[Mikola Lysenko (mikolalysenko)]

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

Fixes #1774

Summary

Every e2e-extended (mode_migration_vlt, *) leg has been red on the nightly since #1648 (run 40639505834): vlt_pinned_matrix_migration_dry_run_parity and vlt_pinned_matrix_migration_hosted_then_vendored fail on vlt 0.0.0-32, 1.0.0-rc.14 and 1.2.0 (Linux, and the Windows/macOS legs).

Root cause

#1648 made plain vendor with a hosted pin in scope an eject: the dry run reports eject_planned (not vendor_would_revert_redirect), and the wet run no longer emits vendor_takeover_reverted_redirect. #1648 updated the PR-tier migration suites (mode_migration_bun.rs, _npm, _cargo, _pypi) and the in-process vlt test (in_process_vendor_bun_takeover/vlt.rs), but missed mode_migration_vlt.rs. Its tests are #[ignore] (real vlt) and run only in the nightly, so PR and merge-queue CI never saw it. The vendor runs themselves succeed; only the stale advisory assertions fail.

Same #1648 fallout as #1770 (e2e_bun_lockb.rs, fixed in #1771 / #1772); a different file, so no overlap.

Changes (test-only, crates/socket-patch-cli/tests/mode_migration_vlt.rs)

• New helper assert_vendored_over_hosted(eject, ..): only the scan --mode vendored takeover may announce vendor_takeover_reverted_redirect; the plain vendor eject must not. Both must report no redirect_revert_failed and the redirect_vlt_reinstall_required warning from the warm-store heal (the contract in_process_vendor_bun_takeover/vlt.rs asserts).
• hosted_then_vendored, vendor driver: the staged manifest must be byte-identical after the eject.
• dry_run_parity: vendor --dry-run must plan exactly one eject (eject_planned), for the hosted pin and not the bystander, and must not emit vendor_would_revert_redirect; the wet vendor uses the helper.
• Doc comments say eject, not takeover, for plain vendor.

Everything after the vendor runs (pure vendored lock, fresh-checkout installs, exact vendor --revert, hosted re-scan) is unchanged.

Evidence

Run locally with the real vlt releases the nightly pins (scripts/install-vlt.sh, SOCKET_PATCH_VLT_E2E_{JS,VERSION,REQUIRED}, LANG=C), cargo test -p socket-patch-cli --all-features --test mode_migration_vlt -- --include-ignored vlt_pinned_matrix:

vlt	origin/main	this branch
1.2.0	FAILED 12 passed / 2 failed (panics at :312 and :360, same as nightly)	ok, 14 passed
0.0.0-32	(nightly red)	ok, 14 passed
1.0.0-rc.14 + upgrade 1.2.0	(nightly red)	ok, 14 passed

Also: cargo fmt --all -- --check clean; cargo clippy --locked -p socket-patch-cli --all-features --test mode_migration_vlt -- -D warnings reports the same 2 pre-existing duplicate_mod errors as main and nothing new. (The sandbox's TLS proxy is not in reqwest's webpki roots, so I pre-seeded the harness's npm package cache with curl; the test code is unchanged by that.)

Windows/macOS legs are not run on PRs; the change is platform-neutral (no cfg), so no full-scope dispatch.

/code-review high

Fixed: eject tied to the hosted purl with an exact count; manifest-untouched check on the eject; heal warning asserted for both drivers (verified the scan takeover emits it on all three vlt releases); bool instead of a driver-name string; stale doc wording and wrapping.

Not fixed:

• Staging a different patch in the manifest (as the PR-tier bun/yarn suites do) would need a second hosted record and artifact in this real-vlt fixture. The PR-tier in_process_vendor_bun_takeover/vlt.rs already covers "the manifest is not the eject's input"; this nightly leg now checks the manifest is untouched.
• The empty "Start fix" commit is the claim marker; the PR is squash-merged.

CI cost

No workflow change. Lean PR run: same ~28 Linux jobs, ~8 min, before and after. These legs stay nightly-only.

https://claude.ai/code/session_011f9aePnKQscLjaCGykq81B

---

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 ccb4d43. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Ready for review at ccb4d431 (burn-down agent).

• CI: 32 success + 50 skipped on this head, 0 failing; mergeable into main.
• Bugbot: reviewed ccb4d43, no findings, 0 open threads.
• Reviewer note: test-only change to mode_migration_vlt.rs; those legs are nightly-only, so the nightly run after merge is the real confirmation.
• Slack announcement not sent this run (Slack send tool unavailable to the agent).

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review at ccb4d431 (burn-down agent).

• CI: 82/82 check runs green or skipped (32 success, 50 skipped); mergeable, no conflicts.
• Bugbot: reviewed ccb4d431, no findings; no unresolved review threads.
• Test-only change to the nightly mode_migration_vlt tests; the nightly legs confirm after merge.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. Test-only fix for the nightly mode_migration_vlt legs (all vlt versions, all OSes), red since #1648 made plain vendor over a hosted pin an eject. The suite still asserted the old takeover advisories; it now expects eject_planned in the dry run and no takeover event from the wet vendor, while scan --mode vendored still must announce its takeover.

Risk: low. One test file; no product code. These tests are #[ignore] (real vlt) and run only nightly, so PR CI doesn't exercise them.

Look here

• assert_vendored_over_hosted — the new shared contract: takeover event only for the scan driver, no redirect_revert_failed, warm-store redirect_vlt_reinstall_required for both.
• hosted_then_vendored vendor driver — eject must leave .socket/manifest.json byte-identical.
• dry_run_parity eject plan — exactly one eject_planned, for the hosted pin only.

Verified

• Read the full diff. Asserted codes exist where expected (commands/vendor.rs:1940 eject_planned, vlt_heal.rs:35 redirect_vlt_reinstall_required) and match the in-process contract in in_process_vendor_bun_takeover/vlt.rs.
• cargo clippy -p socket-patch-cli --all-features --test mode_migration_vlt -- -D warnings: only the pre-existing duplicate_mod error (same on main), nothing new. I did not re-run the real-vlt legs here (needs the pinned vlt releases); the PR body shows them red on main and green on this branch for vlt 1.2.0, 0.0.0-32 and 1.0.0-rc.14.
• CI 32 passed / 50 skipped of 82 on the head; ci-ok green. Cursor Bugbot: passed. CHANGELOG.md untouched. No overlap with CI: update bun.lockb offline takeover test (#1770) #1771 (different file).

Changes I made: none. (Nit, not blocking: two re-flowed doc-comment lines at the top of the file and in the dry-run-parity doc run past 100 columns; rustfmt doesn't wrap comments, so CI is fine.)

Open questions: none.

Auto-merge (squash) is armed, so approving sends it straight to the merge queue.

---

Generated by Claude Code