Fix Composer global home lookup order (#586)

[Mikola Lysenko (mikolalysenko)]

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

Fixes #586

Summary

Global Composer discovery (scan -g, apply -g, vex -g) now finds a Composer 2 global install in the XDG config dir even when an old ~/.composer directory is still around and composer isn't on PATH.

Root cause

When COMPOSER_HOME is unset and composer global config home can't run (composer not on PATH, e.g. a composer.phar install), get_composer_home falls back to composer_home_candidates (crates/socket-patch-core/src/crawlers/composer_crawler.rs). That list probed ~/.composer before the XDG home, which is Composer 1's rule. Composer 2's Factory::getHomeDir checks $XDG_CONFIG_HOME/composer (or ~/.config/composer) first on an XDG system (any XDG_* env var set, or /etc/xdg exists), and only then ~/.composer. So a leftover ~/.composer, even an empty one, hid the real global install and global commands silently found nothing (exit 0).

Fix

• On an XDG system, probe the XDG home first, then ~/.composer, matching Composer 2. XDG detection mirrors Composer's useXdg. An XDG_CONFIG_HOME in the supplied environment also counts.
• An empty XDG_CONFIG_HOME means ~/.config, as in Composer. A relative one is skipped rather than guessed, so it can't put a directory Composer wouldn't use ahead of ~/.composer.
• Windows and non-XDG systems keep their current order. ~/.config/composer stays as a last probe everywhere.
• The candidate list is built on the blocking pool (via Fix Composer vendor-dir config cascade (#439, #686) #1408's default_composer_home), since the XDG check stats /etc/xdg.
• composer_home_candidates keeps its signature, so it doesn't conflict with Fix Composer vendor-dir config cascade (#439, #686) #1408, which also calls it.

Trade-off: Composer 1 still checks ~/.composer first. socket-patch can't tell which Composer major is installed without running composer, and when composer is on PATH its own answer already wins. So the fallback follows Composer 2. Composer 1 is end-of-life, and packagist stopped serving it on 2025-09-01.

Tests (red → green)

Issue case	Test	Before fix	After
#586: stale empty ~/.composer + real ~/.config/composer, XDG system	covgap_crawlers_composer_crawler::get_vendor_paths_global_xdg_home_outranks_stale_dot_composer	FAILED (left: [], the issue's exact symptom)	ok
#586: XDG_CONFIG_HOME set, ~/.composer/vendor also present	covgap_crawlers_composer_crawler::get_vendor_paths_global_xdg_config_home_outranks_stale_dot_composer	FAILED (picked ~/.composer/vendor)	ok
Candidate order: XDG vs non-XDG, empty, relative and supplied XDG_CONFIG_HOME	composer_crawler::tests::test_home_candidates_* (3 unit tests)	new	ok

Commands run locally (Linux, as root):

• cargo fmt --all -- --check: ok
• cargo clippy --workspace --all-features -- -D warnings: ok
• cargo test --workspace --all-features --no-fail-fast: all 281 composer tests pass. 13 tests fail only in this sandbox: chmod-based write-failure tests that can't fail as root (*_write_failure_*, *unremovable*, relax_loop_must_not_traverse_symlinked_root, etc.), plus e2e_bun_lockb, which needs network access to patches-api.socket.dev. None touch Composer code. CI runs them as non-root.
• The Composer e2e suites (e2e_vendor_composer_build, e2e_redirect_composer_build) exercise project mode, not the global-home fallback, so they weren't re-run locally. CI runs them.
• No wrapper (npm/, pypi/, gem/) changes are needed: they only dispatch to the binary.

/code-review high

I fixed every confirmed finding in ab0ffcb except one. Not fixed: "Composer 1 users with both dirs get the XDG dir". That is the intended trade-off described above.

🤖 Generated with Claude Code

https://claude.ai/code/session_011PgPZPKvQDRtAkxyHR3Pe8

---

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 ab0ffcb. Configure here.}

[Mikola Lysenko (mikolalysenko)]

[agent] coverage failed on ab0ffcb in -p socket-patch-cli --test in_process_edge_cases (job 114727527547). This is the intermittent failure tracked in #1416: the same binary failed on #1410 and passed on a re-run of the same SHA. This PR changes only the Composer global-home fallback in crawlers/composer_crawler.rs and its core tests, and doesn't touch that binary. Every other test target in the job passed, and all Composer tests pass locally. No fix exists yet; #1416 is waiting for someone with log access to get the panic text (the log tool here returns only the last 5,000 of ~16,000 lines). Re-running the failed job once.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Ready for review (burn-down agent)

• Head: ab0ffcb
• CI: 36/36 non-skipped checks green (50 skipped by path filters), after one re-run of the coverage job's CI flake: coverage job fails intermittently in vendor / in_process_edge_cases #1416 intermittent in_process_edge_cases failure, which this PR doesn't touch. Mergeable, no conflicts, no CHANGELOG.md change.
• Bugbot: reviewed ab0ffcb, no issues found; no open review threads.
• Reviewer focus: crawlers/composer_crawler.rs. The global-home fallback now probes the XDG Composer home before ~/.composer, matching Composer 2 (fixes Composer global-home fallback checks ~/.composer before the XDG home, so scan -g finds no Composer 2 global packages when a stale ~/.composer exists and composer isn't on PATH #586).

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

Final review brief

What it does: Fixes #586. When neither COMPOSER_HOME nor composer global config home answers, socket-patch now probes the global Composer home in Composer 2's order. On an XDG system it checks $XDG_CONFIG_HOME/composer (or ~/.config/composer) before ~/.composer. An XDG system is one where any XDG_* env var is set or /etc/xdg exists, the same test as Composer's useXdg(). Before this, a stale ~/.composer beat the real XDG home.

Risk: low. It only changes the last-resort fallback. Windows ordering is the same as before. On non-XDG systems, ~/.config/composer is still probed last, so nothing that used to be found is lost.

Look here:

• composer_crawler.rs#L386-L391:`` the candidate list is built on run_blocking because it stats `/etc/xdg`
• composer_crawler.rs#L421-L434:``` the XDG decision and uses_xdg()
• composer_crawler.rs#L454-L473:```` the XDG candidate goes first, an empty XDG_CONFIG_HOME falls back to `/.config`, a relative one is skipped, and `/.config/composer` is de-duplicated
• covgap_crawlers_composer_crawler.rs#L184-L250:````` the two new Composer global-home fallback checks ~/.composer before the XDG home, so scan -g finds no Composer 2 global packages when a stale ~/.composer exists and composer isn't on PATH #586 regression tests, which use a serial env guard

Verified:

• Read the full diff and compared it with Composer 2's Factory::getHomeDir().
• Local test runs at ab0ffcb: composer_crawler unit tests 38/38, covgap_crawlers_composer_crawler 5/5, global_probe_spawn_e2e 10/10, crawler_composer_e2e 33/33.
• Both new regression tests fail against the main-branch composer_crawler.rs.
• CI ci-ok and clippy are green on ab0ffcb, and Bugbot is clean.
• No unresolved threads, and CHANGELOG.md is untouched.

Changes I made: none.

Open questions (non-blocking):

1. uses_xdg() reads the real process env and /etc/xdg, not the injected var closure. Production is unaffected, and the tests call composer_home_candidates_for(.., xdg) with an explicit flag instead.
2. There are two intended differences from Composer. A relative XDG_CONFIG_HOME is skipped, where Composer would resolve it against its cwd. With HOME unset, the code still probes an absolute XDG home, where Composer would throw.

Auto-merge is armed, so approving sends this straight to the merge queue.

---

Generated by Claude Code