CI: core covgap tests join their suites (#1089)

[Mikola Lysenko (mikolalysenko)]

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

Refs #1089 (core-crate slice of child 6; the scan slice is #1583, update merged in #1580). With this and #1583, no covgap_* / coverage_fix_* file is left in the repo. The issue's other children (sentence-assertion counts) stay open.

Summary

The core crate had five tests/covgap_*.rs files. Each was its own test binary, with helpers mirrored from the suite it called its companion. Each one now lives in that suite and uses the suite's helpers:

Old	New
covgap_api_blob_fetcher.rs (6 tests)	blob_fetcher_edges_e2e.rs. The mirrored dummy_client, proxy_client, manifest_with_after_hashes and dir_entry_count are gone
covgap_crawlers_npm_crawler.rs (8)	crawler_npm_e2e.rs. The mirrored options_at and stage_npm_pkg are gone
covgap_patch_apply.rs (4)	crawler_npm_e2e.rs, next to the twin-healing success tests. It uses the suite's index.js patch fixture (ungated and renamed from vlt_patch/vlt_apply) and stage_npm_pkg
covgap_crawlers_composer_crawler.rs (5)	crawler_composer_e2e.rs, using its EnvVars restore guard and global_opts instead of a private EnvVarGuard and global_options_at
covgap_update_state.rs (1)	renamed to update_state_env.rs. It blanks HOME, so it has to stay its own process (see its header)

Cleanups from /code-review high:

• The five older global-home tests in crawler_composer_e2e.rs now restore env through EnvVars instead of hand-written save/restore. Before, a panic left a stubbed HOME/PATH in place for the rest of the binary.
• Function-local imports that the new file-level ones made redundant are removed.
• The read-only blobs-dir test in blob_fetcher_edges_e2e.rs now restores the write bit in a drop guard.

scripts/ci-test-durations.json (the Windows shard balancer) moves the removed targets' timings onto their new homes and adds update_state_env. No product code changes.

CI cost: test-only. A lean PR run is still about 28 Linux jobs and ~8 min, before and after. There are four fewer test binaries to compile and link in every job that builds the core tests (coverage, test-release, Windows/macOS shards). The shard file prices each binary at 2 s of link time. All 24 moved tests still run, with the same assertions.

Test evidence

• Test counts per binary before → after: blob_fetcher_edges_e2e 14+6 → 20, crawler_npm_e2e 86+8+4 → 98, crawler_composer_e2e 39+5 → 44, update_state_env 1. All pass locally.
• cargo test -p socket-patch-core --all-features --test '*': all 32 integration binaries pass. The lib unit tests have 4 failures in this sandbox: copy_tree, vlt_heal, pypi_poetry and pypi_requirements tests whose chmod denials don't hold as root. They are in src/, which this PR doesn't touch.
• cargo fmt --all -- --check is clean. cargo clippy --locked --workspace --all-features -- -D warnings is clean, and so is clippy on the four touched test targets.
• python3 -m pytest scripts/tests -q: 342 passed, 6 skipped. This includes the shard tests that read ci-test-durations.json.
• The moved tests include cfg(unix) and cfg(not(windows)) tests and helpers that are now ungated, so one full-scope run was dispatched on 709b391. It passed, Windows and macOS shards included: https://github.com/SocketDev/socket-patch/actions/runs/38783841518
• This PR moves tests, so there is no red→green regression test.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JH8KWf5dc23jDNhurGwAsN

---

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 709b391. Configure here.}