Fix yarnrc compressionLevel comment refusal (#370)

[Mikola Lysenko (mikolalysenko)]

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

Fixes #370

Summary

A .yarnrc.yml line such as compressionLevel: 0 # keep yarn default is yarn's default (yarn reads it as 0 and writes cacheKey: 10c0). socket-patch counted the trailing comment as part of the value, so it refused the project in both vendored mode (vendor_yarn_berry_cache_unsupported) and hosted mode (redirect_yarn_berry_cache_unsupported). Every npm purl in the project was left unpatched with exit 1.

Root cause

yarnrc_compression_level (crates/socket-patch-core/src/vendor/yarn_berry_lock.rs) took everything after compressionLevel: as the value. The vendored gate (refuse_unsupported_compression) and the hosted gate (preflight_yarn_berry_hosted) both read the knob through it, so fixing the reader fixes both.

Fix

The value is now read as a YAML scalar would be:

• A quoted value ('0' / "0") ends at its closing quote. Anything after it, such as a comment, is ignored.
• A plain value ends before a # that follows whitespace (space or tab).
• A # with no whitespace before it stays part of a plain value, as in YAML. So 0#x is not the default and is still refused, so an ambiguous value still blocks.

Tests (red → green)

Test	Layer	Without fix	With fix
yarnrc_compression_level_drops_a_trailing_comment	reader unit	FAIL	pass
yarnrc_compression_level_keeps_an_unseparated_hash	fail-closed guard	pass	pass
commented_default_compression_level_vendors	vendored gate	FAIL	pass
berry_hosted_preflight_mirrors_the_rewriter_gates (+2 asserts)	hosted gate	FAIL	pass
yarn_berry_vendor_commented_default_compression_level	real yarn 4.12.0 e2e: full vendored capstone (fresh checkout --immutable --check-cache, VEX, revert) with the commented knob	FAIL (vendor_yarn_berry_cache_unsupported)	pass

The e2e capstone also asserts that the lock real yarn writes has cacheKey: 10c0. That shows yarn itself reads the commented line as 0.

Local results

• cargo clippy --workspace --all-features -- -D warnings: clean.
• cargo test --workspace --all-features: all pass except three covgap_commands_vendor state-write-failure tests. They depend on chmod 0o555 blocking writes, which root ignores in this container, and the file is unrelated to this change.
• e2e_vendor_yarn_berry_build + e2e_redirect_yarn_berry_build on yarn 4.12.0: 14/14 + 14/14.
• cargo fmt: the new lines follow rustfmt. I didn't run cargo fmt --all, because main itself isn't fmt-clean (it would reformat about 120 unrelated files) and CI doesn't check formatting.
• Wrappers (npm/, pypi/, gem/): no change needed. The change is entirely inside the Rust core.

Per-issue checklist

• Yarn berry vendored and hosted modes refuse compressionLevel: 0 # comment in .yarnrc.yml as a non-default compression level #370: vendored and hosted no longer refuse compressionLevel: 0 # comment; non-default levels with comments still refuse. Covered by the reader, gate and e2e tests above.

CI

All checks are green on 933280e. Two backtest legs that can't reach this code failed on the first attempt and passed on their one re-run each:

• Poetry patch compatibility / native (ubuntu-latest, 1.3.2)
• Bun patch compatibility / native (ubuntu-latest, 1.1.39)

Bugbot reviewed 933280e and found no issues.

🤖 Generated with Claude Code

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] Bun patch compatibility / native (ubuntu-latest, 1.1.39) failed on 933280e: 43/44 cells passed. The one failure is workspace-get-search vendored, check vexLedgersDeleted (vex with the ledgers deleted, scripts/backtest-bun.py:1131).

I don't think this failure comes from this PR. The only code changed is yarnrc_compression_level, and its two callers are yarn-berry-only: refuse_unsupported_compression (vendored berry) and preflight_yarn_berry_hosted (hosted berry). The Bun backtest never reaches either of them.

The same leg passed on the heads of the other open agent PRs, all based on the same main. This run also logged a mock-API request transport failed retry in the sibling workspace hosted cell. No fix exists to port. I'll re-run the failed job once when the workflow run finishes; a second failure gets treated as real.

---

Generated by Claude Code

[Mikola Lysenko (mikolalysenko)]

[agent] A second check also failed on 933280e: Poetry patch compatibility / native (ubuntu-latest, 1.3.2). Two cells failed, direct agent (appliedExactlyOne) and crlf hosted (vexLedgersDeleted, warmInstallReplacesUpstream). The other 3/5 cells passed.

The reasoning is the same as for the Bun leg above: this PR only changes the yarn berry .yarnrc.yml reader, which no Poetry code path calls. The same workflow passed on the other PR heads built on the same main (most recently #503's 0be19d0, at 22:17). Both suites failed within a minute of each other (22:38), on vex and apply checks against the mock patch API.

I've used this check's one re-run (rerun_failed_jobs on run 36936009039). The Bun 1.1.39 leg gets its re-run once its workflow run finishes. If either fails a second time, I'll treat it as real and investigate it from the captured logs.

---

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 933280e. Configure here.}

[Mikola Lysenko (mikolalysenko)]

[agent] Ready for review at 933280e.

• CI: 96/96 non-skipped checks green (plus 1 neutral: a cancelled duplicate Bugbot run). The earlier Bun 1.1.39 and Poetry 1.3.2 cell failures noted above passed on re-run.
• Bugbot: reviewed 933280e (22:59Z) — no issues found.
• Mergeable against main, no unresolved review threads.
• Slack announcement pending: the burn-down agent has no Slack send tool this run.

---

Generated by Claude Code