fix(ci): fresh checkout of the event ref every run; robust tkinter skip #34

Merged

ykp merged 3 commits from fix/ci-fresh-checkout into main 2026-09-29 11:40:13 +02:00

Conversation 0 Commits 3 Files changed 3 +108 -24

ykp commented 2026-09-29 09:05:29 +02:00

Owner

Copy link

Fixes the persistent CI failures on the 3.12/3.14 legs (failing since run 119, even on docs-only commits, while local runs were green). Two compounding root causes:

1. Stale workspace pinning CI to old code (workflow issue)

The workflow cloned only when pyproject.toml was missing. The self-hosted runner reuses a persistent workspace directory, so once a leg holds a leftover (stale/partial) checkout it is silently reused forever: pyproject.toml exists -> no clone -> git not even installed -> never heals. Evidence: failures on docs-only commits, the first attempt at fixing the skip still failing (run 130 - its workspace never received the new code), and contaminated workspaces containing unrelated junk (src/main/.../zte_dummy_legacy/, demo.txt).

Fix: wipe the workspace and clone the event ref fresh on every run - GITHUB_HEAD_REF for pull_request events, GITHUB_REF_NAME for push, refs/pull/N/head fetch fallback, hardcoded FORGE_URL/REPO_SLUG fallbacks for runners that do not export the GITHUB_* context.

2. pytest.importorskip("tkinter") broke under pytest 9.1 (code/test issue)

Since pytest 9.1, importorskip only catches ModuleNotFoundError by default and re-raises plain ImportErrors at collection time. The python-slim CI images ship the tkinter/ package but not libtk8.6.so, so import tkinter raises ImportError (not ModuleNotFoundError) -> collection error libtk8.6.so: cannot open shared object file interrupting the whole run. (The 3.9 image lacks the package entirely -> ModuleNotFoundError -> skipped -> explains its green leg; CI pip install pulled pytest 9.x between runs 118 and 119, which is when the failures started.)

Fix: guard the tkinter import with try/except ImportError and skip via a gui_skip marker helper (tkinter is None or not HAS_DISPLAY). No behavior change when tkinter works.

Also

• README: drop the hardcoded test count (drifts every increment).

Verification

• pytest: 198 passed, 4 skipped (headless); ruff clean.
• The workflow change can only be fully validated by CI itself; this PR's own run exercises the fresh checkout on all three legs.

Fixes the persistent CI failures on the 3.12/3.14 legs (failing since run 119, even on docs-only commits, while local runs were green). Two compounding root causes: ## 1. Stale workspace pinning CI to old code (workflow issue) The workflow cloned only when `pyproject.toml` was missing. The self-hosted runner reuses a persistent workspace directory, so once a leg holds a leftover (stale/partial) checkout it is silently reused **forever**: `pyproject.toml` exists -> no clone -> git not even installed -> never heals. Evidence: failures on docs-only commits, the first attempt at fixing the skip still failing (run 130 - its workspace never received the new code), and contaminated workspaces containing unrelated junk (`src/main/.../zte_dummy_legacy/`, `demo.txt`). **Fix:** wipe the workspace and clone the event ref fresh on every run - `GITHUB_HEAD_REF` for pull_request events, `GITHUB_REF_NAME` for push, `refs/pull/N/head` fetch fallback, hardcoded `FORGE_URL`/`REPO_SLUG` fallbacks for runners that do not export the `GITHUB_*` context. ## 2. `pytest.importorskip("tkinter")` broke under pytest 9.1 (code/test issue) Since pytest 9.1, `importorskip` only catches `ModuleNotFoundError` by default and **re-raises plain ImportErrors at collection time**. The python-slim CI images ship the `tkinter/` package but not `libtk8.6.so`, so `import tkinter` raises `ImportError` (not `ModuleNotFoundError`) -> collection error `libtk8.6.so: cannot open shared object file` interrupting the whole run. (The 3.9 image lacks the package entirely -> `ModuleNotFoundError` -> skipped -> explains its green leg; CI `pip install` pulled pytest 9.x between runs 118 and 119, which is when the failures started.) **Fix:** guard the tkinter import with `try/except ImportError` and skip via a `gui_skip` marker helper (`tkinter is None or not HAS_DISPLAY`). No behavior change when tkinter works. ## Also - README: drop the hardcoded test count (drifts every increment). ## Verification - `pytest`: **198 passed, 4 skipped** (headless); `ruff` clean. - The workflow change can only be fully validated by CI itself; this PR's own run exercises the fresh checkout on all three legs.

ykp added 1 commit 2026-09-29 09:05:30 +02:00

fix(ci): fresh checkout of the event ref every run; robust tkinter skip ... ea6b487b26

Two compounding CI problems (3.12/3.14 legs failing since run 119 on
docs-only commits while local runs were green):

1. Stale workspace pinning CI to old code. The workflow cloned only when
   pyproject.toml was missing, so a leftover workspace on the
   self-hosted runner was silently reused forever (pyproject.toml
   present -> no clone -> no git needed -> never heals). This workspace
   was even contaminated with unrelated junk (src/main/..., demo.txt).
   Fix: wipe the workspace and clone the event ref fresh on every run
   (GITHUB_HEAD_REF for pull_request, GITHUB_REF_NAME for push, with
   hardcoded fallbacks for runners that do not export the context).

2. pytest 9.1 changed importorskip: by default it only catches
   ModuleNotFoundError and re-raises plain ImportErrors at collection
   time. On images where the tkinter package exists but _tkinter cannot
   load (python slim images ship tkinter/ but not libtk8.6.so),
   pytest.importorskip("tkinter") failed collection with
   ImportError: libtk8.6.so instead of skipping. Fix: guard the import
   with try/except ImportError and skip via a gui_skip marker helper.

README: drop the hardcoded test count (drifts every increment).

pytest: 198 passed, 4 skipped (headless); ruff clean.

ykp added 1 commit 2026-09-29 09:12:58 +02:00

ci: retrigger (transient leg failures — same commit green on the mirrored event) 7b3aad96ea

ykp added 1 commit 2026-09-29 09:26:41 +02:00

ci(checkout): stdlib tarball download with retries — drop git/apt ... cc44d53518

The git+apt checkout added two fragile network steps to every job
(apt-get update/install + git clone over the flaky DNS here). Replace
with a single urllib download of the forge tarball API (event sha, ref
fallback, 4 attempts each) extracted with tarfile — pure stdlib, works
in the bare slim images. pip install gains --retries/--timeout for the
same reason. Checkout logic verified end-to-end against the live forge
(archive fetched, extracted flat, pyproject.toml asserted present).

ykp merged commit b9d1507319 into main 2026-09-29 11:40:13 +02:00

ykp referenced this pull request from a commit 2026-09-29 11:40:15 +02:00

Merge pull request 'fix(ci): fresh checkout of the event ref every run; robust tkinter skip' (#34) from fix/ci-fresh-checkout into main

ykp referenced this pull request 2026-09-29 14:44:34 +02:00

020c(gui): Generate scenario button — sidecar-only output + CLI parity #35

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

ykp/gengnss!34

No description provided.