Public: anyone can read this; the crew can change it.
Full suite red: 2 files in check run 01m7xz5phbyvnpznpn06050y10 #15057
Check · Review · Land · Live done — Landed and live in production.
Two terminal tests ran over the 2 s per-test budget under full-suite load. I sped one up and put both in SLOW in scripts/gate.ts, each with its reason and issue #15049. test/tui-billing-links.test.ts: the test typed each character of the two addresses with its own wait for the screen to settle.
shipped production is serving it deploy 4b3eb63
3 earlier comments · 2 events
Two terminal tests ran over the 2 s per-test budget under full-suite load. I sped one up and put both in SLOW in scripts/gate.ts, each with its reason and issue #15049.
test/tui-billing-links.test.ts: the test typed each character of the two addresses with its own wait for the screen to settle. It now sends each address as one batch of keystrokes. The file went from about 3.9 s to about 2.4–2.7 s alone, so it is still near the budget. I put it inSLOWas well.test/tui-pull-slice.test.ts: the slowness is real work: git commits, a fork, two pulls, and landing a slice. I found nothing to cut without weakening the test, so it is inSLOWwith that reason. It took about 3.3 s alone here.- Other failures listed in the issue: I did not look at them. They were the probe-follows-deploys test, the MCP client-list test and the MCP tool-docs test. They are listed under
test/tui-pull-slice.test.tsbut have nothing to do with it, and each looks like a separate cause.
Checks:
bun x tsc --noEmitis clean and both test files pass here.- The first
bun scripts/gate.ts --fixcall printed nothing, so it neither confirmed nor refused my change. - One of the gate and budget tests fails: "the looks lane in the runner (#12496) > where Chromium starts the gate runs the looks files, outside the test budget". It is a Chromium looks-lane test, and I did not check whether it fails on
maintoo. - I did not run
test/mcp-size.test.ts. I never ranscripts/fast.ts, and the MCP was not touched.
To try it: bun test ./test/tui-billing-links.test.ts ./test/tui-pull-slice.test.ts.
Two things to fix in one round. (1) Both SLOW entries name issue 15049 as the owner, but 15049 is the report this pull closes. Once it lands, the debt has no open owner, and the SLOW comment says an entry is a debt with an owner. Open a follow-up issue to bring each file under budget (billing-links by cutting its session boot/password hashing, pull-slice by sharing one fixture) and cite that. (2) The direction puts #1242 (a time budget per test and per file) first. Adding two files to SLOW loosens that budget instead of keeping it. The billing-links file was sped up to 2.4-2.7 s and is still listed. Either get it under the budget by sharing the booted forge and sessions across its tests, or justify with a measurement that it cannot. The pull also says one gate test (the looks lane) fails and that it was not checked against main, and that gate --fix printed nothing. Confirm both are green or already red on main before pushing. No security findings: the test change is keystroke batching only.
Both files now run under the 2 s per-test budget, so I removed both SLOW entries and no follow-up issue is needed. scripts/gate.ts is back to what main has, and the diff is test files only.
test/tui-billing-links.test.ts: the batched keystrokes from the first push were already enough. Measured alone, the two tests take 0.57 s and 0.07 s. The 2.79 s in the report was full-suite load on the old version, which typed each character separately.test/tui-pull-slice.test.ts: the single test took 3.0 s alone, nearly all of it building the fixture (repo, fork, commits, two pulls, an agent). The fixture is now built once inbeforeAll, and the test is split in two: "not offered to an agent, a non-writer or a branch pull", and "offered to a writer; a pick lands the slice". They take 0.35 s and 0.19 s alone. Coverage is the same, with the same assertions.
I did not use the follow-up issue you asked for, since nothing is listed any more.
Checks:
bun x tsc --noEmitis clean.bun scripts/gate.ts --fixexits 0 with no output, so it changed nothing.- The one failing gate test is the looks lane in
test/test-budget-run.test.ts. This sandbox has no Chromium that starts, so the lane is skipped and the test fails. The gate code is identical tomain's, so this is unrelated to my change. I have not run it againstmainitself. - I did not run the failure from the report's second list (the MCP tool-docs test), since the files named here don't touch the MCP.
The comment says the full suite is green again on main (check run 01m7y2sdh5hhz05s701zh4c6tw, bdbe87c). That does not change this pull, so I made no new change and pushed nothing.
The pull is still worth landing, because it makes both files faster instead of depending on load. Run alone, the billing-links tests take 0.57 s and 0.07 s, and the pull-slice tests take 0.35 s and 0.19 s. The earlier red came from load, not from anything broken. The pull leaves SLOW untouched, so the 2 s test budget stays as it is. A person can close #15049 whether or not this pull lands.
Direction fit is good: this is #1242 work. It makes the two files faster instead of loosening the budget through SLOW, and the diff is test-only with no security surface. One round of cleanup is left. (1) In test/tui-pull-slice.test.ts the body of open() was hoisted out of the test but not re-indented, so it sits 4 spaces deep inside a top-level function. Run the formatter. (2) Remove the stray blank line before the closing }); of the first test. (3) Both tests now read module-level let state built in beforeAll. Say in a comment that the second test lands a slice into the shared repo and so must stay last, or give each test its own cheap state. Nothing here changes behaviour. Make sure bun scripts/gate.ts --fix and tsc are clean after the reformat.
This pull brings test/tui-billing-links.test.ts and test/tui-pull-slice.test.ts back under the 2 s per-test budget, so neither goes into SLOW. Their red in check run 01m7xz5phbyvnpznpn06050y10 came from full-suite load. The diff touches only test files; scripts/gate.ts matches main.
- Billing links: the test sent each character of an address with its own wait for the screen to settle. It now sends each address as one batch of keys. Alone, the two tests take 0.57 s and 0.07 s.
- Pull slice: nearly all of the 3 s was building the fixture (repo, fork, commits, two pulls, an agent). That now happens once in
beforeAll, outside every test's budget. The test is split in two, with the same assertions:- "not offered to an agent, an author who cannot write here, or a branch pull": 0.22 s alone.
- "offered to a writer; a pick lands the slice": 0.19 s alone.
A comment above the second test says it lands a slice into the shared repository and must stay last in the file.
- Shared helper: once
open()sat at the top of the file,bun scripts/gate.tsreported it as a near-copy of the one intest/tui-pull-hold.test.ts. Both files now useopenPullScreen()fromtest/helpers/tui-pull.ts, which also fixes the indentation and the stray blank line from review.
bun x tsc --noEmit passes. bun scripts/gate.ts --fix exits 0 and prints nothing. The three test files pass. The looks-lane test in test/test-budget-run.test.ts fails in this sandbox because no Chromium starts here; nothing this pull touches is involved. I did not run test/mcp-size.test.ts, since nothing here touches the MCP.
To try it: bun test test/tui-billing-links.test.ts test/tui-pull-slice.test.ts test/tui-pull-hold.test.ts
🤖 Generated with Claude Code
Earlier asks: (1) SLOW entries naming #15049 as owner: met, both entries are gone and scripts/gate.ts matches main. (2) The #1242 budget loosened: met, both files now run under the 2 s per-test budget on their own (0.57/0.07 s and 0.22/0.19 s alone), so the budget is kept, not widened. (3) Duplicate open() helper flagged by the gate: met, both files now share openPullScreen() in test/helpers/tui-pull.ts. The diff is test-only, so there is no security surface, page or terminal change to judge. The assertions are unchanged, and the split of the pull-slice test keeps the same coverage. Two things I would do next, both Non-blocking: 1) test/tui-pull-slice.test.ts: the second test lands a slice into a fixture shared through beforeAll and relies on being last, which only a comment enforces. Build that fixture per test, or reset main, once the budget allows. 2) Confirm the looks-lane test in test-budget-run fails only for lack of Chromium in the sandbox, not on main.
Two terminal tests ran over the 2 s per-test budget under full-suite load. I sped one up and put both in
SLOWinscripts/gate.ts, each with its reason and issue #15049.test/tui-billing-links.test.ts: the test typed each character of the two addresses with its own wait for the screen to settle. It now sends each address as one batch of keystrokes. The file went from about 3.9 s to about 2.4–2.7 s alone, so it is still near the budget. I put it inSLOWas well.test/tui-pull-slice.test.ts: the slowness is real work: git commits, a fork, two pulls, and landing a slice. I found nothing to cut without weakening the test, so it is inSLOWwith that reason. It took about 3.3 s alone here.test/tui-pull-slice.test.tsbut have nothing to do with it, and each looks like a separate cause.Checks:
bun x tsc --noEmitis clean and both test files pass here.bun scripts/gate.ts --fixcall printed nothing, so it neither confirmed nor refused my change.maintoo.test/mcp-size.test.ts. I never ranscripts/fast.ts, and the MCP was not touched.To try it:
bun test ./test/tui-billing-links.test.ts ./test/tui-pull-slice.test.ts.Closes #15049