Skip to content

Catch up co-core 0.46 → 0.48 (0.48.2) #206

Closed

Description

@gregoryfoster

Filed by the coordinator of epic #154, with the operator's approval (2026-11-15). This runs next. #185 is waiting on CannObserv/watcher#420.

Why

  • cannobserv tagged v0.47.0, v0.48.0 and v0.48.1 at 2026-11-14T20:16Z, 2026-11-14T23:25Z and 2026-11-15T04:07Z. All three versions are on the co-pypi index for co-core, co-core-aio and co-core-sync. That was read unbounded as co-pypi-reader@ at 04:2xZ and again at 05:4xZ [measured].
  • The repo rule is to pin the current minor (AGENTS.md, Environment & Tooling). The pins at pyproject.toml:447-449 read >=0.46.0,<0.47, so the lock can't take 0.48.
  • Two minors at once. co-core-aio and co-core-sync 0.48.x require co-core>=0.48,<0.49, so the three packages move together. 0.47 adds nothing we'd stop at.

What 0.47 and 0.48 change [stated by CHANGELOG.md at v0.48.1]

  • 0.47.0 is a breaking minor, in the WordPress and transcript models.
    • cannobserv.ext.wordpress.blocks.get_wp_automation_disclosure_block and prepare_block_attribute are removed (#672).
    • TranscriptSpeakerModel.relationship_id is renamed role_assignment_id. The old name still validates (#469 part C).
    • It adds the Automation Disclosure block builder, the transcript serializer, and co_core_sync.services.disclosure_catalog (#676).
  • 0.48.0 is a breaking minor: DisclosureServiceModel.service_type now takes a catalog value only. A label raises ValueError (#678).
  • 0.48.1 is a patch.
    • It adds disclosure_catalog.add_service_type (#680).
    • Drive's files.get, files.list and files.delete now retry transient failures (num_retries=3). Writes are still sent once (#601).
    • A retried delete whose first attempt landed can now raise GdriveApiError(404).

The closure diff, already run [measured 2026-11-15T05:4xZ by the coordinator]

The wheels were fetched with bash scripts/co-pypi.sh uv run --no-project python scripts/co_pypi_wheels.py --dest <dir> --version 0.46.1 --version 0.48.1 co-core co-core-aio co-core-sync. Then scripts/co_core_closure_diff.py 0.46.1 0.48.1 --wheels <dir> printed:

58 files at 0.46.1, 58 at 0.48.1. 4 not identical:
- `version-only` co_core/__init__.py
- `version-only` co_core_aio/__init__.py
- `version-only` co_core_sync/__init__.py
- `changed` co_core_sync/drivers/gdrive.py
Requires-Dist:
- co_core: unchanged
- co_core_aio: removed `co-core<0.47,>=0.46`; added `co-core<0.49,>=0.48`
- co_core_sync: removed `co-core<0.47,>=0.46`; added `co-core<0.49,>=0.48`
Third-party imports (at 0.48.1): google, google_auth_oauthlib, googleapiclient, httpx, pydantic, redis, requests, urllib3 (unchanged).
Dynamic imports: the same two co-core-sync `drivers/` sites.

changed means read it.

  • drivers/gdrive.py is in the closure through co_core_sync.drivers.
  • Replicator has no Drive writer. gdrive is refused as provider_disabled (src/worker/aliases.py:49, src/worker/replicate.py:401), and /etc/replicator/replication-aliases.json holds no gdrive- alias [measured].
  • --diff shows the retry and nothing else [measured by the coordinator]: DRIVE_NUM_RETRIES = 3, and .execute(num_retries=DRIVE_NUM_RETRIES) on files().delete, .get and .list. Re-read it from your own wheels and say so in the pin comment.

No code change is expected. Re-run the diff from the wheels you lock, and paste it in.

Steps

  1. List the index unbounded first (a lock probe at <1). If something newer than 0.48.1 has appeared, widen to it and say so here.
  2. Move all three pins to >=0.48.0,<0.49. The lock will take 0.48.1.
    • Nothing here depends on 0.48.1's changes, so the patch floor stays at .0, per the rule in DEPENDENCIES.md. If you floor at 0.48.1 instead, say why here.
    • Run uv lock --upgrade-package for the three packages through bash scripts/co-pypi.sh uv …, in an isolated venv (the Converge storage on the cluster content-addressed model #114 rule; the worktree steps are in the develop-in-worktrees notes and DEPENDENCIES.md).
    • Expect the lock diff to be the three co-core* entries plus the requires-dist mirror.
  3. Update every literal of the current minor. git grep -nE '0\.46|<0\.47' -- ':!uv.lock' ':!docs/plans' on f2a4333 lists them [measured]:
  4. Gates. Save the full output of each to files.
    • The default suite. The baseline on f2a4333 is 2062 passed, 1 skipped, both clean and loaded.
    • -m integration on a scratch broker.
    • -m browser in the delegated scope.
    • -m gcs.
    • tests/ci and tests/deploy (181 together with tests/test_deploy.py and tests/test_load_env.py on the host).
    • ruff.
    • CI green by job.
    • The real-wheel closure test, with REPLICATOR_TEST_CLOSURE_WHEELS filled at 0.36.0 and 0.37.0 by the fetcher.

Since #202: a session started before 2026-11-15T05:33Z lacks the replicator group. Run id -nG first, and use sg replicator -c '…' if replicator is missing (DEPLOYMENT.md, The service user).

Done when

  • The lock holds 0.48.1 ×3 from the index, with hashes, and uv lock --check is clean.
  • scripts/deploy.sh (via shipping-work) has put the merge SHA live, and on that build:
    • worker ready shows capture, browser_fetch and persist as enabled;
    • the release venv (/srv/replicator/live/.venv) has 0.48.1 ×3;
    • the fetch-policy replay is clean;
    • the worker runs as replicator with NRestarts=0.
  • The pin comment carries the closure-diff report.

The deploy restarts the worker.

Activity

  1. gregoryfoster commented on Nov 15, 2026

    @gregoryfoster
    MemberAuthor

    Pre-flight: widen to 0.48.2 (coordinator, 2026-11-15T16:4xZ)

    Correction to the body's target. cannobserv tagged v0.48.2 at 2026-11-15T06:51:22Z, about 75 minutes after I filed this. It's on the co-pypi index for all three packages [measured 16:4xZ, unbounded]. So step 1's "widen" applies now: the lock takes 0.48.2. The pins stay >=0.48.0,<0.49.

    What 0.48.2 changes [stated by its CHANGELOG]. It's a patch, with no signature changes and no removals:

    • find_by_ulid is added on person, organization, role and assignment (#699).
    • GdriveDeleteIfPresent and GdriveDriver.delete_if_present are added. upload_one's REPLACE now uses them (#692).

    The closure diff, re-run [measured, using wheels fetched by co_pypi_wheels.py]:

    58 files at 0.46.1, 58 at 0.48.2. 5 not identical:
    - `version-only` co_core/__init__.py
    - `changed` co_core/effects/gdrive.py
    - `version-only` co_core_aio/__init__.py
    - `version-only` co_core_sync/__init__.py
    - `changed` co_core_sync/drivers/gdrive.py
    Requires-Dist: co_core unchanged; co_core_aio and co_core_sync: removed `co-core<0.47,>=0.46`; added `co-core<0.49,>=0.48`
    Third-party imports (at 0.48.2): unchanged.
    

    Read with --diff, the two changed files hold only Drive changes:

    • effects/gdrive.py adds the GdriveDeleteIfPresent dataclass and the GdriveDeleteOutcome enum, and adds the new effect to the GdriveEffect union.
    • drivers/gdrive.py adds 0.48.1's num_retries and 0.48.2's delete_if_present.

    Replicator has no Drive writer (the body's provider_disabled reasoning), so no code change is expected still. Paste the 0.46.1 → 0.48.2 report into the pin comment.

    Literal sites: unchanged from the body. tests/test_dependency_index.py records "nothing past 0.48.2".

  2. changed the title [-]Catch up co-core 0.46 → 0.48 (0.48.1)[/-] [+]Catch up co-core 0.46 → 0.48 (0.48.2)[/+] on Nov 15, 2026
  3. gregoryfoster commented on Nov 15, 2026

    @gregoryfoster
    MemberAuthor

    Pickup review (implementer, 2026-11-15T20:1xZ)

    Body + pre-flight checked against origin/main = f2a4333.

    Holds as briefed

    • Index, unbounded: a scratch lock at >=0.46.0,<1 (19:59:58Z) resolves co-core, co-core-aio, co-core-sync 0.48.2. Nothing past it. The pre-flight's widening stands; pins >=0.48.0,<0.49, floor .0.
    • Literal sites: git grep -nE '0\.46|<0\.47' -- ':!uv.lock' ':!docs/plans' on f2a4333 gives exactly the body's list.
    • Closure diff, from my own fetched wheels (co_pypi_wheels.py --version 0.46.1 --version 0.48.2): identical to the pre-flight's — 58/58 files, 5 not identical, 2 changed, both Drive. Same result with tests/ added as a root. Read with --diff: effects/gdrive.py adds GdriveDeleteIfPresent, GdriveDeleteOutcome and the union member; drivers/gdrive.py adds DRIVE_NUM_RETRIES = 3 on files().delete/.get/.list and delete_if_present. Nothing else.
    • No Drive writer: KNOWN_PROVIDERS = ("gcs",) (src/worker/aliases.py:62, reasoning at :45-52); /etc/replicator/replication-aliases.json has 0 gdrive- matches [measured]. No gdrive/Gdrive use in src/ or scripts/ beyond the alias pattern and docstrings.

    Corrections to the body's "What 0.47 and 0.48 change" — from a whole-wheel diff -rq (dist-info excluded), not the CHANGELOG:

    1. 0.47.0's removals ship in none of the three wheels. The CHANGELOG names cannobserv.ext.wordpress.blocks. get_wp_automation_disclosure_block and prepare_block_attribute appear in no .py of the 0.46.1 wheels; at 0.48.2 they appear only in docstrings of two new modules (pure/adapters/wordpress/blocks.py, disclosure.py). Same shape as Catch up co-core 0.45 → 0.46 (co-pypi index; closure diff version-only) #199's wp_view_*_url.
    2. 0.48.0's service_type break has no 0.46.1 surface here. pure/models/disclosure_service.py is new on this move. And at 0.48.2 the label refusal is raised by the WordPress adapter (disclosure.py:_service_type_value) when a block is built, not at model construction — the model only refuses blank.
    3. The transcript rename does ship (pure/models/transcript_speaker.py). Outside the closure, also changed or added: co/v1 exceptions.py, urls.py, disclosure_catalog.py; pure/models/transcript*.py, disclosure_*; workflows/upload.py; co-core-sync runner.py and services/ (_co_v1, amendment_settings, assignment, organization, person, role, sponsorship_settings, new disclosure_catalog). Nothing in src/, scripts/ or tests/ reaches any of them.

    Conclusion unchanged: no code change expected. Proceeding: worktree + isolated venv; I'll confirm the fetched wheels' sha256 against the new lock's hashes before pasting the report.

  4. gregoryfoster commented on Nov 15, 2026

    @gregoryfoster
    MemberAuthor

    Build report (implementer, 2026-11-15T20:2xZ) — PR #207, branch head cae0dce9364d

    Lock: 0.48.2 ×3 from co-pypi, with hashes; diff = the three co-core* entries + requires-dist mirror; uv lock --check clean. The six diffed wheels (0.46.1, 0.48.2) match the old and new lock's sha256 — the report in the pin comment is from the locked wheels. Floor stays .0.

    Gates (worktree, isolated venv; full output saved per gate):

    Gate Result Baseline / brief
    default, clean 2062 passed, 1 skipped 2062 / 1 on f2a4333 (re-measured here)
    default, loaded 2062 passed, 1 skipped same
    -m integration (scratch redis-server 7.0.15, loopback) 57 passed, 0 skipped —
    -m browser (user scope, Delegate=yes) 31 passed —
    -m gcs (test buckets) 17 passed, 0 skipped —
    tests/ci tests/deploy test_deploy.py test_load_env.py 181 passed 181
    real-wheel closure (REPLICATOR_TEST_CLOSURE_WHEELS at 0.36.0/0.37.0) test_reproduces_136_for_0_36_to_0_37 passed, not skipped —
    ruff check + format clean —
    CI (PR run 43756037853) lint ✅ test ✅ gcs ✅ browser ✅ push-to-main run pending the merge

    test_dependency_index.py::test_the_locked_versions_sit_inside_the_pin went red on the lock move as expected, green after the >=0.49 bound.

    Open before close: merge, CI on the merge SHA, scripts/deploy.sh (restarts the worker), the Done-when checks on the live build, and codebase_update from the main checkout for the manifest edit. Paused for the operator before shipping.

  5. added a commit that references this issue on Nov 15, 2026
  6. gregoryfoster commented on Nov 15, 2026

    @gregoryfoster
    MemberAuthor

    CR round 1 (blanket --fix), PR #207 head 096afd362d60:

    • CR 1 — fixed: the pin comment now records that neither changed Drive module can load here: googleapiclient isn't installed (no [gdrive] extra, absent from uv.lock), drivers/gdrive.py imports it at module top, and importing src.worker.main + src.storage.local_store loads neither module [measured].
    • CR 2 — held: the manifest's codebase_update runs from the main checkout after the merge (SocratiCode indexes by absolute path).
    • Also verified in review: the version attributions in the pin comment against the 0.47.0 / 0.48.0 / 0.48.1 wheels, and cannobserv #469 #601 #663 #664 #672 #678 #692 by title.
      Post-fix gate: uv lock --check clean, ruff clean, default 2062 passed / 1 skipped; CI lint/test/gcs/browser ✅ on 096afd362d60. Still paused before shipping.
  7. added a commit that references this issue on Nov 15, 2026
  8. gregoryfoster commented on Nov 15, 2026

    @gregoryfoster
    MemberAuthor

    Shipped — 59fb240 live (PR #207, merged 2026-11-15T23:42Z)

    • Pins >=0.48.0,<0.49 ×3; lock 0.48.2 ×3 from co-pypi with hashes; uv lock --check clean. Floor stays .0.
    • Pin comment carries the 0.46.1 → 0.48.2 closure report from the locked wheels (sha256-matched), the --diff reading, and CR 1's measured fact: neither changed Drive module can load here.
    • Literals moved in AGENTS.md, docs/DEPENDENCIES.md, the manifest, tests/test_dependency_index.py (>=0.49).
    • Commits: 59e1b8d (pin + lock + test), cae0dce (docs), 096afd3 (CR 1), merge 59fb240.

    Done when, on the live build [measured 23:44–23:46Z]:

    • scripts/deploy.sh 59fb24082f7e… exited 0; /srv/replicator/live → releases/59fb24082f7e; CI run 43776215180 green on lint, test, gcs, browser.
    • worker ready (build 59fb24082f7e, 23:44:16Z): persist, capture, browser_fetch all enabled.
    • /srv/replicator/live/.venv: co-core, co-core-aio, co-core-sync 0.48.2.
    • Fetch-policy replay: started, 10 hosts applied, complete; 0 warning-or-above lines since start.
    • Main PID 39128 runs as replicator, NRestarts=0, no replicator-failure records.

    Known limits:

    • SocratiCode: the manifest description and the dependency-contract artifact re-embedded at 23:43Z (34/34 artifacts indexed), but the code index's incremental update was refused — another process held this project's index lock (capped driver and MCP both). The changed files are pyproject.toml, uv.lock, one test and docs; the lock holder's watcher or the next update catches them up.
    • Local -m integration, -m browser and -m gcs ran on cae0dce; CR 1 (096afd3) changed only a pyproject.toml comment, and CI re-ran browser and gcs on it.

    For #154:

  9. gregoryfoster commented on Nov 16, 2026

    @gregoryfoster
    MemberAuthor

    Verified (coordinator, epic #154, 2026-11-15T23:5xZ)

    I read all of this in this check-in [measured]:

    • Range: f2a4333..59fb240. Three commits (59e1b8d, cae0dce, 096afd3) and the merge 59fb240 (PR #206: co-core 0.46 → 0.48, two-minor catch-up #207, merged 23:42:01Z). main is clean, at origin/main.
    • CI by job: lint, test, gcs, browser and update-uv-graph passed on 59fb240; lint, test, gcs and browser passed on 096afd3.
    • Live:
      • /srv/replicator/live points to releases/59fb24082f7e, and the previous release is f2a43335c6b0. /run/replicator/build-id.env reads BUILD_ID=59fb24082f7e.
      • The worker is pid 39128, started at 23:44:13Z, running as replicator:replicator. Its cwd is the release, it has 0 maps under /home, and NRestarts=0. NoNewPrivileges, ProtectSystem=strict, ProtectHome and PrivateTmp are all still on.
      • The release venv holds co-core, co-core-aio and co-core-sync at 0.48.2.
      • The journal since the start has 18 lines, all INFO, plus 7 non-JSON systemd and check_redis_floor lines. worker ready shows persist, capture and browser_fetch as enabled. The policy replay read 503 messages and applied 10 hosts.
      • journalctl -t replicator-failure is empty, and no unit has failed. Root disk is at 55%.
    • Index, unbounded: the newest version is still 0.48.2 for all three packages (23:5xZ).
    • Suites on 59fb240, run by me:
      • Default suite: 2061 passed, 2 skipped, both clean and loaded. The extra skip is test_deploy.py:1049 (not run from an exe.dev session), because I launched the run detached from my session. Run in the foreground, the same test passes. So the result matches your 2062 passed, 1 skipped.
      • Host deploy tests: 180 passed, 1 skipped, with the same skip. That makes 181.
      • ruff check, ruff format --check and uv lock --check are all clean.
    • SocratiCode: this closes the known limit you recorded. The code index's incremental update had completed by 23:4xZ. codebase_search returns test_dependency_index.py with "nothing past 0.48.2" and the >=0.49 bound, and codebase_context_search returns DEPENDENCIES.md at >=0.48.0,<0.49.

    Carried to the epic:

    Nothing is owed.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions