Skip to content

chore(release): 26.11.25 - #55

Merged
ancongui merged 9 commits into
mainfrom
fix/small-bugs-26.11.25
Nov 19, 2026
Merged

ancongui merged 9 commits into
mainfrom
fix/small-bugs-26.11.25

Conversation

@ancongui

Copy link
Copy Markdown
Collaborator

What and why

Release 26.11.25: three small bugs left open after the 26.11.24 bug sweep, all found by its code reviews. dev ci --run could record evidence for tests that never ran, or overwrite a developer's settings, and dev docs check failed on test fixture images while it skipped every image of a managed worktree.

Changes

  • dev ci --run blocks the evidence for a check piped into the home folder through sudo tee, or into a dotfile or dot folder there (pytest | tee ~/log ~/.bashrc, whatever the order of the targets). The same writes with no check beside them are still the runner's setup.
  • dev docs check leaves images in test fixture folders (fixtures, __fixtures__, testdata) out of the diagram rules, and judges only the folders inside the checkout, so a managed worktree has its images checked again.
  • Docs, the CHANGELOG section and the version files at 26.11.25.

How to verify

  • dev ci --run: 8 steps green in 38m42s on Python 3.13 (unit tests 35m22s, Semgrep gate, gitleaks, osv-scanner, trivy, zizmor).
  • New tests fail without each fix: the sudo tee and dotfile cases in tests/test_gitx.py, the fixture and checkout-location cases in tests/test_docs_check.py.
  • An independent code review found two gaps in the first version; both are fixed and the re-review approved the tree.
  • bin/dev docs check passes on the whole repository.

Risk and rollback

Small and local: two classification rules in dev ci --run and one folder rule in dev docs check. A workflow step that pipes a check into a home dotfile or through sudo tee into the home folder now blocks local evidence instead of running or being skipped; such a step should write its log inside the checkout. To roll back, release 26.11.24 again.

Links

  • Tracked for later: B-igkq6ces6s8 (a check redirected into a home dotfile), B-igkq6ck9fgy (a check piped to sudo tee outside the home folder), and the larger items left from the sweep.

dev ci --run skipped 'pytest | sudo tee ~/pytest.log' as runner setup, so the tests never ran and evidence was still recorded, and it ran 'pytest | tee ~/.bashrc' as written, overwriting the developer's own settings. A home-folder write through sudo or into a dotfile or dot folder now blocks the evidence when a check runs beside it; with no check beside it, it is still the runner's setup.

Root cause: _prep_one returned setup for every home write under sudo and the plain home kind for dotfiles, so _classify skipped the sudo pipeline as setup and ran the dotfile pipeline as a check

Verified: python3 -m unittest tests.test_gitx tests.test_ci_commands tests.test_workflow_ci tests.test_cistatus: 173 OK; the new test failed first on 5 of its cases
dev docs check failed on a test fixture image without a title (tests/fixtures/visual/capture/folder/pic.svg), the second time a deliberately plain fixture tripped the diagram rules meant for the pictures readers see. The check now skips SVGs under fixtures, __fixtures__ and testdata folders.

Root cause: the SVG scan in docs.check skipped only build and dependency folders, so every test fixture image had to meet the diagram rules

Verified: python3 -m unittest tests.test_docs_check OK (the new test failed first on its three fixture folders); bin/dev docs check: 47 SVGs, 0 problems
The dev ci --run reference and the runner-setup table now say that a check piped to sudo tee or into a home dotfile blocks the evidence, the dev docs check reference says it leaves test fixture images out, and the CHANGELOG has a 26.11.25 section for both.

Verified: bin/dev docs check docs/commands.md docs/git-and-github.md CHANGELOG.md: 0 issues, 139 links OK
Release 26.11.25 ships the two fixes; the four version files name the release the CHANGELOG section describes.

Verified: git diff -U0 on the four files: one version line each, 26.11.24 to 26.11.25
The code review found that 'pytest | tee ~/log ~/.bashrc' still ran and overwrote ~/.bashrc, while the reverse order was blocked. A home write now takes the most severe of its targets (root, then dotfile, then plain home); '~' itself no longer counts as a dotfile, and the root message names the program.

Root cause: _prep_one returned on the first home target of a writer, so a dotfile after a plain home file was never looked at

Verified: python3 -m unittest tests.test_gitx tests.test_ci_commands tests.test_workflow_ci tests.test_cistatus tests.test_docs_check: 179 OK; the extended test failed first on the reversed targets
The code review found that dev docs check skipped every SVG of a checkout whose own path held a skipped folder name: every managed worktree (.worktrees), and since the fixture rule, a checkout under fixtures or testdata. The skip now looks only at the path inside the checkout. The image test class also moved above the module's main block, so running the file directly runs it.

Root cause: the SVG scan tested the absolute path's parts, so the checkout's parent folders counted as folders to skip

Verified: python3 -m unittest tests.test_docs_check OK (the new checkout cases failed first); bin/dev docs check on the repository exits 0
The docs and CHANGELOG said every check piped to sudo tee blocks the evidence; only a write into the home folder does. The CHANGELOG also names the worktree image fix and the release date.

Verified: bin/dev docs check CHANGELOG.md docs/commands.md docs/git-and-github.md: 0 issues
dev pr merge failed on every pull request with 'cannot read the repository's merge settings: Unknown JSON field: autoMergeAllowed': gh's repo view has no such field, so 26.11.24 could merge nothing. The repo view no longer asks for it; only a merge with --auto reads GitHub's allow_auto_merge from the REST API, and merges nothing when it cannot read it. The tests' fake gh now refuses JSON fields the real gh does not have.

Root cause: the auto-merge fix asked gh repo view --json for autoMergeAllowed, a field gh does not have, and the tests' fake gh answered any field, so no test saw gh refuse it

Verified: python3 -m unittest tests.test_gitx OK; with the old code every merge test fails against the stricter fake, as dev pr merge 55 failed for real
The 26.11.25 section names the dev pr merge failure every 26.11.24 user hits and its fix.

Verified: bin/dev docs check CHANGELOG.md: 0 issues
@ancongui
ancongui merged commit 2105b3c into main Nov 19, 2026
5 checks passed
@ancongui
ancongui deleted the fix/small-bugs-26.11.25 branch November 19, 2026 07:42
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant