t41372/long-screen - CodSpeed

long-screen

Blog Docs Changelog

Performance History

Latest Results

feat(export): add adjustable screenshot output sizes Let PNG downloads and clipboard copies shrink the kept crop by percentage or a custom longest-edge limit, with live output dimensions. Preserve native tiles, paged exports and archival projects. Downsample rows in Rust with alpha-weighted area averaging so long images do not need a full-image canvas. Update all 21 catalogues and cover cropping, resizing, clipboard output and responsive layouts. Validation: formatting, type checks and lint passed, along with 387 Deno tests, five Rust tests and Chrome/WebKit export regressions. The full coverage gate still reports existing catalogue floors, analytics coverage and the scenarios floor.

main

15 days ago

perf(core): walk a native row in three spans so its sampling loop stays branch-free Measuring each native row's middle half (for rounded bar tops) put a range check and a second sum into the learner's per-sample loop, and CodSpeed's layers_learn benchmarks lost 16.5-17.7% on test_3/6/7/8. The row is now walked left of the middle half, through it, and right of it, adding to the row sum in the same left-to-right order as one walk, so every statistic is unchanged. layers_learn locally: test_7 3.07 -> 2.67 ms, test_3 1.84 -> 1.59 ms. Scenario fingerprint identical; four real recordings byte-identical at tolerance 0.

fix/open-defects-20261025

17 days ago

docs: record the merged nested scrolling and the round's final comparison The nested-scroll layers landed after the round history was written: CAPABILITIES now lists one level of nested containers as partly supported, the history moves that work out of the withdrawn table (keeping only the rare-scroll pane, 2d), points each kept branch at its saved patches, and adds the final 32-recording comparison against the start: 0.82x time overall, own_a 1.19x, and the source of every recording whose output changed.

fix/open-defects-20261025

17 days ago

ci(codspeed): run the benchmarks only when the Rust workspace changes The kernel and pipeline benchmarks measure code under rust/ alone, yet every pull request and push to main ran them for 15-30 minutes, including README, web-shell and workflow changes that cannot move the numbers. Both triggers now filter on rust/** and this workflow file; Run workflow still starts a run by hand. Pull requests keep comparing against the latest main run that measured the kernels.

main

24 days ago

feat(web): count page views and runs on the hosted site with PostHog The hosted site now counts how often the printer is used. On long-screen.yi-ting.com and long-screen.pages.dev, without Do Not Track or Global Privacy Control, main.js adds assets/analytics.js (src/analytics.ts), which bundles PostHog's slim no-external SDK (posthog-js 1.438.12) and sends a page view plus run_started and run_finished events queued by src/ui/analytics.ts. Events carry settings, the core version, the codec family, a frame-count bucket, elapsed seconds, the canvas count and diagnostic codes; never a file name, pixels or message text. A stopped run ends as `stopped`, a failed start as `start_failed`, and a run the browser killed is reported as `interrupted` on the next load. The SDK is bundled rather than loaded from PostHog's CDN because the page is cross-origin isolated and PostHog's scripts send no Cross-Origin-Resource-Policy header. Cookieless mode with in-memory persistence keeps nothing in the browser, `advanced_disable_flags` skips the remote config so the dashboard cannot switch on replay or surveys, and the CSP gains https://us.i.posthog.com in connect-src only. Self-hosted copies, local development and the portable build never load the SDK, and main.js has no dynamic import(), which the portable build refuses. The help page's privacy section (zh and en, static/index.html byte-identical to the zh catalogue), the README and AGENTS.md say what is counted and by whom. tests/browser/analytics.test.ts runs the demo and a stopped run in Chrome and WebKit under the deployed CSP, checks the exact events and that no payload, cookie or storage key identifies the user, and that a GPC page never fetches the SDK; LONGSCREEN_ANALYTICS_LIVE=1 sends to the real project.

feat/posthog-analytics

24 days ago

docs: record the npm 0.1.1 release through trusted publishing npm now has a trusted publisher for long-screen, and re-running the npm-v0.1.1 run's failed job published 0.1.1 with provenance once the npm environment was approved; all three registries are at 0.1.1.

main

24 days ago

feat(ui): show the core library version beside the commit on the plate The plate's commit hash is exact but hard to read. scripts/build.ts now also stamps the long-screen crate's version (rust/long-screen/Cargo.toml), so the plate reads v0.1.1-dev a210785. The version always shows; the commit link still appears only when the build knows its commit. Both carry a title saying what they are. Checked at 390, 1024, 1440 and 1920 px in en and zh: nothing overflows, the 390 px plate keeps 22 px before the status light with a -dirty label, and the how-to stays above the fold.

feat/quality-20261018

24 days ago

docs: record the new-recordings quality round and its invariants ARCHITECTURE §十 gains the two rules that had no entry yet: the row-vote band on a side the change test leaves empty (translucent iOS 26 bars) and sticky bands that slide at a small speed of their own (Bilibili's late header collapse), each with the measured failure behind it. §十二 records the three speedups that only touch non-SIMD paths or add a colour memo. docs/history/2026-10-quality-round.md keeps the round's root causes, the 32-recording paired comparison against main c65e33e, the routes that were tried and dropped, and what is still open.

feat/quality-20261018

24 days ago

Latest Branches

0%

fix: resolve the defects left open after the 2026-10-18 quality round#12

17 days ago

e985ef5

fix/open-defects-20261025

0%

feat(web): count page views and runs on the hosted site with PostHog#10

24 days ago

8f4e699

feat/posthog-analytics

×2.3

fix: treat screen-fixed bars and floating chrome as fixed UI on new recordings#9

24 days ago

c0cb56a

feat/quality-20261018

© 2026 CodSpeed Technology

Home Terms Privacy Docs