[Mikola Lysenko (mikolalysenko)]

LLM Description written by Claude Code:claude-opus-5-5

Fixes #641

Summary

scan --mode vendored (and vendor) on a PDM project whose pdm.lock was written by PDM before 2.11 now warns pypi_pdm_stale_install when the env PDM installs into still holds the upstream release. Before, it said nothing, the next pdm sync kept the unpatched bytes, and vex attested not_affected with a "re-run your install" hint that doesn't work on those releases. The warning names the remedy that works for that release line, and vex's vendored_tree_out_of_sync hint now names it too.

Root cause

PDM < 2.11 (lock_version 2 / 4.3 / 4.4) keeps a release already installed at the same version on pdm sync / pdm install. Hosted mode already accounted for this (redirect_pdm_stale_install_risk plus the installed-byte probe). The vendored flavors had an installed-byte probe only for Pipenv and Hatch, so the PDM flavor never checked.

Changes

• vendor/pypi.rs: new pdm_stale_install_warning, the PDM twin of the Pipenv/Hatch probes. It reuses stale_install_sites over non_hatch_local_venv_site_packages, which already resolves PDM's env (.pdm-python, an activated or in-project venv, PEP 582 __pypackages__). It runs on a fresh vendor, an in-sync re-run, and a superseding re-wire. A copy inherited from a --system-site-packages base interpreter is named separately with a remedy that reaches it.
• vendor/pypi_pdm.rs: stale_install_remedy gives pdm sync --reinstall on PDM 2.x. For lock_version 2 (PDM before 2.0, which has no --reinstall), it says to remove the installed copy and then run pdm sync. stale_install_lock / stale_install_note build the VEX addendum, only for a package the governing pdm.lock wires to a vendored artifact, and read pdm.lock once per run.
• utils/pdm_lock.rs: keeps_installed_release is now shared by hosted (redirect/pdm.rs) and vendored mode.
• commands/vex.rs: the vendored_tree_out_of_sync detail appends the PDM remedy.
• Docs: a new row in the CLI_CONTRACT.md warning table plus the vex hint note, and docs/testing/pdm-compatibility.md.
• The npm/PyPI/gem wrappers only pass warning codes through, so they need no change.

Test evidence

Issue	Test	Red without fix → green with fix
#641 (lock 4.3/4.4 venv, fresh + in-sync re-run; 4.4.1/4.5.0 silent; patched env silent)	vendor::pypi::tests::pdm_vendor_warns_about_a_stale_install_before_pdm_2_11	failed left: 0 right: 1 (only vendor_prebuilt_downloaded) → ok
#641 (PDM 1.x lock_version 2, PEP 582 __pypackages__)	vendor::pypi::tests::pdm_1x_vendor_warns_about_a_stale_pep582_install	failed left: 0 right: 1 → ok
#641 (vex hint)	vendor::pypi_pdm::tests::stale_install_note_follows_the_lock_and_its_wiring	new helper → ok
#641 end to end with real PDM	e2e_vex_build pdm::pdm_vendored_install_then_manifest_less_vex (new stale_install_flow: the warning fires, real pdm sync keeps upstream six, and the named remedy installs the patched wheel)	PDM 2.10.4 on pre-fix sources: pdm 2.10.4 lock 4.4 stale install: [] FAILED → ok on 2.10.4 (lock 4.4), 1.4.5 (lock 2) and 2.29.2 (lock 4.5.1, silent)

Commands run locally:

• cargo fmt --all -- --check: ok
• cargo clippy --workspace --all-features -- -D warnings: ok
• SOCKET_PATCH_PDM_E2E_REQUIRED=1 SOCKET_PATCH_PDM_E2E_VERSION={2.10.4,1.4.5,2.29.2} cargo test -p socket-patch-cli --all-features --test e2e_vex_build -- pdm:: --ignored: 2 passed per release
• cargo test --workspace --all-features --no-fail-fast: 228 of 239 suites passed. The 11 failing tests come from this sandbox and are outside the code this PR touches:
  • 9 tests inject a write failure by making a file read-only, which root ignores (the container runs as uid 0).
  • 2 e2e_bun_lockb tests need network access to patches-api.socket.dev and the bun release CI pins (bun 1.4.4 is installed locally).

Review

I ran /code-review high and fixed findings 1-7 in dd2a447: base-interpreter copies, the VEX note limited to packages pdm.lock wires, one pdm.lock parse per run, the superseding path, a shared lock-version predicate, parse_pypi_purl, and the wording. I didn't fix this one:

• The probe runs twice per package, because pypi_prelude runs from both service_preflight and the real run. The Pipenv and Hatch probes already share this cost through the same prelude. Splitting the prelude is a separate refactor.

https://claude.ai/code/session_01JGE2ajfBnCgDHnaCM2D1NK

---

Note

Low Risk
Additive warnings and clearer VEX hints on vendored PyPI/PDM paths; no change to patch application or auth behavior.

Overview
Fixes #641 by surfacing when vendored PDM wiring leaves upstream bytes in the live env on locks written by PDM < 2.11 (lock_version 2 / 4.3 / 4.4), where pdm sync does not replace a same-version install.

Vendored vendor / scan now emits pypi_pdm_stale_install (mirroring Pipenv/Hatch): it probes PDM’s install target (venv or PEP 582 __pypackages__), names pdm sync --reinstall on PDM 2.x or manual removal before sync on lock_version 2, and treats inherited --system-site-packages copies separately. The warning runs on fresh vendor, in-sync re-run, and superseding re-wire.

keeps_installed_release in pdm_lock is shared with hosted redirect_pdm_stale_install_risk. vex appends the same PDM remedy to vendored_tree_out_of_sync when the governing pdm.lock wires the package to a vendored artifact.

CLI contract and PDM compatibility docs document the new code; unit and e2e PDM tests assert warning text, silence on 2.11+ locks, and that the named remedy actually installs the patched wheel.

^{Reviewed by Cursor Bugbot for commit dd2a447. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit dd2a447. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Labeled Ready for review (burn-down agent).

• Head: dd2a4472
• CI: 38/38 non-skipped checks green, no failures; mergeable, no conflicts
• Bugbot: reviewed dd2a447, no findings
• Reviewer focus: the new pypi_pdm_stale_install warning in vendor/pypi_pdm.rs and the shared keeps_installed_release helper in utils/pdm_lock.rs (now used by both hosted and vendored paths).

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. PDM releases before 2.11 (pdm.lock lock_version 2 / 4.3 / 4.4) keep a package that is already installed at the same version on pdm sync. Because of that, vendoring a patch left the unpatched upstream bytes live, and nothing warned about it. This PR adds a pypi_pdm_stale_install warning to vendored PDM, matching the existing Pipenv/Hatch warnings. The warning names pdm sync --reinstall, or for lock_version 2 (PDM before 2.0) tells you to remove the installed copy first. vex's vendored_tree_out_of_sync hint now gives the same remedy.

Risk: low. It only adds a warning and extends a hint. It doesn't change patch application, lock wiring or auth. The lock-version check is now one shared predicate, and hosted mode behaves the same as before.

Look here

• vendor/pypi.rs:910: the new probe, including the split for --system-site-packages base copies
• vendor/pypi.rs:1250: where the probe runs (fresh vendor, in-sync re-run, superseding re-wire; refusals return before it)
• vendor/pypi_pdm.rs:123: the remedy for each release line, and :156, which adds the VEX note only for packages pdm.lock wires to a vendored artifact
• commands/vex.rs:1062: pdm.lock is loaded once per run and skipped in global mode
• utils/pdm_lock.rs:30: keeps_installed_release, now shared with hosted mode

Verified

• Read the full diff (8 files, +501/-7).
• keeps_installed_release covers every lock_version the parser accepts that keeps installs (2/4.3/4.4). 4.4.1/4.5.x stay silent.
• The unit tests cover 4.3, 4.4, 4.4.1 and 4.5.0, PEP 582 on lock_version 2, and a patched env staying silent. The Windows Lib\\site-packages layout is handled.
• The e2e test runs real PDM 2.10.4, 1.4.5 and 2.29.2. The PR description says the new tests fail on the pre-fix code.
• No CHANGELOG edit, no leftover debug code.
• CI: ci-ok and clippy green on dd2a447 (38/38 per the burn-down). Bugbot: clean on dd2a447. No unresolved threads.

Changes I made: none.

Open questions: none blocking. As the PR description notes, the probe runs twice per package (preflight plus the real run), the same as Pipenv and Hatch.

Auto-merge is armed, so approving sends this straight to the merge queue.

---

Generated by Claude Code