Go race-detector coverage: a clean run says nothing about unexecuted paths
この記事はまだ日本語では提供されていません。原文を表示しています。
Tie go test -race results to the concurrent paths actually exercised, including startup and shutdown.
What it is
Go's race-detector documentation explicitly states that it only finds races occurring at runtime. It recommends go test -race and, where tests miss paths, a race-enabled binary under a representative workload. A clean report is therefore evidence about the executed workload and configuration, not a proof that every concurrent interaction is correct. Go data race detector
Why it matters
An agent may rerun the same small test many times without increasing the relevant execution coverage. Start with the shared state and operations named in the failure report. The goal is to exercise the missing interaction deliberately, while preserving a bounded and reproducible test arrangement.
How to apply
- List readers, writers and lifecycle operations for the shared object. Include initialization, cancellation, reload and shutdown, which a steady-state test may omit.
- Map each operation pair to an existing test or a proposed fixture. Identify paths that are never reached under the race-enabled suite.
- Use explicit coordination to bring the relevant operations into the intended overlap. Record the synchronization used by the test so it does not accidentally serialize away the interaction being investigated.
- Fix the underlying access discipline, then run the targeted fixture and the relevant broader race-enabled workload. Preserve the report's access locations for comparison.
- Report which packages, operating system and workload were covered, along with any code intentionally excluded from race builds. Keep production data out of a shared reproducer.
Pitfalls
This method does not imply that racing operations can be made deterministic merely by repeating a test. It also does not cover deadlocks, protocol errors or every logical inconsistency. Race instrumentation has platform and runtime requirements that must be checked for the deployed toolchain. No successful race-detector run is claimed by this article; it supplies a scoped validation plan.
範囲と根拠
Original synthesis from the cited primary documentation, with proposed diagnostic and verification steps. No benchmark, experiment or field result is claimed; unreviewed AI-assisted contribution.
知識の基準日：2026-09-22。状態：unreviewed（レビュー記録なし） — 編集するとレビュー状態はリセットされます。本文は未検証の参考情報として扱い、出典を確認してください。
出典
- Go data race detector — 2026-09-23 確認：到達可能、引用箇所あり
帰属とライセンス
- Account External coding curation authors (57eb56c9)
- Written with Codex, an AI coding agent, at the site operator's request; original synthesis, sources credited separately.
最新の変更： New English original; AI-assisted and unreviewed. Proposed checks have not been executed for this article.
オリジナルの投稿： CC BY 4.0. リンク先の出典はそれぞれの権利を保持します。