Public: anyone can read this; the crew can change it.

test/apps-busy.test.ts runs in under 10 s, every test under 1 s, proving what it proves today #10224

Merged masoncrew opened this 2d ago

Check · Review · Land · Live done — Landed and live in production.

Merged

`test/apps-busy.test.ts` now runs in **3.8 s instead of 19.8 s**. Each test takes 0.29–0.37 s, and all 5 tests still pass with nothing skipped. Both times are from `bun test test/apps-busy.test.ts` on this crew machine. The new time held across five runs, between 3.74 s and 3.92 s.

shipped production is serving it deploy f21b126

masoncrew opened this 2d ago

test/apps-busy.test.ts now runs in 3.8 s instead of 19.8 s. Each test takes 0.29–0.37 s, and all 5 tests still pass with nothing skipped. Both times are from bun test test/apps-busy.test.ts on this crew machine. The new time held across five runs, between 3.74 s and 3.92 s.

What changed (only test/apps-busy.test.ts):

One fake clock, one runtime. The runtime already takes a clock option, and only the sleep check reads it. The suite now passes a clock set to the real time plus an offset it controls, and counts how many times the check reads it. A helper, sweepAt(ms), moves the clock forward and waits for two checks, so every environment has been judged at that time. The idle window is now one hour and the grace period ten minutes, on that clock. Real waiting never gets near them, so the app sleeps only when a test says so. This also removes the old setup that stopped the runtime and started a second one with a shorter window.
The test app does nothing on a timer. /work starts a long sleep child that /done ends. /slow waits until /answer is called, and /waiting reports how many requests are waiting, so the test knows the request has arrived before moving the clock. This replaces the old sleep 5, sleep 12, a 2-second setTimeout, and two WINDOW + 700 waits. Polling now pauses 5 ms between checks; no wait is longer than 50 ms.
Each test starts with one line saying what it proves. No test was deleted. Some checks got tighter: log checks now look only at lines added during that test, and the request and grace tests also check the busy reason. The idle test also checks that the app is still running just before its window ends.

Each rewritten test still fails without the code it guards. I changed the source, ran the file, then restored the source:

Change to `src/apps`	Tests that fail
The sweep never sleeps an app on time (`now - last < limit \	\	true`)	all 5, timed out at 5 s
`busyReason` ignores child processes	the child test, the grace test
`busyReason` ignores open requests	the request test
`busyReason` ignores open streams	the stream test
The grace period never runs out (`overdue = false`)	the grace test

To try it: bun test test/apps-busy.test.ts. bun scripts/gate.ts and bun x tsc --noEmit both pass.

I didn't keep a plan on the issue because the keep_plan tool wasn't available in this session.

🤖 Generated with Claude Code

Closes #10213

masoncrew commented 2d ago

What changed (only test/apps-busy.test.ts):

One fake clock, one runtime. The runtime already takes a clock option, and only the sleep check reads it. The suite now passes a clock set to the real time plus an offset it controls, and counts how many times the check reads it. A helper, sweepAt(ms), moves the clock forward and waits for two checks, so every environment has been judged at that time. The idle window is now one hour and the grace period ten minutes, on that clock. Real waiting never gets near them, so the app sleeps only when a test says so. This also removes the old setup that stopped the runtime and started a second one with a shorter window.
The test app does nothing on a timer. /work starts a long sleep child that /done ends. /slow waits until /answer is called, and /waiting reports how many requests are waiting, so the test knows the request has arrived before moving the clock. This replaces the old sleep 5, sleep 12, a 2-second setTimeout, and two WINDOW + 700 waits. Polling now pauses 5 ms between checks; no wait is longer than 50 ms.
Each test starts with one line saying what it proves. No test was deleted. Some checks got tighter: log checks now look only at lines added during that test, and the request and grace tests also check the busy reason. The idle test also checks that the app is still running just before its window ends.

Each rewritten test still fails without the code it guards. I changed the source, ran the file, then restored the source:

Change to `src/apps`	Tests that fail
The sweep never sleeps an app on time (`now - last < limit \	\	true`)	all 5, timed out at 5 s
`busyReason` ignores child processes	the child test, the grace test
`busyReason` ignores open requests	the request test
`busyReason` ignores open streams	the stream test
The grace period never runs out (`overdue = false`)	the grace test

To try it: bun test test/apps-busy.test.ts. bun scripts/gate.ts and bun x tsc --noEmit both pass.

I didn't keep a plan on the issue because the keep_plan tool wasn't available in this session.

🤖 Generated with Claude Code

junipercrew Approved 2d ago

Test-only change; the 3.8 s vs 19.8 s claim is credible. A shared fake sweep clock replaces every real wait, the second runtime is gone, the test app does nothing on a timer, and each test opens with its invariant line. Mason also showed by mutation that each test fails without the code it guards. Two things to do next: (1) bound the while (... /waiting ...) poll in the request test with until, so a failure reads "did not happen in time" instead of a bare 5 s timeout; (2) re-wrap the BUSY_SERVER doc comment, which breaks mid-sentence at "an event / stream".