Implementation Plan: Burn down the 99 grandfathered /tmp-literal test offenders

Branch: fix/tmp-literal-offender-burndown | Issue: Closes #1842 | Spec: spec.md | Research: research.md

Summary

Sweep the 98 live (+1 stale) grandfathered /tmp-literal test files (tmp_ratchet_baseline.txt) off literal /tmp: category-A write-leaks → tmp_path/fixtures with teardown (real isolation, FR-007 — not /dev/shm/mkdtemp evasion); category-B path-literals → non-/tmp absolute sentinels preserving mock/assertion semantics. Then a single gate WP empties the baseline and flips the ratchet from "frozen baseline (>50)" to a self-consistent hard gate (fragment-constructed /tmp/ needle + __file__ self-exclude + a positive self-test replacing the >50 floor). Closes #1842; sibling to the structural PR #2429.

Technical Context

Language/Version: Python 3.11 Testing: pytest (tmp_path fixture), the ratchet tests/architectural/test_no_tmp_paths_in_tests.py Project Type: single project — test-suite hygiene sweep + one gate-mechanism change Constraints: NFR-001 (no xfail/skip/delete/loosening); real isolation not evasion (FR-007); ruff+mypy --strict clean; no new suppressions; gate must not flag itself. Scale/Scope: 98 files across 22 dirs + the ratchet file + the baseline file.

Critical sequencing (the whole plan hinges on this)

  • Conversion WPs never touch tmp_ratchet_baseline.txt. A file that is still baselined but no longer contains /tmp/ is simply skipped by _collect_violations (no violation). So converting files while leaving them baselined keeps the ratchet green throughout, and — crucially — the anti-vacuity >50 floor never trips mid-sweep. The baseline file is owned solely by the gate WP → no parallel-edit conflicts.
  • The gate WP runs last (depends on every conversion WP): only once all 98 files are /tmp-free can the baseline be emptied and the hard gate turned on without self-failure.

Charter Check

  • No masking / no weakening (NFR-001) — real conversions; FR-007 forbids the substring-evasion shortcuts. ✅
  • Non-vacuous / red-first — the gate's positive self-test flags a synthetic offender; cat-A residue-freedom verified across the set. ✅
  • Canonical sources — the self-reference fix mirrors the existing test_no_legacy_terminology.py pattern (fragment needle + __file__ exclude), not a new mechanism. ✅
  • No new suppressions; ruff+mypy --strict clean. ✅

Implementation Concern Map → Work Packages

Conversion WPs are grouped by directory to keep each ≤ ~15–18 files and give one owner per file (no overlap). All conversion WPs are independent (parallel); the gate WP depends on all of them.

WPScopeFilesDep
WP01tests/specify_cli/ group A (~first 15)~15—
WP02tests/specify_cli/ group B (~remaining 14; drop stale cli/commands/test_review.py)~14—
WP03tests/sync/13—
WP04tests/charter/ (7) + tests/status/ (4)11—
WP05tests/doctrine/ (6) + tests/agent/ (6)12—
WP06tests/next/ (3) + runtime/ (2) + unit/ (3) + integration/ (3) + cli/ (3) + dossier/ (3) + contract/ (1)~18—
WP07tests/adversarial/ (2) + audit/ (1) + auth/ (2) + core/ (1) + git_ops/ (1) + glossary/ (2) + kernel/ (1) + paths/ (2) + the 2 non-gate architectural/ files~14—
WP08Gate: empty tmp_ratchet_baseline.txt; make the gate file genuinely literal-free (all 14 /tmp/ lines) — fragment-construct the _TMP_LITERAL needle (:73) AND the self-test payload (:182), reword the 9 docstring/comment + 3 assertion-message lines (4,13,22,83,110,133,143,147,170,157,158,185); keep __file__ self-exclude in _collect_violations (belt-and-suspenders); replace test_baseline_is_non_empty_anti_vacuous (>50) with a positive self-test; add the FR-007 cat-A isolation-adoption check; assert SC-001's exact grep on the gate file returns 02 (test_no_tmp_paths_in_tests.py, tmp_ratchet_baseline.txt)WP01–07

Per-file method (every conversion WP)

1. Read each baselined file; classify per literal occurrence: A (writes under /tmp → route through tmp_path/fixture with teardown) or B (arbitrary absolute path in mock/test-data/assertion → non-/tmp POSIX sentinel, e.g. a clearly-fake /nonexistent/... or a tmp_path-derived string, preserving the exact assertion). 2. Preserve intent (C-001): absolute-path-rejection tests stay absolute-path tests; Windows-literal edge cases keep cross-platform coverage; assertion-message substrings updated in lockstep. 3. Run the file's tests green; confirm no /tmp/ remains (grep), no residue for cat-A. 4. Do not edit the baseline (WP08 owns it).

Project Structure

kitty-specs/tmp-literal-offender-burndown-01KWWRW2/  spec · research · plan · tasks
tests/**/<98 offender files>                          # WP01–07 (by dir group)
tests/architectural/test_no_tmp_paths_in_tests.py     # WP08 gate flip
tests/architectural/tmp_ratchet_baseline.txt          # WP08 empty

Structure Decision: single project; 7 parallel conversion WPs (partitioned by directory, disjoint file ownership) + 1 dependent gate WP.