mirror of
https://github.com/allaunthefox/Research-Stack.git
synced 2026-07-31 03:05:21 +00:00
6100565578
Addresses Devin Review on PR #11 ("Staged regression test passes vacuously — git diff runs in real repo, not temp repo"). Root cause: scripts/math-first/require_math_evidence.py hardcoded cwd=REPO_ROOT on the git subprocess. REPO_ROOT is computed from __file__, which always resolves to the REAL repo path -- even when the test invokes the script with cwd=tmp_path. So 'git diff --cached --name-only' queried the real (clean) repo, returned an empty list, and the script exited 0 via the noop short-circuit. The regression test then printed OK while never actually exercising the --staged classification logic. Fix: Introduce _git_toplevel() in require_math_evidence.py that runs 'git rev-parse --show-toplevel' from the *inherited* cwd. Both _files_from_staged() and _files_from_git_diff() now use that toplevel as their subprocess cwd. Pre-commit and CI both invoke the script from inside the repo root anyway, so behaviour in production is unchanged; only the test (and any other caller that runs the script from a non-Research-Stack directory) now works correctly. Test hardening: test_require_math_evidence.py's staged regression now drives the script via plain subprocess with cwd=tmp_path (no runpy wrapper, no os.chdir trickery) and covers three sub-cases against the same temp repo: (a) math-track-only staged -> expect exit 1 (negative) (b) math-track + receipt staged -> expect exit 0 (the bug) (c) math-track + claims.yaml -> expect exit 0 (registry) Sub-case (a) is the key addition: with the pre-fix script, this case wrongly returns 0 (real repo is clean -> noop), so the test fails loudly. Verified locally by temporarily reverting the cwd fix -- the test correctly reports 'FAIL staged_regression_negative: expected exit 1, got 0'. Docs: docs/math-first-tooling.md notes that receipt-required-for-math-content is a no-op in the CI pre-commit job (pre-commit's --from-ref/--to-ref mode does not touch the index, so --staged returns an empty list and the hook exits 0). PR-scope enforcement of the same rule lives in the dedicated require-evidence CI job. This addresses the Devin Review info comment on .pre-commit-config.yaml. Co-Authored-By: Allaun Silverfox <bigdataiscoming+9i37y6j2@protonmail.com> |
||
|---|---|---|
| .. | ||
| math-first | ||
| archive-and-delete-v2.sh | ||
| bf4prover.py | ||
| clean-tailscale-refs.sh | ||
| finalize-monorepo.sh | ||
| ingest_watcher.py | ||
| pist_orchestrator.py | ||
| populate-open-webui-knowledge-expanded.py | ||
| populate-open-webui-knowledge.py | ||
| remove-tailnet-nodes-api.sh | ||
| reset-tailnet.sh | ||
| review_agent.py | ||
| sciencehub_mcp.py | ||
| SCIENCEHUB_README.md | ||
| zotero_corpus_pipeline.py | ||