[harleypig]

Summary

The tests/vim half of #136. The tests/shell half landed in #141.

• Shared launcher (tests/vim/helpers/vim.bash). All twelve Vim test files built the same environment and the same Vim command line themselves: HOME and the XDG directories in the test's temp directory, a PATH holding only a stub directory, and timeout 30 env -i … vim -N -u .vimrc -es. They now load one helper, following the shell half's load helpers/<name> pattern:

  • vim_fixture, called from setup, sets $repo, $vim, $vimrc, $home, $stubs (holding only sh), $report and the $vim_env array.
  • headless_vim [VAR=VALUE...] [--] [VIM ARGS...] runs Vim in that environment, with any extra variables taking precedence, and returns Vim's exit status.
  • vim_env_unset NAME... removes a variable for tests that need it unset (vimrc.bats unsets XDG_DOTVIM, xdg.bats unsets XDG_STATE_HOME and XDG_DATA_HOME).
  • with_node is the node tripwire stub, which coc.bats and load.bats each had a copy of.

  Each file's report.vim scaffolding and its own submodule check are left as they were. Their contents differ from file to file, so they are not a shared launcher.

• oldfiles.bats → fold.bats (from Small tidy-ups: envsubstitute help quote, old shell-lint warnings, proselint config path #44). The file tests CustomFoldText() in the late fold.vim. Its name and header still referred to the deleted OLD_FILES/. The new name matches set.bats and maps.bats, and the header now says what the file checks.

• .claude/TESTS.md gets a Shared helpers section saying where each suite's helpers live and naming both of them (v0.2.0).

Refs #44
Closes #136

For files added later: other tests/vim/*.bats files being written at the same time as this PR still carry their own copy of the launcher. Each one should switch to load helpers/vim and call vim_fixture/headless_vim once it and this PR have both merged.

Test plan

• bats tests/vim with a fresh, empty TMPDIR: 75/75 ok, which is the same count as on master before the change. TMPDIR was still empty afterwards.
• bats tests/shell with a fresh, empty TMPDIR: all ok, TMPDIR still empty afterwards.
• shellcheck -s bash and shfmt -i 2 -s -bn -ci -sr -d (the repo's shfmt style) on the helper: both clean.
• markdownlint (the CI image, pinned by digest) on .claude/TESTS.md: clean.

Testing evaluation

I ran this myself to the qa-tester's standard; there is no team-managed: sentinel.

1. Verdict: already covered. The change is a behaviour-preserving refactor of the test harness, and the 75 existing tests are what cover it. The risk is that a test is still green but no longer able to fail, so each break below was checked for red after the refactor.
2. Automatable: yes. bats, tests/vim/*.bats.
3. Falsifiers. Each break was applied after the refactor, confirmed red, then restored:
   • .vimrc XDG_DOTVIM fallback pointed at /nowhere → vimrc.bats red in 3 tests (unset, empty, symlink). This shows vim_env_unset XDG_DOTVIM really leaves the variable unset.
   • vim_env_unset made a no-op → xdg.bats "unset, Vim uses the defaults" red.
   • headless_vim drops its extra VAR=VALUE args → xdg.bats "set, Vim uses them" red.
   • with_node stub left non-executable → coc.bats "with node" and "Tab at the start of a line" red.
   • Real PATH (host has node) in vim_env → coc.bats "without node, CoC … not loaded" red.
   • set foldtext= commented out in fold.vim → fold.bats red in all 3 tests.
   • algorithm:patience commented out in set.vim → set.bats red.
   • A call to an undefined function appended to set.vim → load.bats "starts with no error message" red in both tests (with and without node).
   • helptags.vim made to always rebuild → helptags.bats "tags newer … left alone" red.
   • shfmt --indent 2 → --indent 4 in the sh ftplugin → sh-ftplugin.bats red.
4. Disposition: nothing new is owed. The falsifiers above were the work this PR needed.
5. Entry-point reach: not applicable. This is test harness code, not a guard or a gate.

Manual verification

1. From a checkout with submodules initialised (git submodule update --init) and BATS_LIB_PATH pointing at bats-support/bats-assert.
2. Run d=$(mktemp -d); TMPDIR=$d bats tests/vim; ls -A "$d".
3. Success: 1..75, every line ok, exit 0, and ls -A prints nothing.
4. Failure looks like any not ok line, or a leftover entry in $d.
5. Run: 1..75, 75 ok, 0 not ok, exit 0. The temp directory was left with 0 entries.

Review

opt-out: no named review skill (code-review or similar) ran on this test-only refactor. In its place, the falsifiers above stand as evidence, together with a self-review of the diff, which turned up the points below for a reviewer to check.

• Behaviour change. vimrc.bats and xdg.bats now also get XDG_CONFIG_HOME and XDG_CACHE_HOME inside the temp directory. Before, they left both unset, which falls back to $home/.config and $home/.cache, also inside the temp directory. Neither file tests those two variables.
• Argument parsing. headless_vim treats leading VAR=VALUE words as environment variables. A file argument containing = therefore has to come after --. statusline.bats and sh-ftplugin.bats already pass -- before their file paths.
• What to check: whether report and home belong in vim_fixture. report is unused by helptags.bats and sh-ftplugin.bats, and has a SC2034 suppression with its reason. Also check whether moving with_node (two copies, not three) goes beyond the issue's "launcher".

Completion report

1. Functionality works: met. All 75 tests pass, and the falsifiers above confirm the tests can still fail.
2. Access model: not applicable. Test harness, no access surface.
3. Permissions: not applicable, same reason.
4. Audit logging: not applicable, same reason.
5. Validation: not applicable. No input boundary.
6. Errors handled: met. Vim's exit status is still returned to each caller, and each caller keeps its own || vim_status=$? or || true handling.
7. Testing complete: met. See the testing evaluation above.
8. Documentation: met. .claude/TESTS.md Shared helpers, the helper's own header, and the updated test-file headers.
9. Completion report: this section.

Co-Authored-By: AI Agent noreply@harleypig.com