test/body-limits.test.ts runs in under 10 s, every test under 1 s, proving what it proves today #10225

test/body-limits.test.ts now runs in about 5.3 s, down from 14.0 s. One test still takes about 2.9 s, so the "no test over 1 s" item is not met. That last wait is in the server code under src/, which is outside this issue's scope, so I filed #10226 for it.

Timings are from bun test test/body-limits.test.ts in this sandbox, not the crew machine: 14.04 s before; 5.34 / 5.26 / 5.31 s after. Both runs had 22 tests and 33 expect() calls. No test was deleted or skipped, and only this file changed.

What made it slow, and what changed

• The "misanswered" test sat out its own 2 s socket timer, because the connection stayed open after the answer. Its trailing GET /api/v1/health now sends Connection: close, so the test ends on that answer. It still expects ["413","200"]. Now about 10 ms.
• The three tests that wait for the server to hang up were the stranger cap on /api/v1/fleet/ops, the never-ending body on /oauth/register, and the never-ending anonymous push. The server ends a connection with server.timeout(request, 1), and Bun only fires socket timeouts every 4 s. So the hang-up comes 0–4 s after the answer however fast the client sends; I measured four send rates and all closed at about 4.0 s.

- The three connections now start together in beforeAll, and their tests run last, so the other tests run during the wait. - Never-ending bodies are sent in 64 KiB chunks instead of 4 KiB, so they reach the cap sooner. - The first of the three to wait takes about 2.9 s; the other two then finish almost instantly.

• Each rewritten test now starts with a one-line comment stating what it proves. Helpers: pump (a paced chunked body), chunkedPost (the request head) and a RawOutcome type.

Shown to fail without the code they guard (I edited src/http/limits.ts, then put it back)

• closeAfter made a no-op: all three hang-up tests fail.

- Stranger cap: the 401 comes without connection: close. - Never-ending body (read and unread): closedByServer is false.

• Close the connection without draining the rest of the body: "the next request on a reused connection…" gets 400 instead of 401. This fails with both the old file and the new one.
• "Misanswered" test: I found no change to src/ that makes it fail, with the old file or the new one. I tried no drain, cancelling the read, closeAfter as a no-op, and forcing a close; Bun keeps the connection's framing by itself. It is kept as it was, apart from the Connection: close line.

Checks

• bun scripts/gate.ts and bun x tsc --noEmit pass.
• The file passed 22/22, three runs in a row.
• scripts/design-lint.ts doesn't apply, since nothing under src/web/ changed.

Not done

• The keep_plan tool wasn't available in this run, so I didn't keep a plan on the issue.
• #10226 asks for the server to close the connection as soon as the answer is sent. I made #1242 wait for it.

To try it: bun test test/body-limits.test.ts

🤖 Generated with Claude Code

Closes #10215

test/body-limits.test.ts now runs in about 5.3 s, down from 14.0 s. One test still takes about 2.9 s, so the "no test over 1 s" item is not met. That last wait is in the server code under src/, which is outside this issue's scope, so I filed #10226 for it.

Timings are from bun test test/body-limits.test.ts in this sandbox, not the crew machine: 14.04 s before; 5.34 / 5.26 / 5.31 s after. Both runs had 22 tests and 33 expect() calls. No test was deleted or skipped, and only this file changed.

What made it slow, and what changed

• The "misanswered" test sat out its own 2 s socket timer, because the connection stayed open after the answer. Its trailing GET /api/v1/health now sends Connection: close, so the test ends on that answer. It still expects ["413","200"]. Now about 10 ms.
• The three tests that wait for the server to hang up were the stranger cap on /api/v1/fleet/ops, the never-ending body on /oauth/register, and the never-ending anonymous push. The server ends a connection with server.timeout(request, 1), and Bun only fires socket timeouts every 4 s. So the hang-up comes 0–4 s after the answer however fast the client sends; I measured four send rates and all closed at about 4.0 s.

- The three connections now start together in beforeAll, and their tests run last, so the other tests run during the wait. - Never-ending bodies are sent in 64 KiB chunks instead of 4 KiB, so they reach the cap sooner. - The first of the three to wait takes about 2.9 s; the other two then finish almost instantly.

• Each rewritten test now starts with a one-line comment stating what it proves. Helpers: pump (a paced chunked body), chunkedPost (the request head) and a RawOutcome type.

Shown to fail without the code they guard (I edited src/http/limits.ts, then put it back)

• closeAfter made a no-op: all three hang-up tests fail.

- Stranger cap: the 401 comes without connection: close. - Never-ending body (read and unread): closedByServer is false.

• Close the connection without draining the rest of the body: "the next request on a reused connection…" gets 400 instead of 401. This fails with both the old file and the new one.
• "Misanswered" test: I found no change to src/ that makes it fail, with the old file or the new one. I tried no drain, cancelling the read, closeAfter as a no-op, and forcing a close; Bun keeps the connection's framing by itself. It is kept as it was, apart from the Connection: close line.

Checks

• bun scripts/gate.ts and bun x tsc --noEmit pass.
• The file passed 22/22, three runs in a row.
• scripts/design-lint.ts doesn't apply, since nothing under src/web/ changed.

Not done

• The keep_plan tool wasn't available in this run, so I didn't keep a plan on the issue.
• #10226 asks for the server to close the connection as soon as the answer is sent. I made #1242 wait for it.

To try it: bun test test/body-limits.test.ts

🤖 Generated with Claude Code

Left #10215 open: 1 unticked item remain (bun test test/body-limits.test.ts reports under 10 s on the crew machine, quoted in the pull with before and after; no test over 1 s.).