niceeval 0.11.4-canary.21 → 0.12.0

package/src/sandbox/vercel.test.ts

@@ -3,7 +3,7 @@
 // download-directory.test.ts;这里只证明 vercel provider 自己的接线,不重新验证模板本身的
 // ignore/剥离/写盘逻辑)。fake `vsb.runCommand` / `vsb.readFileToBuffer`,不连真实 Vercel
 // API——真实 Vercel Sandbox 行为归 E2E(../../docs/engineering/testing/e2e/README.md)。
-import { afterEach, describe, expect, it } from "vitest";
+import { afterEach, describe, expect, it, vi } from "vitest";
 import { mkdtemp, readFile, rm } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
@@ -11,6 +11,7 @@ import { VercelSandbox } from "./vercel.ts";
 
 let roots: string[] = [];
 afterEach(async () => {
+  vi.useRealTimers();
   await Promise.all(roots.map((r) => rm(r, { recursive: true, force: true })));
   roots = [];
 });
@@ -124,6 +125,8 @@ describe("VercelSandbox.runCommand · 执行身份", () => {
 // cases: docs/engineering/testing/unit/sandbox.md「Sandbox 复用」——能力归属。
 // 寿命只从当前 session 的远端元数据读(createdAt + timeout),续期走 SDK 的 extendTimeout;
 // plan 上限拒绝续期时如实报 ready:false,不把请求值当成已生效(见 vercel.ts 的注释)。
+// remainingMs = createdAt + timeout - Date.now();CI 并行下 wall clock 在 before/after 两次
+// 读数之间只要前进就会把刚好够的续期压到 ready:false——用假时钟固定 Date.now。
 describe("VercelSandbox.ensureLifetime", () => {
   function sessionFake(opts: { timeout: number; createdAtOffsetMs?: number; extend?: (ms: number) => void }) {
     const session = { createdAt: new Date(Date.now() - (opts.createdAtOffsetMs ?? 0)), timeout: opts.timeout };
@@ -138,6 +141,8 @@ describe("VercelSandbox.ensureLifetime", () => {
   }
 
   it("当前 session 的远端剩余时间够时直接确认,不调 extendTimeout", async () => {
+    vi.useFakeTimers();
+    vi.setSystemTime(new Date("2026-01-01T00:00:00.000Z"));
     let extended = 0;
     const sandbox = makeSandbox(sessionFake({ timeout: 1_200_000, extend: () => (extended += 1) }));
 
@@ -148,7 +153,10 @@ describe("VercelSandbox.ensureLifetime", () => {
   });
 
   it("剩余不够时按缺口续期,续到够就确认", async () => {
+    vi.useFakeTimers();
+    vi.setSystemTime(new Date("2026-01-01T00:00:00.000Z"));
     const asked: number[] = [];
+    // timeout 10min,已活 9min → 剩余 1min;要 2min → 按缺口续 1min。
     const sandbox = makeSandbox(
       sessionFake({ timeout: 600_000, createdAtOffsetMs: 540_000, extend: (ms) => asked.push(ms) }),
     );
@@ -156,11 +164,12 @@ describe("VercelSandbox.ensureLifetime", () => {
     const result = await sandbox.ensureLifetime(120_000);
 
     expect(result.ready).toBe(true);
-    expect(asked).toHaveLength(1);
-    expect(asked[0]).toBeGreaterThan(0);
+    expect(asked).toEqual([60_000]);
   });
 
   it("plan 拒绝续期(HTTP 400)时报 ready:false 并带上 provider 理由", async () => {
+    vi.useFakeTimers();
+    vi.setSystemTime(new Date("2026-01-01T00:00:00.000Z"));
     const sandbox = makeSandbox(sessionFake({ timeout: 600_000, createdAtOffsetMs: 590_000 }));
 
     const result = await sandbox.ensureLifetime(1_800_000);

package/src/show/index.ts

@@ -38,7 +38,7 @@ import { detectLocale, t } from "../i18n/index.ts";
 import { currentSample, filterExperiments } from "../sample/index.ts";
 import { matchExperimentSelector } from "../shared/aggregate.ts";
 import { panelCapabilityOf } from "../report/model/panel.ts";
-import { formatMetricValue, formatUSD, verdictMark } from "../report/model/format.ts";
+import { formatMetricValue, formatTimeDistance, formatUSD, verdictMark } from "../report/model/format.ts";
 import { renderAlignedRows, type ColumnAlign } from "../report/model/text-layout.ts";
 import {
   annotatedSourceResult,
@@ -396,7 +396,7 @@ async function renderCompareSlice(
     id: "compare",
     title: "Compare",
     navigation: true,
-    render: () => ({ type: Table, props: { data: deltaTableContent(data) } }),
+    render: () => ({ type: Table, props: { data: deltaTableContent(data, io.locale) } }),
   };
   const table = await renderHostPageText(
     page,
@@ -570,9 +570,10 @@ function usageMatrixCellOf(rows: readonly UsageResult[]): UsageMatrixCell {
   };
 }
 
-/** 一格的 7 列文案;`historical` 时在最后一列(成本)追加 `↩`——与 `DeltaTable` 把 `↩` 叠在
- *  同一格末尾同一条排版纪律(faces.ts `deltaConditionCellText`),不是新发明的标注位置。 */
-function usageMatrixCellText(cell: UsageMatrixCell, historical: boolean): string[] {
+/** 一格的 7 列文案;`staleSinceMs` 在场时在最后一列(成本)追加相对时距(experiment-table.md
+ *  「时效不写字」——时距是这一格的数据,不是叠加的装饰符号),与 `DeltaTable` 的历史执行标注
+ *  同一条排版纪律、同一份 `formatTimeDistance` 入口。 */
+function usageMatrixCellText(cell: UsageMatrixCell, staleSinceMs: number | undefined): string[] {
   const cost = cell.costUSD !== undefined ? formatUSD(cell.costUSD) : MISSING_MARK;
   return [
     cell.turns !== undefined ? String(cell.turns) : MISSING_MARK,
@@ -581,15 +582,15 @@ function usageMatrixCellText(cell: UsageMatrixCell, historical: boolean): string
     cell.cacheRead !== undefined ? formatMetricValue(cell.cacheRead) : MISSING_MARK,
     cell.out !== undefined ? formatMetricValue(cell.out) : MISSING_MARK,
     cell.requests !== undefined ? String(cell.requests) : MISSING_MARK,
-    historical ? `${cost} ↩` : cost,
+    staleSinceMs !== undefined ? `${cost} ${formatTimeDistance(staleSinceMs)}` : cost,
   ];
 }
 
 /**
  * `--usage` 在对照范围(`--exp` 出现两次以上)下的逐 eval 用量矩阵
  * (docs/feature/reports/show/usage.md「范围化的用量表」)。配对身份、缺席占位、跨快照携带的
- * `↩` 时效标注,复用 `deltaTableData` 已经算好的这份判定——与 `renderCompareSlice` 消费同一次
- * 计算(条件解析、eval id 配对、watermark/carried 派生的 `historical`),不重新实现一遍;这个
+ * 相对时距标注,复用 `deltaTableData` 已经算好的这份判定——与 `renderCompareSlice` 消费同一次
+ * 计算(条件解析、eval id 配对、watermark/carried 派生的 `historical`/`staleSinceMs`),不重新实现一遍;这个
  * pivot 只在它之上叠一层用量字段的逐条件合计,数据源是 `usageRowsOf` 已有的逐 attempt 行,按
  * experimentId + evalId 分组求和。每个条件一组 7 列(`USAGE_MATRIX_METRIC_COLUMNS`),缺席
  * 条件整组落 `—`。
@@ -618,7 +619,7 @@ async function renderUsageCompareSlice(
       const deltaCell = row.cells[condition];
       if (!deltaCell) return USAGE_MATRIX_METRIC_COLUMNS.map(() => MISSING_MARK);
       const usageCell = usageMatrixCellOf(byConditionByEval.get(`${condition}\u0000${row.key}`) ?? []);
-      return usageMatrixCellText(usageCell, deltaCell.historical);
+      return usageMatrixCellText(usageCell, deltaCell.historical ? deltaCell.staleSinceMs : undefined);
     });
     return [row.key, ...cellsText];
   });