E2E: default image Open WebUI v0.12.0-slim

CLAUDE.md

@@ -24,7 +24,7 @@ Docker-based E2E tests live in `tests/e2e/` (human guide: `docs/testing.md`). Ho
 
 ```bash
 tests/e2e/run.sh <suite>                     # gemini | azure | n8n | infomaniak | filters | all
-tests/e2e/run.sh --image v0.11.3-slim gemini # A/B another Open WebUI release (default v0.11.4-slim)
+tests/e2e/run.sh --image v0.11.4-slim gemini # A/B another Open WebUI release (default v0.12.0-slim)
 tests/e2e/run.sh --ref <branch> <suite>      # test files from a git ref; --src DIR / --file PATH=FILE
 tests/e2e/run.sh --keep --only 'azure.rag' azure   # keep container; then --reuse --name <name>
 tests/e2e/run.sh --strict-known all          # as in CI: a tagged check that passes while its marker applies is FAIL
@@ -33,7 +33,7 @@ tests/e2e/check_owui_api.sh [TAG|latest]     # static check of open_webui import
 
 - **What to run:** changed `pipelines/google/*` → `gemini`; `pipelines/azure/*` → `azure`; `pipelines/n8n/*` → `n8n`; `pipelines/infomaniak/*` → `infomaniak`; `filters/*` → `filters` (+ `gemini` for `google_search_tool`); `tests/e2e/**` → `all`. New Open WebUI release → `check_owui_api.sh latest`, then `run.sh --image <new tag> all`.
 - **Results:** `PASS` / `FAIL` / `KNOWN`. Exit 1 only on FAIL; exit 2 = setup error (Docker/container, leftover `NAME-data` volume without `--force`, an option without value, unknown suite, git ref or `--file` path, `--only` regex matching no group, driver error); 130/143 after Ctrl-C/SIGTERM. KNOWN = fails the way a bug registered in `tests/e2e/harness/known_<area>.py` does (its `evidence` matches the detail, or its log signature shows up with `since=`); a tagged check that fails differently is FAIL. KNOWN does not fail the run; its fixing branch may still be an unmerged/unpublished PR. Markers are **version-gated**: each `KnownIssue` names its `file` and the `fixed_in` version; once the tested file has that version, a failure is FAIL ("regression of known …") and a pass is listed as an obsolete marker ("drop marker"). So when you fix such a bug, bump the file's `version:` and set `fixed_in` to it; drop the markers once the fix is merged. `--only` is a Python regex: `'gemini.(api|image)'`, not `\|`. Output (git-ignored): `tests/e2e/out/<run>/` with `driver.txt`, `results.json`, `summary.md`, `server.log`, `mocks.txt`, staged `functions/` (LF); results are written for interrupted or timed-out runs too (partial).
-- **Baseline** (v0.11.4-slim, `--strict-known`): `main` → 0 FAIL, 0 KNOWN (no known bug registered), no obsolete markers. Check counts per suite and the scenario list: `docs/testing.md` → "What is tested" (keep the counts there only, not here). The `filters` suite needs network for the tiktoken encodings, edits the container's `/etc/hosts` and trust store, and its `offline` group needs a fresh container (no `--reuse`).
+- **Baseline** (v0.12.0-slim, `--strict-known`; v0.11.4-slim and v0.11.3-slim give the same): `main` → 0 FAIL, 0 KNOWN (no known bug registered), no obsolete markers. Check counts per suite and the scenario list: `docs/testing.md` → "What is tested" (keep the counts there only, not here). The `filters` suite needs network for the tiktoken encodings, edits the container's `/etc/hosts` and trust store, and its `offline` group needs a fresh container (no `--reuse`).
 - **Driver checks:** `<suite>.timeout` (`E2E_SUITE_TIMEOUT`, default 900 s; whole run `E2E_TIMEOUT`, default 1800 s), `<suite>.crash`, `<suite>.no-checks`, `<suite>.interrupted`, `run.server-log` (errors or secrets logged outside every suite's scan window; names the `function_<id>`), `run.mocks-log` (errors in `mocks.txt`). Ctrl-C cleans up within seconds; a killed `run.sh` leaves the container, the volume and an orphan driver in the container, which `--reuse` stops. Suites are discovered from `tests/e2e/suites/` (no registration).
 - **Add coverage** for every behaviour change: a `t.check(...)` in `tests/e2e/suites/<suite>.py` (API path `t.owui.chat`, browser path `t.browser().chat`, upstream request via `mock.last()`), mock behaviour in `tests/e2e/mocks/mock_<provider>.py`. Keep `tests/e2e` Ruff-clean. Evidence of a new known bug must not match unrelated failures: the CI meta-test (main's files with `E2E_MOCK_FAULT=500`) allows no KNOWN (`REQUEST_SIDE_KNOWN` in `.github/workflows/e2e.yml`, empty today, is only for bugs that can show nothing but the request sent upstream), so the evidence must include proof that the upstream answered (e.g. `HTTP 200` plus the mock's answer).
 - **Gotchas:** `valves/update` REPLACES all valves (use `update_valves`, which merges); refresh `/api/models?refresh=true` after model/filterIds changes; settings in the data volume override env vars; only the browser path (socket.io + saved chat) exercises event emitters, saved content, usage and sources; background tasks (`/api/v1/tasks/*`) call pipes with `__task__` set and `__event_emitter__=None`; the first Gemini install pip-installs `google-genai` (~30-60 s); in Git Bash prefix your own `docker exec`/`docker cp` with `MSYS_NO_PATHCONV=1` per command (do not export it; `run.sh` unsets it because `git -C /c/...` needs path conversion); never remove Docker images, only your containers/volumes. Open WebUI 0.12 loads functions as `function_<id>_<32 hex>` (the harness normalizes log lines to `function_<id>`) and renamed `__metadata__["user_prompt"]` to `base_user_prompt`; `check_owui_api.sh` only checks names, so run all suites against every new Open WebUI release.

docs/google-gemini-integration.md

@@ -31,8 +31,8 @@ Open WebUI runs `pip` for this requirement when the function is saved and again
 Open WebUI skips the automatic install when `ENABLE_PIP_INSTALL_FRONTMATTER_REQUIREMENTS=false` or `OFFLINE_MODE=true` is set. In that case, and generally for production, multi-worker or multi-replica deployments, install the SDK yourself. For Docker, bake it into your image, which also avoids the in-process `websockets` downgrade:
 
 ```dockerfile
-# Use the same tag you deploy, e.g. v0.11.4, v0.11.4-slim or main
-FROM ghcr.io/open-webui/open-webui:v0.11.4
+# Use the same tag you deploy, e.g. v0.12.0, v0.12.0-slim or main
+FROM ghcr.io/open-webui/open-webui:v0.12.0
 RUN pip install --no-cache-dir "google-genai>=1.68.0,<3"
 ```
 

docs/testing.md

@@ -25,7 +25,7 @@ Container and volume are removed afterwards.
 
 - **Docker** (Docker Desktop on Windows/macOS, Docker Engine on Linux).
 - **bash**: Linux/macOS shell, or **Git Bash** on Windows.
-- Network access for the first image pull (`ghcr.io/open-webui/open-webui:v0.11.4-slim`,
+- Network access for the first image pull (`ghcr.io/open-webui/open-webui:v0.12.0-slim`,
   ~1 GB), for `pip install google-genai` when the Gemini pipe is installed and, in the
   `filters` suite, for the tiktoken encodings (`cl100k_base`, `o200k_base`) the driver
   downloads into the container's `TIKTOKEN_CACHE_DIR` (without them the suite fails).
@@ -37,34 +37,34 @@ Open WebUI container, whose Python already ships `aiohttp`, `httpx`,
 ## Quick start
 
 ```bash
-tests/e2e/run.sh                 # all suites against ghcr.io/open-webui/open-webui:v0.11.4-slim
+tests/e2e/run.sh                 # all suites against ghcr.io/open-webui/open-webui:v0.12.0-slim
 tests/e2e/run.sh gemini          # one suite (gemini, azure, n8n, infomaniak, filters)
 tests/e2e/run.sh azure n8n       # several suites
-tests/e2e/run.sh --image v0.11.3-slim gemini   # another Open WebUI version
+tests/e2e/run.sh --image v0.11.4-slim gemini   # another Open WebUI version
 pixi run e2e gemini              # same via pixi (Linux)
 ```
 
 Typical output (abridged):
 
 ```text
-starting owui-e2e-111759-1234 from ghcr.io/open-webui/open-webui:v0.11.4-slim
+starting owui-e2e-111759-1234 from ghcr.io/open-webui/open-webui:v0.12.0-slim
 Open WebUI healthy after 20s (http://localhost:57765)
-Open WebUI 0.11.4, suites: gemini, azure, n8n, infomaniak, filters
+Open WebUI 0.12.0, suites: gemini, azure, n8n, infomaniak, filters
 function versions: google_gemini.py 1.20.0, azure_ai_foundry.py 3.0.1, n8n.py 2.3.1, ...
 === gemini ===
 [PASS ] gemini.load  pipelines/google/google_gemini.py loads (create, import, activate)
 [PASS ] gemini.api.stream  API stream without websocket session: answer streamed, thinking in the <details type="reasoning" done="true" duration="N"> block
 ...
 [PASS ] gemini.tools.builtin  built-in tool, browser stream=True: Open WebUI runs it, ...
 ...
---- gemini: 415s
+--- gemini: 525s
 ...
 === infomaniak ===
 ...
 [PASS ] infomaniak.models.name-prefix  NAME_PREFIX valve changes the model names, and changing it back restores them
 ...
-SUMMARY: 565 PASS, 0 FAIL, 0 KNOWN in 1213s
-total runtime: 1249s
+SUMMARY: 580 PASS, 0 FAIL, 0 KNOWN in 1263s
+total runtime: 1299s
 output: tests/e2e/out/20261009-111759-owui-e2e-111759-1234
 ```
 
@@ -97,7 +97,7 @@ SIGTERM the script exits with `130` / `143`; see [Interrupted runs](#interrupted
 | Option | Default | Meaning |
 | --- | --- | --- |
 | `[suite ...]`, `-s, --suites a,b` | `all` | `gemini`, `azure`, `n8n`, `infomaniak`, `filters` (the modules in `tests/e2e/suites/`), `all` |
-| `-i, --image IMAGE` | `$OWUI_IMAGE` or `ghcr.io/open-webui/open-webui:v0.11.4-slim` | image, or just a tag (`v0.11.3-slim`) |
+| `-i, --image IMAGE` | `$OWUI_IMAGE` or `ghcr.io/open-webui/open-webui:v0.12.0-slim` | image, or just a tag (`v0.11.4-slim`) |
 | `--only REGEX` | – | only scenario groups whose `<suite>.<group>` matches (Python regex), e.g. `--only gemini.api`; see below |
 | `--ref REF` | – | test the function files as of a git ref (`git show REF:path`): a branch, tag or commit |
 | `--src DIR` | – | test the function files from another checkout / worktree (not together with `--ref`) |
@@ -796,9 +796,9 @@ Python under `tests/e2e/` follows the repo's Ruff settings:
 ## Static Open WebUI API check
 
 ```bash
-tests/e2e/check_owui_api.sh            # against v0.11.4
+tests/e2e/check_owui_api.sh            # against v0.12.0
 tests/e2e/check_owui_api.sh latest     # against the latest Open WebUI release
-tests/e2e/check_owui_api.sh v0.12.0    # any tag or branch
+tests/e2e/check_owui_api.sh v0.11.4    # any tag or branch
 ```
 
 It reads every `from open_webui... import ...` and `import open_webui...` in
@@ -836,7 +836,7 @@ and makes billed requests (about 40 small ones on flash models, with the default
 
 ```bash
 export GOOGLE_API_KEY=...                       # from the environment only, never as an argument
-tests/e2e/smoke/gemini-tools.sh                 # v0.11.4-slim, gemini-3-flash-preview + gemini-2.5-flash
+tests/e2e/smoke/gemini-tools.sh                 # v0.12.0-slim, gemini-3-flash-preview + gemini-2.5-flash
 tests/e2e/smoke/gemini-tools.sh --image v0.11.3-slim --only R3,R14   # some risks, another image
 tests/e2e/smoke/gemini-tools.sh --dry-run-mock  # no key: the same run against the e2e Gemini mock
 ```

tests/e2e/check_owui_api.sh

@@ -2,7 +2,7 @@
 # Static compatibility check: does an Open WebUI release still provide every
 # open_webui.* symbol, method and injected __param__ the functions in this repo use?
 #
-# usage: tests/e2e/check_owui_api.sh [TAG]      TAG: v0.11.4 (default) | latest | any git ref
+# usage: tests/e2e/check_owui_api.sh [TAG]      TAG: v0.12.0 (default) | latest | any git ref
 #
 # Reads pipelines/**/*.py and filters/*.py, fetches the matching Open WebUI backend
 # modules at TAG (via `gh api` when authenticated, otherwise raw.githubusercontent.com
@@ -25,7 +25,7 @@ set -euo pipefail
 
 HERE=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 REPO=$(cd "$HERE/../.." && pwd)
-TAG=${1:-v0.11.4}
+TAG=${1:-v0.12.0}
 GH_REPO=open-webui/open-webui
 CACHE=$(mktemp -d)
 trap 'rm -rf "$CACHE"' EXIT

tests/e2e/run.sh

@@ -25,7 +25,7 @@ HERE=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 REPO=$(cd "$HERE/../.." && pwd)
 
 # Default Open WebUI image (.github/workflows/e2e.yml reads this line).
-DEFAULT_IMAGE=ghcr.io/open-webui/open-webui:v0.11.4-slim
+DEFAULT_IMAGE=ghcr.io/open-webui/open-webui:v0.12.0-slim
 
 FUNCTION_FILES=(
   pipelines/google/google_gemini.py

tests/e2e/smoke/gemini-tools.sh

@@ -4,7 +4,7 @@
 # harness cannot confirm. Manual and opt-in: it needs a Gemini API key and
 # makes billed requests (about 40 small ones).
 #
-#   GOOGLE_API_KEY=... tests/e2e/smoke/gemini-tools.sh [--image v0.11.4-slim] \
+#   GOOGLE_API_KEY=... tests/e2e/smoke/gemini-tools.sh [--image v0.12.0-slim] \
 #       [--model gemini-3-flash-preview] [--model-25 gemini-2.5-flash]
 #   tests/e2e/smoke/gemini-tools.sh --dry-run-mock   # no key: against the e2e mock
 #