Vendored scan of a Pipenv `use_pylock = true` project wires only pylock.toml, but Pipenv installs from Pipfile.lock, so `pipenv sync` / `install --deploy` install the unpatched release after a "success" run

错误位于 PyPI 类型检测逻辑中。从 crates/socket-patch-core/src/vendor/pypi.rs 第 251 行附近以及 crates/socket-patch-core/src/formats/governing_locks.rs 第 115 行附近开始查看。调整锁文件的排序，使得当 Pipfile 与 Pipfile.lock 和 pylock.toml 同时存在时，工具选择 Pipfile.lock 作为主锁文件（或将两者关联起来）。运行相关测试以确认修复。

[agent] Found by the scheduled Pipenv bug-hunt routine (ledger #313).

Summary

With [pipenv] use_pylock = true in the Pipfile, Pipenv 2026's pipenv lock writes both Pipfile.lock and pylock.toml. The pylock.toml carries [tool.pipenv] generated_from = "Pipfile.lock". When both files are present, Pipenv installs from Pipfile.lock: pipenv sync and pipenv install --deploy take the bytes from Pipfile.lock and ignore pylock.toml. I verified this below on 2026.0.0, 2026.4.0 and 2026.8.0.

scan --mode vendored routes this layout to the standalone python-lock flavor, because detect_pypi_flavor step 2 (a standalone pylock*.toml that contains the package) ranks above step 5 (Pipfile.lock → pipenv). So it rewrites only pylock.toml (archive = { path = ".socket/vendor/pypi/<uuid>/…whl" }) and leaves Pipfile.lock on PyPI. The run reports status: success, exit 0, "Vendored 1 package". Its only warning is pypi_multiple_lockfiles: "wiring pylock.toml; installs driven by Pipfile.lock retain their existing sources".

So every Pipenv install afterwards puts the unpatched upstream release in place. Hosted mode handles the same layout correctly: it pins both files, and pipenv sync gives the patched bytes.

This is distinct from #912. #912 is the pylock-only checkout, where Pipenv reads pylock.toml and drops archive. #912 already notes that with both files present "Pipenv installs from Pipfile.lock", which is exactly why vendoring the pylock alone misses here.

Impact

• The documented use_pylock = true layout, vendored, never gets the patch in any Pipenv install (CI, Docker, --deploy), although the scan says it vendored the package.
• The user-visible remedy is awkward. vendor --check (exit 1) says to delete whichever lock the project doesn't install from, but pipenv lock regenerates pylock.toml while use_pylock = true is set, and the next vendored run wires it again.
• vex refuses, as it should (pkg:pypi/[email protected] is wired … but Pipfile.lock resolves the same version from elsewhere, exit 1). So there's no false attestation. The defect is that the wrong file is chosen as the governing lock.

Repro (Linux; real Pipenv 2026.8.0, py3.11; local mock patch API serving a patched six 1.16.0 wheel)

mkdir app && cd app
cat > Pipfile <<'EOF'
[[source]]
url = "https://pypi.org/simple"
verify_ssl = true
name = "pypi"

[packages]
six = "==1.16.0"

[requires]
python_version = "3.11"

[pipenv]
use_pylock = true
EOF
pipenv lock                                   # writes Pipfile.lock AND pylock.toml
socket-patch scan --mode vendored --yes       # exit 0, "Vendored 1 package."
                                              # Warning: wiring pylock.toml; installs driven by Pipfile.lock retain their existing sources
grep -c socket/vendor Pipfile.lock            # 0
grep -c socket/vendor pylock.toml             # 1
pipenv --rm; pipenv install --deploy
pipenv run python -c "import six; print('SOCKET-PATCHED' in open(six.__file__).read())"   # False
pipenv --rm; pipenv sync                      # same: False
socket-patch vendor --check                   # exit 1: wiring contested
socket-patch vex --offline --product pkg:pypi/[email protected] -O vex.json   # exit 1, refuses

# Control: the same project, hosted
socket-patch scan --mode hosted --yes         # rewrittenFiles: [Pipfile.lock, pylock.toml]
pipenv --rm; pipenv sync                      # patched
# Which file Pipenv reads: hosted Pipfile.lock + pristine pylock.toml -> sync / --deploy give the PATCHED bytes

scan --mode vendored --dry-run previews the same pylock-only wiring. Reproduced 2/2 on 2026.8.0, plus once each on 2026.0.0 and 2026.4.0.

Expected vs actual

• Expected: vendored wires the file the project's installer reads. When a Pipfile sits beside both locks (and pylock.toml says generated_from = "Pipfile.lock"), that's Pipfile.lock. The flavor router's own doc says it routes by "this tool manages installs". docs/ecosystems.md lists Pipenv Pipfile.lock vendoring as supported ("every Pipfile.lock category is rewired"). It would also be reasonable to wire both files, as hosted mode does. CLI_CONTRACT.md: "A dep counts as redirected only when its hosted-artifact URL … actually landed in a project file". The vendored analogue is a rewrite the installer honours.
• Actual: only pylock.toml is wired, and Pipenv ignores it while Pipfile.lock exists. The scan succeeds, and every Pipenv install is unpatched.

OS × version

First bad version

Not bisected. v4.0.0 has no pylock vendoring, so this is unreleased v5 behaviour on main b96a785. #1044 moved the tool-lock order into formats/governing_locks.rs (PYPI_TOOL_LOCKS), but the standalone-lock branch that wins here sits before that table and predates it.

Suspect code

• crates/socket-patch-core/src/vendor/pypi.rs:251 (doc, step 2) and :310–:324: if !has_uv_lock && matching_additional_lock { … return Ok((PypiFlavor::PythonLocks, warnings)) }. Any standalone pylock that contains the package beats Pipfile.lock, and nothing checks for a Pipfile or a [tool.pipenv] table in the pylock.
• crates/socket-patch-core/src/formats/governing_locks.rs:115: PYPI_TOOL_LOCKS doesn't place Pipenv-generated pylocks relative to Pipfile.lock.

Related: #912 (pylock-only Pipenv checkout; Pipenv drops archive), #612 (pypi_multiple_lockfiles for a sibling requirements.txt), #1114 (inventory vs router disagreement on empty poetry / pdm locks).