Vendored pnpm refuses a user exact-pin override (`left-pad: 1.3.0`) in pnpm-workspace.yaml with a misleading "does not match package.json" error, though the same pin in package.json is taken over

[mikolalysenko]

[agent] Found by the scheduled pnpm bug-hunt routine (ledger #303).

Summary

Vendored pnpm takes over a user's exact-version override of the package it vendors (pnpm_lock.rs OverrideDisposition::Takeover: "the user's pin already forces every tar-fs to this exact version, so redirecting the same key preserves their semantics"). That takeover only looks at package.json pnpm.overrides. On pnpm 10.5+, and always on pnpm 11 / 12 (which ignore package.json overrides), the override lives in pnpm-workspace.yaml. There, the same bare-key exact pin is refused with vendor_override_conflict:

pnpm-lock.yaml carries an override key left-pad for left-pad that does not match package.json's left-pad@1.3.0 — the two override maps must agree (run pnpm install to re-sync them) before vendoring

package.json has no overrides at all, and the lock and the workspace file already agree. The suggested pnpm install changes nothing, so the error points the user at the wrong file and the wrong fix.

The behaviour is inconsistent across key shapes and files:

• bare key left-pad: 1.3.0 in package.json pnpm.overrides (pnpm 9.15.9): taken over, frozen install patched, revert restores it
• versioned key left-pad@1.3.0: 1.3.0 in pnpm-workspace.yaml (10.34.5, 12.8.1): taken over, frozen install patched, revert byte-exact
• bare key left-pad: 1.3.0 in pnpm-workspace.yaml (10.34.5, 11.28.3, 12.8.1): refused

Impact

On pnpm 11 / 12, exact-pinning a vulnerable transitive dependency in pnpm-workspace.yaml overrides: is the standard way to force a version, and vendoring can't patch that package. The refusal is loud and writes nothing for that package. But from a hosted project the takeover un-hosts it first (#853), and the error sends users to package.json and pnpm install, which don't help.

Repro (Linux, Node 22, main 4646693)

I used a local mock patch API that serves left-pad@1.3.0 with a marker prepended to index.js.

API="--api-url http://127.0.0.1:8787 --org acme --api-token x"
mkdir p && cd p
echo '{"name":"c","version":"1.0.0","dependencies":{"left-pad":"1.3.0"}}' > package.json
printf 'packages:\n  - .\noverrides:\n  left-pad: 1.3.0\n' > pnpm-workspace.yaml
pnpm install            # the lock records `overrides: left-pad: 1.3.0`
socket-patch scan --mode vendored --json --yes $API
# exit 1, partial_failure, left-pad: failed vendor_override_conflict ("… does not match package.json's `left-pad@1.3.0` …")
# change the key to `left-pad@1.3.0: 1.3.0` → exit 0, the value becomes the file: spec, frozen install patched

Expected vs actual

• Expected: the workspace file is treated like package.json for overrides. An exact pin equal to the vendored version, under either the bare or the name@version key, is taken over in place, and vendor --revert restores the user's value. That's how classify_pkg_override handles package.json, and on pnpm ≥10.5 the workspace file is the authoritative override map (Vendored pnpm 10.5+ with overrides: in pnpm-workspace.yaml: the new package.json pnpm.overrides shadows the user's overrides, so frozen installs fail and a re-lock drops them #360 / Fix vendored pnpm shadowing workspace overrides (#360) #785). A genuine conflict (a range, a different version or a selector) stays a refusal, and its message names the file that actually carries the override. CLI_CONTRACT.md lists vendor_override_conflict only for "a user-authored override/resolution … already exists"; the takeover exception exists in the code but not in the docs.
• Actual: the bare-key workspace pin is refused, and the detail blames package.json.

Matrix (vendored scan exit / outcome)

pnpm	bare key in pnpm-workspace.yaml	versioned key in pnpm-workspace.yaml	bare key in package.json
9.15.9	1 / refused	not run	0 / taken over, frozen patched
10.34.5	1 / refused	0 / taken over, frozen patched, revert byte-exact	not run
11.28.3	1 / refused	not run	n/a (pnpm 11 ignores package.json overrides)
12.8.1	1 / refused	0 / taken over, frozen patched, revert byte-exact	n/a

Each refused cell was reproduced at least twice. I tested Linux only (no OS-specific code is involved). Not a regression: release 4.0.0 gives the same refusal and message on 12.8.1.

Suspect code

• crates/socket-patch-core/src/vendor/pnpm_lock.rs:617: classify_pkg_override derives effective_key from package.json alone, so with no package.json override it is our canonical left-pad@1.3.0.
• pnpm_lock.rs:1404 (check_lock_override) then sees the lock's left-pad key ≠ effective_key and refuses with the package.json wording. pnpm_lock.rs:1852 (check_workspace_override) would refuse it the same way.
• A fix would classify the workspace overrides: section with the same Takeover / Ours / conflict rules when it holds the override.

---

Backlog review — 2026-10-08

Priority: P1 → P2. Existing pnpm override shapes cause a refusal; failures are visible rather than unsafe successful patch claims.

[mikolalysenko]

[agent] Triage: priority:p1. pnpm (npm-family). The override takeover only reads package.json pnpm.overrides and misses a bare-key exact pin in pnpm-workspace.yaml. This is also one of the three triggers in #853, but #853's root cause (takeover restores the hosted pin before the vendored gates run) is separate, so they are cross-linked rather than clustered.

---

Generated by Claude Code

[mikolalysenko]

[agent] Claiming this issue as part of the pnpm open-issue sweep in draft PR #1007. Branch: agent/fix-pnpm-open-issues. Claim-ID: 2026-10-07T12:44:29Z-pnpm07

[mikolalysenko]

[agent] PR #1007 fixes this issue. It has a regression test that fails on main, CI is fully green (552 checks) and it's ready for review. The issue will close when #1007 merges. The PR description's table gives the root cause, the fix and the test for each issue.