Skip to content

perf(web): R.3 — the board stops waiting its turn - #317

Merged
Willi363363 merged 1 commit into
perf/home-history-limitfrom
perf/home-board-in-parallel
Sep 18, 2026
Merged

Willi363363 merged 1 commit into
perf/home-history-limitfrom
perf/home-board-in-parallel

Conversation

@Willi363363

@Willi363363 Willi363363 commented Sep 18, 2026 •

Copy link
Copy Markdown
Owner

Third step of track R, carrying its own R.3 ⬜ → ✅ tick. Based on #316,
not on staging: both steps edit the
same twenty lines of readHome, and resolving that in GitHub's interface is the
one thing method/01-git-flow.md forbids. Rebased onto staging and retargeted
as soon as #316 merges.

readBoard was awaited on its own line, and only then did readHome open the
Promise.all below it. Nothing in that group fed the board and the board fed
none of them — same viewerId, same clock. The guest path had it twice over:
the day's tile waited behind the board and needs nothing from it either.

The measurement, including the one that says nothing

On this machine the change does not measure. 40 runs of the guest home
against a local Postgres:

before   4.45 ms median
after    4.81 ms median

Inside the noise — a loopback round trip is smaller than the variance. Recorded
that way rather than dressed up, because 09-query-debt.md's rule is that a
note without a measurement is a guess, not that every measurement must flatter
the change.

What does measure is the depth of the waterfall. 25 ms injected per read,
which is roughly a Neon round trip from a Vercel function:

signed in   50.8 ms → 25.8 ms
guest       50.6 ms → 25.4 ms

Exactly one round trip, on both paths. That is where this pays: not on a
developer's loopback, on the deployment the project actually runs.

The tests assert the structure, not a timing

Every read a home needs is in flight before any of them has answered — which is
the finding stated as an assertion, and does not depend on how fast anything is.
Both cases fail against the old shape.

A promise created before a branch is the shape track Q spent five steps on,
so two more cases say a failing board is a rejected readHome on both paths.
Probed by floating it on purpose: the suite then reports 9 passed, 1 error and
the job fails with no failing case — 10-test-debt.md's own Vitest entry,
arriving from the other side.

Checks

  • turbo run test --force --concurrency=1: 10 of 10 tasks, 3 872 passed, 0 skipped.
  • turbo run typecheck lint --force: 22 of 22.
  • pnpm format:check: green.
  • 09-query-debt.md loses its third entry, back at exactly 200 lines.

🤖 Generated with Claude Code

https://claude.ai/code/session_01DsMRxyowr2Ch5RwiAa3t9z

@vercel

vercel Bot commented Sep 18, 2026 •

Copy link
Copy Markdown

The latest updates on your projects. Learn more about Vercel for GitHub.

Project Deployment Actions Updated
wikifake Ready Ready Preview Sep 18, 2026 11:51am UTC

@qodo-code-review

Copy link
Copy Markdown

ⓘ Qodo reviews are paused because your trial has ended. Ask your workspace admin to add credits to resume reviews. Manage billing

`readBoard` was awaited on its own line, and only then did `readHome` open the
`Promise.all` below it. Nothing in that group fed the board and the board fed
none of them: same `viewerId`, same clock. The guest path had it twice over —
the day's tile waited behind the board and needs nothing from it.

**On this machine the change does not measure.** 40 runs of the guest home
against a local Postgres gave 4.45 ms before and 4.81 ms after: inside the
noise, a loopback round trip being smaller than the variance. Recorded that
way rather than dressed up.

What does measure is the depth of the waterfall, with 25 ms injected per read —
roughly a Neon round trip from a Vercel function:

    signed in   50.8 ms → 25.8 ms
    guest       50.6 ms → 25.4 ms

Exactly one round trip, on both paths.

The test asserts the structure and not a timing: every read a home needs is in
flight before any of them has answered. Both cases fail against the old shape.

A promise created before a branch is what track Q spent five steps on, so two
more cases say a failing board is a rejected `readHome` on both paths rather
than an unhandled rejection. Probed by floating it on purpose: the suite then
reports `9 passed, 1 error` and the job fails with no failing case — which is
`10-test-debt.md`'s own entry about Vitest, arriving from the other side.

`turbo run test --force --concurrency=1`: 10 of 10 tasks, 3 872 passed, 0
skipped. Typecheck and lint green over all 22 tasks.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01DsMRxyowr2Ch5RwiAa3t9z
@Willi363363
Willi363363 force-pushed the perf/home-history-limit branch from 63f1462 to ceeb121 Compare September 18, 2026 11:48
@Willi363363
Willi363363 force-pushed the perf/home-board-in-parallel branch from 7a2d30e to 7c2d711 Compare September 18, 2026 11:48
@Willi363363
Willi363363 merged commit 29a17ce into perf/home-history-limit Sep 18, 2026
11 of 12 checks passed

This branch was successfully deployed

1 active deployment
Preview — 7c2d711c Deployed Sep 18, 2026 by vercel[bot]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant