the Phase 12 album routes onto `classes.NewGuardedCoverImporter` needs their cases re-recorded against PHP with cover sidecars. Until then `classes/cover_importer.go` `fetch`'s 2xx path can only run against the live image hosts, which is why `check-phase14.sh --coverage` exempts that one function (72.7%).
|
|
- **`TestCoverFetcherHostLock` failed once under the full `-race` run of `check-phase14.sh --go`** (the first run of 14-06, with the whole fonoteka.go tree under test in parallel). The failure output was not kept, and the test passed in every later run: three `-race` runs of its package under load, two further `--go` runs and two `--all` runs. The gate now keeps the full log of a failed `go test` run (`phase14-failed-<pid>.json` under `$TMPDIR`) so a repeat can be diagnosed. Candidate owner: a quick task if it recurs.
|
|
- **A testcontainers Postgres start can exceed its 60 s wait under the gate's parallel load** (the `updates` package, seen once in `--all` at the coverage stage). The gate refuses, as it must; a rerun passed. Raising the wait in the shared test harnesses is a candidate quick task.
|