[Mikola Lysenko (mikolalysenko)]

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

Fixes #1770

Summary

The nightly e2e-extended legs e2e_bun_lockb (bun 1.0.36 and 1.1.45) have been red on main since #1648 (run 40639505834). That PR changed hosted → vendored from a "takeover" into an "eject":

• vendor --offline now refuses up front with offline_eject_unavailable when a hosted pin is in scope.
• The online eject reads each pin's patch record from the API and reports no vendor_takeover_reverted_redirect event.

#1648 updated the in-process eject tests (vendor_eject_bun_lockb.rs). It missed the real-Bun suite e2e_bun_lockb.rs, whose binary-lock legs run only nightly, so PR and merge-queue CI never saw the failure.

Changes (test-only, crates/socket-patch-cli/tests/e2e_bun_lockb.rs)

• native_binary_hosted_vendored_takeover_roundtrip: the offline step now asserts exit 1, offline_eject_unavailable with the re-run-online remedy, no files written and no request made, for both the dry run and the real run.
• Both online ejects (that test and binary_shared_bundled_record_hosted_pin_is_managed) now pass --api-url/--api-token/--org for the mock. Without them the eject fetched the record from the live public patch API (patches-api.socket.dev). The 1.4.2 leg was green only because that live service happened to serve the fixture's uuid.
• Each online eject now asserts that it read the pin's record view from the mock and reported no takeover-revert event. This replaces the old assertion that such an event was present.

Evidence

On origin/main with bun 1.0.36, native_binary_hosted_vendored_takeover_roundtrip fails with expected redirect_revert_failed, the same panic as the nightly.

With this branch, cargo test -p socket-patch-cli --all-features --test e2e_bun_lockb -- --include-ignored passes that test on bun 1.0.36, 1.1.45 and 1.4.2. These pass too:

• cargo fmt --all -- --check
• cargo clippy --locked --workspace --all-features -- -D warnings

Two tests still fail locally only because of this sandbox's network. Both pass in CI on main today.

• native_binary_alias_and_transitive: GitHub tarball 403 through the proxy.
• The final rollback in binary_shared_bundled_record_hosted_pin_is_managed: it fetches the before-blob from the live patch API.

CI cost

No workflow change. A lean PR run keeps the same job count and minutes (about 28 Linux jobs, about 8 min). These legs stay nightly-only.

/code-review high

Fixed: weak negative-only assertions, the missing no-request check on the offline refusal, and stale "takeover" comments.

Not fixed:

• Deduplicating the two eject invocations into a helper. It's style only and would widen the diff.
• Removing fixture.stage() in the shared test. The later rollback(.., copy_already_original = true) relies on that manifest.
• The empty "Start fix" commit. It's the claim marker, and the PR is squash-merged.

Follow-up (not in this PR)

rollback_refuses_binary_hosted_pin_then_checkout still downloads the before-blob from the live public patch API. Pointing it at the mock isn't enough, because the mock has no blob route. Making it hermetic is a separate test-harness change.

https://claude.ai/code/session_011f9aeP7z55jFbL75HPMkHJ

---

Note

Low Risk
Test-only changes to e2e assertions and mock wiring; no production CLI or runtime behavior is modified.

Overview
Updates nightly real-Bun tests in e2e_bun_lockb.rs so hosted → vendored flows match the eject behavior from #1648 (in-process tests were already updated; these legs only run in extended CI).

Offline vendor over a binary hosted pin now expects exit 1, top-level offline_eject_unavailable with a re-run-online message, unchanged project snapshot, and no mock HTTP traffic (dry and wet), instead of redirect_revert_failed in events.

Online eject invocations in native_binary_hosted_vendored_takeover_roundtrip and binary_shared_bundled_record_hosted_pin_is_managed pass --api-url / --api-token / --org so patch records come from the wiremock, not the public API. Assertions now require a /patches/view/ hit via new helper record_views, and explicitly expect no vendor_takeover_reverted_redirect event. Comments are retitled from takeover to eject.

^{Reviewed by Cursor Bugbot for commit d80f222. Configure here.}

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] Heads-up for the reviewer: #1772 (ci-janitor) fixes the same #1770 failure in the same file with an equivalent change. This PR also asserts the offline refusal makes no request and that the online ejects read the record from the mock (record_views). Both are green with clean Bugbot reviews. Merge one and close the other. Its nightly sibling, mode_migration_vlt, is fixed separately in #1775.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review at d80f222e (burn-down agent).

• CI: 32 success + 50 skipped on this head, 0 failing; mergeable into main.
• Bugbot: reviewed d80f222 (summary updated), no findings, 0 open threads.
• Reviewer note: the duplicate Fix nightly bun.lockb e2e for the hosted-pin eject #1772 has since been closed unmerged, so this is the remaining fix for Nightly e2e_bun_lockb red: offline takeover test predates #1648 #1770. Test-only; legs are nightly-only.
• Slack announcement not sent this run (Slack send tool unavailable to the agent).

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review at d80f222e (burn-down agent).

• CI: 82/82 check runs green or skipped (32 success, 50 skipped); mergeable, no conflicts.
• Bugbot: reviewed d80f222e, no findings; no unresolved review threads.
• Fix nightly bun.lockb e2e for the hosted-pin eject #1772 (ci-janitor, same Nightly e2e_bun_lockb red: offline takeover test predates #1648 #1770 fix) was closed unmerged, so this is the one to take. Test-only change; nightly bun legs confirm after merge.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does. Test-only fix for the nightly e2e_bun_lockb legs (bun 1.0.36 / 1.1.45), red since #1648 turned hosted → vendored into an eject. The offline step now expects the up-front offline_eject_unavailable refusal (no writes, no requests), and both online ejects now point --api-url/--api-token/--org at the mock instead of silently hitting the live patch API.

Risk: low. One test file changed; no product code. The legs it fixes only run nightly, so PR CI can't prove them — I ran them locally (below).

Look here

• record_views helper — counts mock record-view hits to prove the eject read the record from the mock.
• offline refusal assertions — exit 1, error code, remedy text, no files, no requests.
• first online eject and shared-bundled eject — new mock API flags and the inverted "no takeover-revert event" assertion.

Verified

• Read the full diff; the asserted code offline_eject_unavailable and "re-run without --offline" text match commands/vendor.rs:980, and the shared mock_api mounts the patches/view/<uuid> route both tests rely on.
• Ran both changed tests with bun 1.4.4 (--include-ignored): native_binary_hosted_vendored_takeover_roundtrip passes; binary_shared_bundled_record_hosted_pin_is_managed passes every new assertion and fails only at its final rollback, which downloads a blob from the live patch API this sandbox can't reach — exactly the case the description lists (pre-existing, follow-up noted).
• CI 32 passed / 50 skipped of 82 on the head; ci-ok green. Cursor Bugbot: passed, no findings. CHANGELOG.md untouched.

Changes I made: none.

Open questions: none.

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

---

Generated by Claude Code