[Mikola Lysenko (mikolalysenko)]

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

Fixes #1135

Summary

With HOME and CARGO_HOME both unset (env -i, systemd units, stripped
CI/container entrypoints), cargo still builds from the account's passwd home
(e.g. /root/.cargo). socket-patch found no cargo home at all, so
scan --mode agent reported the crate lockfile-only and exited 0 while cargo
compiled the unpatched copy, and scan -g -e cargo found no packages. It now
resolves the cargo home the way cargo does and patches the copy cargo builds.

Root cause

Four places resolved the cargo home on their own, each as $CARGO_HOME, else
utils::fs::home_dir()/.cargo: the cargo crawler, the [patch] config chain
(vendor/cargo_config.rs), the build-cache cleaner
(utils/cargo_build_cache.rs) and the self-update cargo-install check
(update/channel.rs). home_dir() reads only HOME/USERPROFILE. Cargo's
home crate reads HOME on Unix (USERPROFILE on Windows) and, on Unix,
falls back to the passwd entry when HOME is unset.

Fix

• New utils::cargo_home::cargo_home(): $CARGO_HOME when set and non-blank,
  else <home>/.cargo, where <home> is HOME (USERPROFILE on Windows) or,
  with that variable unset or empty on Unix, the passwd entry (as std's
  home_dir does). A relative HOME still means no home (unchanged), so we never patch a shared
  passwd-home cache cargo isn't using.
• All four readers go through it.
• passwd_home() moved from crawlers::gradle_cache to utils::fs (re-exported
  from gradle_cache) and is looked up once per process.
• docs/ecosystems.md (Cargo: shared registry cache) states the rule.

Test evidence

• Red on main: cargo test -p socket-patch-core --lib cargo_home_without_home_env_uses_passwd_home
  failed before the fix (CargoCrawler::cargo_home() returned None), passes after.
• New resolver unit tests in utils/cargo_home.rs (CARGO_HOME wins, blank is unset,
  HOME wins over passwd, unset HOME falls back to passwd (Agent-mode cargo with HOME unset finds no registry cache, reports the crate as lockfile-only and exits 0, while cargo builds the unpatched copy from the passwd home's ~/.cargo #1135), empty HOME falls back to passwd, relative HOME is no home, USERPROFILE ignored on Unix, process-env end to end).
• Binary repro (the issue's global variant): env -u HOME -u CARGO_HOME socket-patch scan -g -e cargo --json
  now reports scannedPackages: 238 from /root/.cargo (the issue reported 0 on main).
• cargo fmt --all -- --check: clean. cargo clippy --workspace --all-features -- -D warnings: clean.
• cargo test -p socket-patch-core --all-features --lib: 6809 passed, 4 failed. The 4
  (copy_tree::relax_loop_must_not_traverse_symlinked_root,
  vlt_heal::an_unremovable_hidden_lock_keeps_every_store_entry,
  pypi_poetry::wire_write_failure_maps_error_and_leaves_lock_untouched,
  pypi_requirements::wire_failure_rolls_back_already_written_files) depend on
  permission denials and fail only because this container runs as root. They
  don't touch this change and pass in CI.
• cargo test -p socket-patch-core --all-features --test '*': 655 passed.
• cargo test -p socket-patch-cli --all-features --bins --lib: 935 passed.
• CLI cargo suites: e2e_safety_cargo_build -- --ignored 10/10, e2e_vendor_cargo_build 40/40,
  e2e_redirect_cargo_build 23/23, e2e_redirect_cargo_shapes 34/34, mode_migration_cargo 30/30.
• The full cargo test --workspace build ran out of local disk while linking every CLI
  test binary, so the cargo-related CLI suites above were run one at a time instead.
  CI runs the full matrix.

Issue	Test
#1135 (agent scan, global scan, config chain)	cargo_crawler::tests::cargo_home_without_home_env_uses_passwd_home, utils::cargo_home::tests::*

/code-review high

Fixed: Unix read USERPROFILE (and Windows read HOME), unlike cargo; a relative
HOME fell through to the passwd home; update/channel.rs was a fifth resolver;
the passwd lookup repeated on every call; utils depended on the Gradle crawler;
the test env guard wasn't panic-safe.
Not changed:

• A whitespace-only CARGO_HOME is still treated as unset. That keeps the existing
  behavior and the repo's env_non_empty convention.
• The crawler-level test overlaps the resolver test. It stays as the regression test
  showing the crawler itself is fixed.
• Other per-user caches (Python, Deno, NuGet) with the same HOME-unset gap are out
  of scope here. Each tool has its own home rule, so they need separate fixes.
• On Windows with USERPROFILE unset, cargo uses the known-folder API. We don't
  replicate that. It is untested in the issue too.

https://claude.ai/code/session_011f9aeNZ3vKbaLRi7eRFxTp

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

Bugbot is paused — your team has an unpaid invoice

Bugbot is paused because your team has an unpaid invoice.

Pay it from the billing page in the Cursor dashboard and Bugbot resumes automatically.

(requestId: 2df4790e-1391-452f-afb5-bd06211571ae)

[cursor]

Bugbot is paused — your team has an unpaid invoice

Bugbot is paused because your team has an unpaid invoice.

Pay it from the billing page in the Cursor dashboard and Bugbot resumes automatically.

(requestId: 99ed0440-467b-4417-ab1a-541a6e93f3d7)

[Mikola Lysenko (mikolalysenko)]

[agent] CI is green on be7e923 (31 success, 50 skipped by path filters). Bugbot did not review: cursor[bot] replied that Bugbot is paused because the team has an unpaid invoice. I'll re-trigger @bugbot review once billing is restored.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

BugBot review

---

Generated by Claude Code

[cursor]

Bugbot is paused — your team has an unpaid invoice

Bugbot is paused because your team has an unpaid invoice.

Pay it from the billing page in the Cursor dashboard and Bugbot resumes automatically.

(requestId: 731d67ec-bb47-4791-8bdb-a07152df6889)

[cursor]

Bugbot is paused — your team has an unpaid invoice

Bugbot is paused because your team has an unpaid invoice.

Pay it from the billing page in the Cursor dashboard and Bugbot resumes automatically.

(requestId: 587d318e-f4bd-443b-80b4-0a87cbc37490)

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

bugbot run

---

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 6566a42. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Ready for review at 6566a42 (burn-down agent).

• CI: 50 check runs: 31 success, 19 skipped, all green (Cursor Bugbot included).
• Bugbot reviewed 6566a42: no issues found. No unresolved review threads.
• Slack announcement not sent this run (Slack posting unavailable); the next run retries.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review at 6566a42 (burn-down agent).

• CI: 49/49 check runs on 6566a42 green or skipped (30 success, 19 skipped); mergeable into main.
• Bugbot reviewed 6566a42 (now that billing is restored): no issues found. No open review threads.
• Slack announcement not sent: the Slack send tool was unavailable this run.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does: Adds one cargo-home resolver, utils::cargo_home::cargo_home(). It uses a non-blank $CARGO_HOME first, then <home>/.cargo. On Unix, an unset or empty HOME now falls back to the passwd entry. Four call sites used to work this out separately (the cargo crawler, the vendor [patch] config, the build-cache cleaner and the self-update channel check), and they all call it now. So env -i and systemd runs patch the ~/.cargo registry that cargo actually builds from (#1135).

Risk: low. Behaviour only changes when HOME is unset or empty on Unix, or on Windows when HOME and USERPROFILE differ (it now follows cargo and uses USERPROFILE). The unsafe getpwuid_r code was moved byte-for-byte from gradle_cache.rs, and gradle_cache still re-exports it.

Look here

• utils/cargo_home.rs:36-56: the resolve order, empty vs relative HOME, and the passwd fallback
• utils/fs.rs:405-444: the moved getpwuid_r FFI and its OnceLock cache
• update/channel.rs:70 with channel.rs:545-550: the self-update channel now uses the resolved cargo home
• crawlers/cargo_crawler.rs:1112-1127:`` the crawler-level regression test

Verified

• Reviewed the unsafe FFI: pointers are read only after a successful call, the buffer is resized on ERANGE and capped at 1 MiB, and pw_dir is null-checked.
• cfg(unix) gating with a non-unix stub.
• No other CARGO_HOME resolvers remain in crates/*/src.
• Affected-module tests: 155/155 pass.
• With the old resolver put back, 3 of the new tests fail, so they do catch the bug.
• cargo clippy -p socket-patch-core --all-features -D warnings (the CI form): clean.
• CI on the head: skipped:19 success:31. Bugbot: success. CHANGELOG.md is untouched. The description matches the code.

Changes I made: none.

Open questions (non-blocking nits)

• NoHomeEnv (cargo_home.rs:62-91) is #[cfg(test)], but all of its users are #[cfg(unix)]. A Windows cargo test build gets a dead_code warning, not a failure. #[cfg(all(test, unix))] would fix it.
• ChannelEnv::from_env has no direct test, and a whitespace-only CARGO_HOME now counts as unset.

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

---

Generated by Claude Code