Fix oracle selftest temp-path collision flake in coverage

[Mikola Lysenko (mikolalysenko)]

Problem

The required coverage job failed in two of the last five merge-group runs, and both entries were evicted from the queue:

• https://github.com/SocketDev/socket-patch/actions/runs/37828927026 (pr-1103 group, job 113488896739)
• https://github.com/SocketDev/socket-patch/actions/runs/37828939156 (pr-1095 group, job 113488942697)

Both fail with error: 1 target failed: -p socket-patch-cli --test e2e_vendored_production. The production tests in that binary are all #[ignore]d, so the failure comes from the shared helper selftests compiled into it. The 1051 and 1030 groups, built on the same base in the same batch, passed coverage, so the change being merged doesn't cause it; it's a flake.

Root cause

tests/common/mod.rs gets compiled twice into e2e_vendored_production (and into other binaries that pull in vlt_e2e_common). It's included directly as common and again as vlt_e2e_common::common (#[path = "../common/mod.rs"] pub mod common;). As a result, each oracle_selftests test runs twice, at the same time, in one process. Three of those tests built their scratch paths from the PID alone:

• git_sha256_file_hashes_real_bytes → /tmp/socket-patch-oracle-<pid>-{a,b}.bin
• write_minimal_manifest_emits_apply_compatible_shape → /tmp/socket-patch-oracle-<pid>-manifest
• write_blob_stages_exact_bytes_at_hash_path → /tmp/socket-patch-oracle-<pid>-blob

scratch_dir also ran remove_dir_all first. When the two copies ran together, each one deleted or overwrote the other's files.

Fix

Each of these tests now gets its own tempfile::tempdir(). The directory is unique per call and removed on drop, so the hand-rolled scratch_dir helper and the manual cleanup are gone. tempfile is already a dev-dependency. No assertions changed.

Proof

Looped the e2e_vendored_production test binary locally with --test-threads=16:

	runs	failures
before (origin/main)	300	11: all four collision signatures seen in both copies (manifest must be valid JSON: EOF, create .socket dir: NotFound, read /tmp/socket-patch-oracle-<pid>-a.bin: No such file, write_blob must stage the exact bytes with left: [])
after	500	0

cargo fmt --check passes on the file I touched. cargo clippy -p socket-patch-cli --all-targets -- -D warnings reports nothing in this file. It does report existing needless-borrow lints in tests/prebuilt_common/mod.rs and two other test files, but CI's clippy job doesn't lint test targets.

Where tests run

No tests were removed or moved. These selftests still run everywhere they ran before.

🤖 Generated with Claude Code

https://claude.ai/code/session_01X2N3HNQzBaBgd8xyi3nxva

---

Generated by Claude Code

---

Note

Low Risk
Test-only fixture isolation change; no production or assertion logic changes.

Overview
Fixes flaky coverage failures in oracle selftests by replacing PID-based /tmp scratch paths with tempfile::tempdir() in three tests (git_sha256_file_hashes_real_bytes, write_minimal_manifest_emits_apply_compatible_shape, write_blob_stages_exact_bytes_at_hash_path).

Because tests/common/mod.rs is linked twice in binaries like e2e_vendored_production, those selftests could run in parallel on the same paths; the old scratch_dir helper (which called remove_dir_all first) could clobber sibling runs. Unique per-call temp dirs and drop-based cleanup remove that race. The scratch_dir helper and explicit file/dir teardown are deleted; test assertions are unchanged.

^{Reviewed by Cursor Bugbot for commit bd9c97f. Configure here.}

---

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 bd9c97f. Configure here.}

[Mikola Lysenko (mikolalysenko)]

Burn-down agent: labeled Ready for review at bd9c97f.

• CI: all check suites on the head are success (317 check runs, ci-ok success); no main-wide failures.
• Bugbot: reviewed this head (Cursor check success); no unresolved review threads.
• GitHub still reports mergeable UNKNOWN after repeated re-queries over ~15 min; a local git merge-tree against main (830749f) is clean. No CHANGELOG.md change.
• Slack announcement not sent this run (Slack send tool unavailable); the next run will retry.

---

Generated by Claude Code