test/forks.test.ts runs under 10 s #3597

scripts/test-times.json now records test/forks.test.ts at 1.5 s instead of 28.9 s. That one line is the whole change.

The file was already fast. #1290 rewrote it to share one forge and one git fixture across the file and to wait on event-loop turns instead of a clock or a sleep, and #2138 kept it that way. The 28.9 s figure was written by #515 and never refreshed, so the shard planner was still treating this file as one of the slowest in the gate.

• Time: bun scripts/test-shards.ts --record test/forks.test.ts measured 1.5 s for the file, beforeAll included. Bun only prints a time for a single test when it is slow, and it printed none here.
• Stability: bun test test/forks.test.ts passed ten times in a row: 8 pass, 0 fail, 1.35–1.48 s each.
• What it proves: the same 8 tests as on main, with no test deleted or changed and no production code touched.
• Not done: the record tool also re-sorted three unrelated crew-* entries in the table. I left those out so the diff only touches the line this issue owns.

To try it: bun test test/forks.test.ts.

🤖 Generated with Claude Code

Closes #3579

scripts/test-times.json now records test/forks.test.ts at 1.5 s instead of 28.9 s. That one line is the whole change.

The file was already fast. #1290 rewrote it to share one forge and one git fixture across the file and to wait on event-loop turns instead of a clock or a sleep, and #2138 kept it that way. The 28.9 s figure was written by #515 and never refreshed, so the shard planner was still treating this file as one of the slowest in the gate.

• Time: bun scripts/test-shards.ts --record test/forks.test.ts measured 1.5 s for the file, beforeAll included. Bun only prints a time for a single test when it is slow, and it printed none here.
• Stability: bun test test/forks.test.ts passed ten times in a row: 8 pass, 0 fail, 1.35–1.48 s each.
• What it proves: the same 8 tests as on main, with no test deleted or changed and no production code touched.
• Not done: the record tool also re-sorted three unrelated crew-* entries in the table. I left those out so the diff only touches the line this issue owns.

To try it: bun test test/forks.test.ts.

🤖 Generated with Claude Code

test/forks.test.ts was already well under 10 s. #1290 rewrote it to build one forge and one git fixture per file and to wait on event-loop turns instead of a clock or a sleep. What was still wrong was the record: scripts/test-times.json had listed it at 28.9 s since #515, so the shard planner was still treating it as one of the slowest files in the gate.

What changed

• scripts/test-times.json: test/forks.test.ts is now 1.5 s, measured with bun scripts/test-shards.ts --record test/forks.test.ts. The record tool also re-sorted three crew-* lines; I left those out so only this file's line changes.
• scripts/test-shards.ts (outside the issue's file list, but it's why the first check failed): the planner in src/checks/tiers.ts already decides that a change to test-times.json alone reaches no tests. The script then fell through to the "name the test files to run, or --gate" usage error and exited 2. Any pull that only touches the timing table, like this one, could never go green. An empty fast-tier plan now prints "the change reaches no tests; nothing to run" and exits 0, the same way the empty matrix slice just below it already does.
• test/shards.test.ts: a new test, "a pull no test reaches is green with nothing to run, not a usage error". It builds a throwaway repo whose only commit changes scripts/test-times.json, hands the script an impact map that reaches nothing, and expects exit 0 with the new message. It fails without the fix.

Not changed

• test/forks.test.ts: no test deleted or edited, so its 8 tests prove what they proved before.
• No other test file.

Checked

• bun test test/forks.test.ts passed ten times in a row, 1.38–1.51 s for the whole file.
• test/shards.test.ts passes 69/69 and tsc --noEmit is clean.

To try it: bun test test/forks.test.ts, then bun test test/shards.test.ts -t "no test reaches".

🤖 Generated with Claude Code