2. 重试掩盖

• 重复的重试可能会掩盖竞态条件，而不是修复它们。
• 更推荐确定性的设置和更强的同步点。
• 仅在有明确后续计划的情况下，将重试作为临时遏制手段保留。

3. 异步竞态条件

• 缺少 await、孤立的 promise 以及延迟的断言会导致非确定性。
• 确保在断言之前完成异步副作用。
• 为每个测试用例隔离测试状态。

4. 非确定性数据源

• 未模拟的时钟/随机性可能导致间歇性失败。
• 使用受控的时钟和确定性种子。
• 在断言依赖于时间戳的地方冻结或注入时间。

优先级排序

按此顺序修复：

1. 高频执行的测试套件中基于时间的等待
2. 掩盖真实竞态条件的重试掩盖
3. 共享的可变状态 / 缺少 await 的模式
4. 输出不稳定的快照过度使用

修复模式

将 Sleep 替换为条件等待

// 之前
await new Promise((resolve) => setTimeout(resolve, 1000));
expect(await screen.findByText('Ready')).toBeVisible();

// 之后
await waitFor(() => {
  expect(screen.getByText('Ready')).toBeVisible();
});

稳定时间

vi.useFakeTimers();
vi.setSystemTime(new Date('2026-01-01T00:00:00Z'));

移除重试掩盖

• 用显式同步替换隐式重试。
• 将断言范围缩小到用户可见的契约。
• 为确定性状态添加聚焦的设置辅助函数。

工作流程

1. 发现：识别一个高价值的不稳定性来源。
2. 选择一个修复：保持范围紧凑且行为不变。
3. 创建分支：git checkout -b test/flake-<area>
4. 实施：用确定性逻辑替换非确定性模式。
5. 验证：重复运行受影响的测试以增强信心。
6. 记录：记录修复前后的不稳定性模式数量。
7. 提交与合并：运行 /commit-and-push，然后运行 /enter-merge-queue。

如果未找到高价值的修复，请不要创建分支。

验证建议

尽可能多次运行目标测试套件：

pnpm exec tsx scripts/ciImpact/runImpactedTests.ts >/dev/null
pnpm exec tsx scripts/ciImpact/runImpactedQuality.ts >/dev/null

对于可疑的测试套件，使用最可靠的最小命令直接循环重新运行。

防护措施

• 不要在一次 preen 运行中将范围扩大到广泛的测试重写。
• 不要为了满足测试而改变生产行为。
• 不要用更弱的断言替换确定性断言。
• 不要降低覆盖率阈值。

质量标准

• 所选区域的不稳定性模式数量减少
• 测试保持有意义且确定性
• 受影响的检查通过
• 变更小且易于审阅

令牌效率

rg -n --glob '**/*.{test,spec}.{ts,tsx}' 'setTimeout\(|waitForTimeout\(|sleep\(' packages . | head -40 || true
rg -n --glob '**/*.{test,spec}.{ts,tsx}' 'retry|retries|flaky|TODO.*flaky|FIXME.*flaky' packages . | head -40 || true
pnpm exec tsx scripts/ciImpact/runImpactedTests.ts >/dev/null
pnpm exec tsx scripts/ciImpact/runImpactedQuality.ts >/dev/null
git commit -S -m "message" >/dev/null
git push >/dev/null

如果失败，请在没有抑制输出的情况下重新运行失败的命令。

每周安装次数

1

代码仓库

a2f0/tearleads

首次出现

1 天前

安全审计

Gen Agent Trust HubPassSocketPassSnykPass

安装于

zencoder1

amp1

cline1

openclaw1

opencode1

cursor1

2. Retry-Masks

• Repeated retries can hide race conditions instead of fixing them.
• Prefer deterministic setup and stronger synchronization points.
• Keep retries as a temporary containment only with clear follow-up.

3. Async Race Conditions

• Missing await, orphaned promises, and late assertions create nondeterminism.
• Ensure async side effects complete before assertions.
• Isolate test state per test case.

4. Non-Deterministic Data Sources

• Unmocked clocks/randomness can cause intermittent failures.
• Use controlled clocks and deterministic seeds.
• Freeze or inject time where assertions depend on timestamps.

Prioritization

Fix in this order:

1. Time-based waits in heavily exercised test suites
2. Retry masks hiding real races
3. Shared mutable state / missing await patterns
4. Snapshot overuse with unstable output

Fix Patterns

Replace Sleep with Condition Wait

// Before
await new Promise((resolve) => setTimeout(resolve, 1000));
expect(await screen.findByText('Ready')).toBeVisible();

// After
await waitFor(() => {
  expect(screen.getByText('Ready')).toBeVisible();
});

Stabilize Time

vi.useFakeTimers();
vi.setSystemTime(new Date('2026-01-01T00:00:00Z'));

Remove Retry Mask

• Replace implicit retries with explicit synchronization.
• Narrow assertions to the user-visible contract.
• Add focused setup helpers for deterministic state.

Workflow

1. Discovery : Identify one high-value flake source.
2. Select one fix : Keep scope tight and behavior-preserving.
3. Create branch : git checkout -b test/flake-<area>
4. Implement : Replace nondeterministic pattern with deterministic logic.
5. Validate : Run impacted tests repeatedly for confidence.
6. Document : Record before/after flake-pattern count.
7. Commit and merge : Run /commit-and-push, then /enter-merge-queue.

If no high-value fix is found, do not create a branch.

Validation Suggestions

Run targeted suites multiple times when possible:

pnpm exec tsx scripts/ciImpact/runImpactedTests.ts >/dev/null
pnpm exec tsx scripts/ciImpact/runImpactedQuality.ts >/dev/null

For suspicious suites, rerun directly in a loop with the smallest reliable command.

Guardrails

• Do not expand scope into broad test rewrites in one preen run.
• Do not change production behavior to satisfy tests.
• Do not replace deterministic assertions with weaker ones.
• Do not reduce coverage thresholds.

Quality Bar

• Flake-pattern count reduced for selected area
• Tests remain meaningful and deterministic
• Impacted checks pass
• Change is small and reviewable

Token Efficiency

rg -n --glob '**/*.{test,spec}.{ts,tsx}' 'setTimeout\(|waitForTimeout\(|sleep\(' packages . | head -40 || true
rg -n --glob '**/*.{test,spec}.{ts,tsx}' 'retry|retries|flaky|TODO.*flaky|FIXME.*flaky' packages . | head -40 || true
pnpm exec tsx scripts/ciImpact/runImpactedTests.ts >/dev/null
pnpm exec tsx scripts/ciImpact/runImpactedQuality.ts >/dev/null
git commit -S -m "message" >/dev/null
git push >/dev/null

On failure, rerun the failing command without suppression.

Weekly Installs

1

Repository

a2f0/tearleads

First Seen

1 day ago

Security Audits

Gen Agent Trust HubPassSocketPassSnykPass

Installed on

zencoder1

amp1

cline1

openclaw1

opencode1

cursor1