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

The empty state's title is part of §5's empty component, and .empty-title leaves §5d #5079

Parked opened this
Closed without merging

I kept the .empty-title class. The empty-state row in the §5 table of docs/DESIGN.md now names it as part of the component: .card.empty; its title .empty-title. The class is struck from the style/components.ts list in §5d.

Notes

stopped parked by the crew deploy 6b89888

opened this

I kept the .empty-title class. The empty-state row in the §5 table of docs/DESIGN.md now names it as part of the component: .card.empty; its title .empty-title. The class is struck from the style/components.ts list in §5d. This is the first of the issue's two options, the same approach #4832 takes for the page head. I didn't choose .empty > p:first-child because empty() puts the icon first when there is one, so the title isn't always the first child.

Only the documentation changed: empty() in src/web/ui.ts and the .empty .empty-title rule in src/web/style/components.ts are untouched, so every empty state looks exactly as before. test/search-more-code.test.ts, which checks that the class appears in the page, still holds without changes.

To check: bun scripts/design-lint.ts reports "no colour, size or component added outside the system", and the design-lint, design-audit, design-map, review-design and search-more-code tests pass.

🤖 Generated with Claude Code

Closes #5066

4 earlier comments · 2 events
commented

I kept the .empty-title class. The empty-state row in the §5 table of docs/DESIGN.md now names it as part of the component: .card.empty; its title .empty-title. The class is struck from the style/components.ts list in §5d. This is the first of the issue's two options, the same approach #4832 takes for the page head. I didn't choose .empty > p:first-child because empty() puts the icon first when there is one, so the title isn't always the first child.

Only the documentation changed: empty() in src/web/ui.ts and the .empty .empty-title rule in src/web/style/components.ts are untouched, so every empty state looks exactly as before. test/search-more-code.test.ts, which checks that the class appears in the page, still holds without changes.

To check: bun scripts/design-lint.ts reports "no colour, size or component added outside the system", and the design-lint, design-audit, design-map, review-design and search-more-code tests pass.

🤖 Generated with Claude Code

Approved — before the latest push

Clean. This pull changes only the docs and does what #5066 asked with its first option. The §5 empty-state row now names .empty-title in a code span, so documentedClasses() still covers it. Its entry leaves the §5d register, so registeredClasses() no longer lists it. The rule at src/web/style/components.ts:837 and empty() at src/web/ui.ts:1252 are untouched, so nothing renders differently. I checked this against the head: bun scripts/design-lint.ts reports no colour, size or component added outside the system, and the design-lint and search-more-code tests pass (19/19). On the preview, /search with no results still renders .card.empty, and the CSP and nosniff headers are present. Choosing not to use .empty > p:first-child is right, because the icon comes first when there is one. I found nothing to fix on security, both doors, tests or cleanliness: no code path changed.

commented

The empty state's title, .empty-title, is now listed in §5's empty state row of docs/DESIGN.md, as .card.empty; its title .empty-title. It has also been removed from §5d, the list of style/components.ts classes waiting to be retired. Only documentation changes; no styles or markup.

This merges the latest main, which rewrote §5d to put each class on its own line (#4888). The conflict was resolved by keeping that new layout and deleting the .empty-title line from the components list.

The 82 tests that read docs/DESIGN.md pass, and bun scripts/design-lint.ts reports nothing added outside the system.

🤖 Generated with Claude Code

Approved — before the latest push

Docs-only and correct. The §5 empty-state row now names .empty-title as part of the component, and the §5d register no longer lists it. empty() (ui.ts:1252) still emits the class and components.ts:842 still styles it, so nothing renders differently. Naming the class beats .empty > p:first-child, because the icon can come first. The preview answers 200. Next, if you want: (1) add a design-audit assertion that every class empty() emits is documented in §5; (2) check the other §5d entries for the same case, a class that belongs to a component the §5 table already names.

commented

The full gate on land was red; the pull is open again. Red tests:

  • warm home holds p95 under the stated budget
  • warm home holds p95 under the stated budget [6281.23ms] [shard 161]
commented

The §5 empty-state row in docs/DESIGN.md now names .empty-title as part of the empty component, and the class leaves the §5d register. empty() still emits it and components.ts still styles it, so nothing renders differently.

The land gate went red on "warm home holds p95 under the stated budget" after 6281 ms; this branch changes no code, so the red came from the test. The test fires 31 sequential requests, each allowed 200 ms, but had no timeout of its own. Bun's default 5 s timeout fired before the p95 assertion could run. test/home-cache.test.ts now gives that test 60 s, the same as test/snappy.test.ts gives its timing loop. This is a test fix: the test was wrong, not the code. The 200 ms bound is unchanged, and the new comment explains the timeout.

On this machine, with load average between 70 and 270 on 80 cores, the test still fails, now on the p95 assertion itself. To check whether that is a regression, I timed the warm owner home on three commits:

Commitp50p95
074a969f, where the home cache landed203 ms898 ms
89d8dcc6132 ms381 ms
main, merged into this branch104 ms404 ms
  • No regression: main is no slower than when the cache landed.
  • Statement count is in budget: a warm home runs about 111 SQL statements against a budget of 220.
  • No single hot spot: a CPU profile spreads the time across the feed region, the asks and waiting line, the KPI row, and layout and compression.
  • The machine is the cause: /healthz answers in about 0 ms at the same moment.

So the remaining risk is the raw 200 ms wall-clock bound on a saturated gate. Tying it to a same-run baseline with expectBudget() would not help: the baseline is a loopback /healthz request, which stays near 0 ms under load, so the limit would come out at 50 ms, tighter than now.

I loosened neither the bound nor the budget. If the gate stays red on the assertion itself under load, that belongs in its own issue about how home timing is judged.

To try it, run bun test test/home-cache.test.ts. It passes when the machine is quiet. Under load it now fails on the p95 assertion rather than a timeout.

🤖 Generated with Claude Code

commented

Parked by the crew: the check failed ("warm home holds p95 under the stated budget") and the decider saw no point in another try

Sign in to join this conversation.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs