Skip to content

Turbocharge Claude Code, Cursor, Codex, Gemini & every coding agent: faster, cheaper, with contextual understanding specific to your codebase.

test: time the machine-speed workload on the wall clock

Shrish Dwivedi04d7da0d6e67
diff --git a/test/machine-speed.ts b/test/machine-speed.ts
index b7aeef01..c05cbe2c 100644
--- a/test/machine-speed.ts
+++ b/test/machine-speed.ts
@@ -10,18 +10,18 @@
* several runs, against what it took where the bars were set. It is never
* below 1, capped at 10, and exactly 1 under TRAIL_PERF=1 (the strict run).
* Not a test file itself.
+ *
+ * Timed with `performance.now()`, the clock the tests use, not
+ * `process.cpuUsage()`: on Windows CPU time moves in steps of about 15.6 ms,
+ * so a 5 ms run reads as 0, the best of seven is 0, and a runner five times
+ * slower than the laptop got a factor of 1.
*/
-/** The workload's CPU ms where the bars were set: an Apple M-series laptop on Node 25 (Node 20 there takes about 7). */
+/** The workload's ms where the bars were set: an Apple M-series laptop on Node 25 (Node 20 there takes about 7). */
const REFERENCE_MS = 5;
let cached: number | null = null;
-function cpuMs(): number {
- const u = process.cpuUsage();
- return (u.user + u.system) / 1000;
-}
-
function workload(): number {
const text = "token=abc def-ghi /var/x?key=1 ".repeat(8 * 1024);
let n = 0;
@@ -32,14 +32,14 @@ function workload(): number {
return n;
}
-/** The workload's best CPU ms here, after a warm-up run. */
+/** The workload's best ms here, after a warm-up run. */
export function workloadMs(): number {
workload();
let best = Infinity;
for (let k = 0; k < 7; k++) {
- const t = cpuMs();
+ const t = performance.now();
workload();
- best = Math.min(best, cpuMs() - t);
+ best = Math.min(best, performance.now() - t);
}
return best;
}