kayg/calternal
Watch
1
0
Fork
You've already forked calternal
 0
Code Issues 190 Pull requests 61 Projects Releases Packages Wiki Activity Actions 26

port: code block language picker and syntax colours (approved mockup) #903

Closed
kayg wants to merge 0 commits from port/codeblock into dev
pull from: port/codeblock
merge into: kayg:dev
Conversation 9 Commits - Files changed - +1013 -16
kayg commented 2026-11-03 18:02:25 +00:00
Owner
Copy link

Ports the approved mockup's code block language picker (apps/mockup CodeLang.svelte, branch review/calendar-2) and adds syntax colours. Work in progress: the gate numbers and screenshots are added as comments below.

What it does

  • A code block's top right corner holds a button that says the language (Plain text, Rust, …). It sits in the place of the opening fence, which is never drawn. Pressing it opens the mockup's glass menu: Last used, Common, More, a filter you type into, ↑ ↓ ↵ esc, a small spring from the button for a pointer, none for the keyboard. The pick is written after the opening fence (```rust); plain text is a bare fence. Nothing is detected from the code.
  • Colours (keywords, strings, comments, numbers, calls) use the mockup's palette and tk-* classes, light and dark.
  • Mockup values win; desktop UI text stays 14 px (--text-ui). The fence lines that are not drawn are the block's 11 px top and bottom padding, as in the mockup.

Cost on the typing path

  • Colours are a lazy chunk (colours), fetched only when a note has a fenced block that names a language we colour. A tiny gate (code/gate.ts) is all that ships with the editor; typing elsewhere costs one look at the line being typed on.
  • Each language's raw lezer parser is its own lazy chunk (code-javascript (also TypeScript), code-python, code-rust, code-css, code-json), fetched only for a language that is in the note, plus code-runtime (the shared @lezer/lr). Shell and SQL use the mockup's own regular-expression rules (no grammar to fetch).
  • A change inside a code block parses and repaints that one block; a change that adds or removes a fence, or a parser arriving, repaints all code blocks once. Blocks over 40,000 characters stay plain. Text outside code costs nothing.
  • CodeMirror's language wrappers (@codemirror/lang-*) and the nested markdown parse were tried first and dropped: they pulled about 7 KB brotli of shared helpers (@codemirror/language, @lezer/common, autocomplete) into the first paint bundle. The raw parsers keep first paint at +1.5 KiB (233.2 → 234.7 KiB, budget 240).
  • Budget lines are added for the new lazy chunks only (perf/budgets.json), with the total updated to the sum.

Decisions

  • The language word is resolved at press time from the page (posAtDOM), and the menu follows its fence through changes while it is open; a stored position went stale when the server rewrote the properties after a save and a quick second pick changed the wrong line (found by --repeat-each).
  • @lezer/* parsers are direct dependencies of packages/editor (bun.lock updated).
  • No duplicate open PR found for this work (searched open PRs for code block / language / highlight).

Tests

  • packages/editor unit tests (511 pass), apps/web unit tests (96 pass), svelte-check (only the existing review4-signout errors).
  • apps/web/e2e/code-block.spec.ts (6 cases) and the perf test type.code_note_prose / type.code_block in apps/web/perf/typing.perf.ts (a note of 400 paragraphs and 80 code blocks).

Not ported

  • The mockup's read-only code file view (Who owes whom, line numbers) is the attachments port's, not this one.
  • Languages: the nine of the mockup (plain, JavaScript, TypeScript, Python, Rust, Shell, JSON, CSS, SQL). Others show their word as the label and stay plain.

🤖 Generated with Claude Code

Ports the approved mockup's code block language picker (`apps/mockup` `CodeLang.svelte`, branch `review/calendar-2`) and adds syntax colours. Work in progress: the gate numbers and screenshots are added as comments below. ## What it does - A code block's top right corner holds a button that says the language (`Plain text`, `Rust`, …). It sits in the place of the opening fence, which is never drawn. Pressing it opens the mockup's glass menu: *Last used*, *Common*, *More*, a filter you type into, ↑ ↓ ↵ esc, a small spring from the button for a pointer, none for the keyboard. The pick is written after the opening fence (` ```rust `); plain text is a bare fence. Nothing is detected from the code. - Colours (keywords, strings, comments, numbers, calls) use the mockup's palette and `tk-*` classes, light and dark. - Mockup values win; desktop UI text stays 14 px (`--text-ui`). The fence lines that are not drawn are the block's 11 px top and bottom padding, as in the mockup. ## Cost on the typing path - Colours are a **lazy chunk** (`colours`), fetched only when a note has a fenced block that names a language we colour. A tiny gate (`code/gate.ts`) is all that ships with the editor; typing elsewhere costs one look at the line being typed on. - Each language's **raw lezer parser** is its own lazy chunk (`code-javascript` (also TypeScript), `code-python`, `code-rust`, `code-css`, `code-json`), fetched only for a language that is in the note, plus `code-runtime` (the shared `@lezer/lr`). Shell and SQL use the mockup's own regular-expression rules (no grammar to fetch). - A change inside a code block parses and repaints that one block; a change that adds or removes a fence, or a parser arriving, repaints all code blocks once. Blocks over 40,000 characters stay plain. Text outside code costs nothing. - CodeMirror's language wrappers (`@codemirror/lang-*`) and the nested markdown parse were tried first and dropped: they pulled about 7 KB brotli of shared helpers (`@codemirror/language`, `@lezer/common`, `autocomplete`) into the first paint bundle. The raw parsers keep first paint at +1.5 KiB (233.2 → 234.7 KiB, budget 240). - Budget lines are added for the new lazy chunks only (`perf/budgets.json`), with the total updated to the sum. ## Decisions - The language word is resolved at press time from the page (`posAtDOM`), and the menu follows its fence through changes while it is open; a stored position went stale when the server rewrote the properties after a save and a quick second pick changed the wrong line (found by `--repeat-each`). - `@lezer/*` parsers are direct dependencies of `packages/editor` (`bun.lock` updated). - No duplicate open PR found for this work (searched open PRs for code block / language / highlight). ## Tests - `packages/editor` unit tests (511 pass), `apps/web` unit tests (96 pass), svelte-check (only the existing `review4-signout` errors). - `apps/web/e2e/code-block.spec.ts` (6 cases) and the perf test `type.code_note_prose` / `type.code_block` in `apps/web/perf/typing.perf.ts` (a note of 400 paragraphs and 80 code blocks). ## Not ported - The mockup's read-only code file view (`Who owes whom`, line numbers) is the attachments port's, not this one. - Languages: the nine of the mockup (plain, JavaScript, TypeScript, Python, Rust, Shell, JSON, CSS, SQL). Others show their word as the label and stay plain. 🤖 Generated with [Claude Code](https://claude.com/claude-code)
kayg added 17 commits 2026-11-03 18:02:25 +00:00 
Lazy grammars per language, painted as tk-* classes by a field that repaints only the changed block.
The opening fence's place holds the language button, which asks the app to open its menu.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The nested markdown parse and CodeMirror's language wrappers pulled 7 KB of shared helpers into the first
paint bundle. Each block is parsed on its own instead, by the parser for its fence word.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A stored position went stale when the server rewrote the properties after a save, so a quick second pick
changed the wrong line. The menu now follows its fence through changes while it is open.
test(web): give saves and the file 15 s on a slow browser
Some checks failed
CI / server (pull_request) Has been cancelled
Details
CI / rust (pull_request) Has been cancelled
Details
CI / web (pull_request) Has been cancelled
Details
Markdown extraction contract / markdown-native-contract (pull_request) Has been cancelled
Details
Markdown extraction contract / markdown-editor-contract (pull_request) Has been cancelled
Details
f4372f3c42
kayg commented 2026-11-03 18:02:47 +00:00
Author
Owner
Copy link

Product (left) vs approved mockup (right), 1440x900 and 390x844, light and dark. Taken by me and checked; the mockup is review/calendar-2 on its own page, so its surroundings differ.

state product mockup
port-codeblock-corner-1440-dark
port-codeblock-corner-1440-light
port-codeblock-corner-390-dark
port-codeblock-corner-390-light
port-codeblock-menu-1440-dark
port-codeblock-menu-1440-light
port-codeblock-menu-390-dark
port-codeblock-menu-390-light
Product (left) vs approved mockup (right), 1440x900 and 390x844, light and dark. Taken by me and checked; the mockup is `review/calendar-2` on its own page, so its surroundings differ. | state | product | mockup | |---|---|---| | port-codeblock-corner-1440-dark | ![](https://git.kayg.org/attachments/3babbddc-55ed-42fb-b60b-1f5a74224b91) | ![](https://git.kayg.org/attachments/efc09868-9c91-4a3c-bfed-d2e428a39ef1) | | port-codeblock-corner-1440-light | ![](https://git.kayg.org/attachments/90bd540e-b836-4323-a73d-56fa0c14cacc) | ![](https://git.kayg.org/attachments/9b1af97f-914d-424b-a374-0ac373d37575) | | port-codeblock-corner-390-dark | ![](https://git.kayg.org/attachments/d60a5da7-ca13-4ab1-b204-2405dcd13110) | ![](https://git.kayg.org/attachments/efeb0b92-94a0-44c0-bc2e-d83cd2afa6d1) | | port-codeblock-corner-390-light | ![](https://git.kayg.org/attachments/a3382f9b-0d62-4353-af82-bc2dac77a321) | ![](https://git.kayg.org/attachments/69bd35e2-9a88-4286-b4dc-f900b2f9e5c2) | | port-codeblock-menu-1440-dark | ![](https://git.kayg.org/attachments/693ffc69-fe8c-44ba-b97b-bf6de49a6b3d) | ![](https://git.kayg.org/attachments/c8f45eb8-d1f9-4fe7-8fae-e86c9208a57b) | | port-codeblock-menu-1440-light | ![](https://git.kayg.org/attachments/be082887-df84-4bbf-814e-3b5706ce9854) | ![](https://git.kayg.org/attachments/a00145ea-b918-4800-9d4a-39e9f15ba465) | | port-codeblock-menu-390-dark | ![](https://git.kayg.org/attachments/7fecfce8-4bed-43e7-8241-5efe8b75a15d) | ![](https://git.kayg.org/attachments/e6bfabc2-b479-440c-ba24-044a789b015a) | | port-codeblock-menu-390-light | ![](https://git.kayg.org/attachments/abc9a20a-385c-4e26-be77-426559a47981) | ![](https://git.kayg.org/attachments/68f4885d-2b1c-4edc-8dc0-d3cccfa8479e) |
port-codeblock-corner-1440-dark-mockup.png
307 KiB
port-codeblock-corner-1440-light-mockup.png
315 KiB
port-codeblock-corner-390-dark-product.png
48 KiB
port-codeblock-menu-1440-dark-mockup.png
337 KiB
port-codeblock-corner-1440-light-product.png
253 KiB
port-codeblock-corner-390-dark-mockup.png
94 KiB
port-codeblock-corner-390-light-mockup.png
100 KiB
port-codeblock-corner-390-light-product.png
61 KiB
port-codeblock-menu-1440-dark-product.png
200 KiB
port-codeblock-menu-1440-light-product.png
285 KiB
port-codeblock-menu-390-dark-mockup.png
126 KiB
port-codeblock-menu-390-dark-product.png
72 KiB
port-codeblock-menu-390-light-mockup.png
132 KiB
port-codeblock-menu-390-light-product.png
88 KiB
port-codeblock-menu-1440-light-mockup.png
346 KiB
port-codeblock-corner-1440-dark-product.png
173 KiB
kayg commented 2026-11-03 20:58:11 +00:00
Author
Owner
Copy link

First paint, exact bytes (brotli, index chunk; dev 41f2efa3 vs this head 78b4c77a)

  • dev: 238,797; this PR: 239,117; delta +320 B (0.31 KiB). First-paint CSS: 19,250 vs 19,250 (identical).
  • It cannot be exactly zero: something in first paint has to notice a fence. What stays there: lib/lazy-code.ts (one view plugin: a regex on the note when it opens, and a regex on typed backticks/tildes until the extension is in) plus the dynamic-import preload map entry. Everything else is lazy.
  • Moved out of first paint: the language button, fence padding, syntax colours, their stylesheet, the language menu (code-ui.svelte chunk, 4.4 KiB), and every parser (code-runtime, code-javascript, code-python, code-rust, code-css, code-json). packages/editor's editor.ts, index.ts and syntax/hidden.ts and cm.css are byte-identical to dev. The colours read the parsers' own styleTags through getStyleTags (a few hundred bytes of the highlighter's shared module) instead of highlightTree/tagHighlighter (1.5 KB).
  • A note without a fence fetches nothing for code (asserted by code-block.spec.ts); a note with a fence fetches code-ui.svelte once, then the parsers for the languages it names. Typing a fence into a note without code loads it too (asserted).
  • Known trade-off: for a note with a fence the button and the 11 px padding arrive when the lazy chunk does (a few tens of ms after the text), so the block settles once.

Gates at 78b4c77a

  • code-block.spec.ts (7 cases), --repeat-each=5, CI=1 --retries=0 --workers=1: chromium-1440 35/35, chromium-390 35/35 (dev-01), webkit-1440 35/35, webkit-390 35/35 (dev-02).
  • editor unit tests 511 pass, web 96 pass, tsc --noEmit clean, svelte-check only the existing review4-signout errors.

Typing latency on the perf VMs (ABAB, dev-vs-dev control, both locks)

Taken 11-03 on perf-01 with perf-lock held on BOTH perf VMs for the whole run. Chromium at 4x CPU slowdown, 10k Home, 4 rounds, order reversed every other round. Cases: dev (41f2efa3), dev2 (the same dist again: the dev-vs-dev control), cb (#903 head), h887 (#887 head c8afa867), hb (#904 head). Each cell is the median over the 4 rounds of that run's p50 (or p95, or CPU per key), ms:

20
metric (median over rounds of per-run value; ms)   dev dev2 cb h887 hb
type.code_block                       59.8(4)    62.7(4)    64.7(4)    65.1(4)    63.5(4)
type.code_block_cpu                   24.6(4)    27.9(4)    29.9(4)    27.5(4)    27.5(4)
type.code_block|p95                  147.4(4)   151.1(4)   155.7(4)   153.2(4)   151.2(4)
type.code_note_prose                  62.1(4)    68.1(4)    67.0(4)    67.7(4)    69.0(4)
type.code_note_prose_cpu              30.0(4)    33.3(4)    33.2(4)    33.1(4)    32.9(4)
type.code_note_prose|p95             156.1(4)   160.1(4)   161.7(4)   160.8(4)   160.2(4)
type.long_note                        61.6(4)    63.2(4)    61.9(4)    63.5(4)    65.7(4)
type.long_note_cpu                    26.9(4)    28.8(4)    28.6(4)    29.0(4)    29.2(4)
type.long_note|p95                   149.9(4)   152.4(4)   153.1(4)   153.1(4)   154.6(4)

Reading it: dev and dev2 are identical bits and differ by up to 6 ms (code_note_prose), so that is the resolution. #903 against dev2: within -1.1 to +2.0 ms p50 and -0.4 to +2.0 ms CPU per key; #904 against #887: within +0.4 to +2.2 ms p50 and about +0.2 to +0.4 ms... see the table; no row separates a PR from its base by more than the control does. The earlier +10 ms (one run, perf-01 under load 9, no control) is not reproduced. The per-keystroke work was also cut: the #903 gate now looks only at typed backticks/tildes, and #904 does nothing on a keystroke until a hidden list exists.

## First paint, exact bytes (brotli, `index` chunk; dev 41f2efa3 vs this head 78b4c77a) - dev: **238,797**; this PR: **239,117**; delta **+320 B** (0.31 KiB). First-paint CSS: 19,250 vs 19,250 (identical). - It cannot be exactly zero: something in first paint has to notice a fence. What stays there: `lib/lazy-code.ts` (one view plugin: a regex on the note when it opens, and a regex on typed backticks/tildes until the extension is in) plus the dynamic-import preload map entry. Everything else is lazy. - Moved out of first paint: the language button, fence padding, syntax colours, their stylesheet, the language menu (`code-ui.svelte` chunk, 4.4 KiB), and every parser (`code-runtime`, `code-javascript`, `code-python`, `code-rust`, `code-css`, `code-json`). `packages/editor`'s `editor.ts`, `index.ts` and `syntax/hidden.ts` and `cm.css` are byte-identical to dev. The colours read the parsers' own `styleTags` through `getStyleTags` (a few hundred bytes of the highlighter's shared module) instead of `highlightTree`/`tagHighlighter` (1.5 KB). - A note without a fence fetches nothing for code (asserted by `code-block.spec.ts`); a note with a fence fetches `code-ui.svelte` once, then the parsers for the languages it names. Typing a fence into a note without code loads it too (asserted). - Known trade-off: for a note with a fence the button and the 11 px padding arrive when the lazy chunk does (a few tens of ms after the text), so the block settles once. ## Gates at 78b4c77a - `code-block.spec.ts` (7 cases), `--repeat-each=5`, `CI=1 --retries=0 --workers=1`: chromium-1440 35/35, chromium-390 35/35 (dev-01), webkit-1440 35/35, webkit-390 35/35 (dev-02). - editor unit tests 511 pass, web 96 pass, `tsc --noEmit` clean, svelte-check only the existing `review4-signout` errors. ## Typing latency on the perf VMs (ABAB, dev-vs-dev control, both locks) Taken 11-03 on perf-01 with `perf-lock` held on BOTH perf VMs for the whole run. Chromium at 4x CPU slowdown, 10k Home, 4 rounds, order reversed every other round. Cases: `dev` (41f2efa3), `dev2` (the same dist again: the dev-vs-dev control), `cb` (#903 head), `h887` (#887 head c8afa867), `hb` (#904 head). Each cell is the median over the 4 rounds of that run's p50 (or p95, or CPU per key), ms: ``` 20 metric (median over rounds of per-run value; ms) dev dev2 cb h887 hb type.code_block 59.8(4) 62.7(4) 64.7(4) 65.1(4) 63.5(4) type.code_block_cpu 24.6(4) 27.9(4) 29.9(4) 27.5(4) 27.5(4) type.code_block|p95 147.4(4) 151.1(4) 155.7(4) 153.2(4) 151.2(4) type.code_note_prose 62.1(4) 68.1(4) 67.0(4) 67.7(4) 69.0(4) type.code_note_prose_cpu 30.0(4) 33.3(4) 33.2(4) 33.1(4) 32.9(4) type.code_note_prose|p95 156.1(4) 160.1(4) 161.7(4) 160.8(4) 160.2(4) type.long_note 61.6(4) 63.2(4) 61.9(4) 63.5(4) 65.7(4) type.long_note_cpu 26.9(4) 28.8(4) 28.6(4) 29.0(4) 29.2(4) type.long_note|p95 149.9(4) 152.4(4) 153.1(4) 153.1(4) 154.6(4) ``` Reading it: dev and dev2 are identical bits and differ by up to 6 ms (code_note_prose), so that is the resolution. #903 against dev2: within -1.1 to +2.0 ms p50 and -0.4 to +2.0 ms CPU per key; #904 against #887: within +0.4 to +2.2 ms p50 and about +0.2 to +0.4 ms... see the table; no row separates a PR from its base by more than the control does. The earlier +10 ms (one run, perf-01 under load 9, no control) is not reproduced. The per-keystroke work was also cut: the #903 gate now looks only at typed backticks/tildes, and #904 does nothing on a keystroke until a hidden list exists.
kayg commented 2026-11-03 20:58:22 +00:00
Author
Owner
Copy link

Exact deltas for the table above (this PR minus the same-bits control dev2, ms): p50 code_block +2.0, code_note_prose -1.1, long_note -1.3; CPU per key +2.0, -0.1, -0.2. Minus dev: p50 +4.9, +4.9, +0.3. The control itself differs from dev by +2.9, +6.0, +1.6 p50, so none of this separates the PR from noise; the first run's +10 ms is not reproduced.

Exact deltas for the table above (this PR minus the same-bits control `dev2`, ms): p50 code_block +2.0, code_note_prose -1.1, long_note -1.3; CPU per key +2.0, -0.1, -0.2. Minus `dev`: p50 +4.9, +4.9, +0.3. The control itself differs from `dev` by +2.9, +6.0, +1.6 p50, so none of this separates the PR from noise; the first run's +10 ms is not reproduced.
kayg commented 2026-11-05 04:08:40 +00:00
Author
Owner
Copy link

First-paint diet (BM15, #947). Head 4c4cdda1 (unchanged by me).

First-paint bytes are brotli q11 from apps/web/scripts/bundle-size.mjs (JS of the first-paint closure = the index chunk; CSS of the first stylesheet). Dev baseline 41f2efa3: 238,797 B JS, 19,250 B CSS. A second figure, "hash-normalised", is the same chunk with the 8-character chunk-file hashes replaced by a fixed string: the hashes are random, and they move the real brotli number by ±300 B between two builds of identical code, so a few-hundred-byte claim is only meaningful in the normalised figure (dev: 238,804). Chunk budgets: every new lazy chunk's line is taken from the first-script line or the idle-screens line so the total stays 404.2; perf/web.sh --bundle-only --gate passes on this head.

first JS first CSS lazy JS
dev 238,797 19,250 122,936
this PR 239,117 (+320; normalised 238,930 = +126) 19,250 (0) 205,936 (+83 KB: the language parsers)

The language list, menu and colours are already lazy (lazy-code.ts is the trigger). The first-screen cost is that trigger (470 raw) plus one export of @lezer/highlight (getStyleTags) that code/blocks.ts imports and the editor shares, plus the preload-list entries of the new chunks. Not ≤ 0 on its own; covered by #875's offset (running total in #947).

code-block.spec: 28 passed on all four projects. Budget: unavoidable total change. The seven code-* chunks are 83 KB of new lazy JS; the total is the sum of the chunk lines, so it cannot stay 404.2 here (it is 488.3 as the author set it). That is lazy weight, not first paint. I left the author's lines.

**First-paint diet (BM15, #947).** Head `4c4cdda1` (unchanged by me). First-paint bytes are brotli q11 from `apps/web/scripts/bundle-size.mjs` (JS of the first-paint closure = the `index` chunk; CSS of the first stylesheet). Dev baseline `41f2efa3`: **238,797 B JS, 19,250 B CSS**. A second figure, "hash-normalised", is the same chunk with the 8-character chunk-file hashes replaced by a fixed string: the hashes are random, and they move the real brotli number by ±300 B between two builds of identical code, so a few-hundred-byte claim is only meaningful in the normalised figure (dev: 238,804). Chunk budgets: every new lazy chunk's line is taken from the first-script line or the idle-screens line so the total stays 404.2; `perf/web.sh --bundle-only --gate` passes on this head. | | first JS | first CSS | lazy JS | |---|---|---|---| | dev | 238,797 | 19,250 | 122,936 | | this PR | 239,117 (**+320**; normalised 238,930 = **+126**) | 19,250 (0) | 205,936 (+83 KB: the language parsers) | The language list, menu and colours are already lazy (`lazy-code.ts` is the trigger). The first-screen cost is that trigger (470 raw) plus one export of `@lezer/highlight` (`getStyleTags`) that `code/blocks.ts` imports and the editor shares, plus the preload-list entries of the new chunks. Not ≤ 0 on its own; covered by #875's offset (running total in #947). `code-block.spec`: **28 passed** on all four projects. **Budget: unavoidable total change.** The seven `code-*` chunks are 83 KB of new lazy JS; the total is the sum of the chunk lines, so it cannot stay 404.2 here (it is 488.3 as the author set it). That is lazy weight, not first paint. I left the author's lines.
kayg commented 2026-11-05 08:26:59 +00:00
Author
Owner
Copy link

Total JS budget 488.3 KB brotli approved by kayg on 11-05 (option A on #947). No other budget changes.

Total JS budget 488.3 KB brotli approved by kayg on 11-05 (option A on #947). No other budget changes.
kayg commented 2026-11-05 15:58:24 +00:00
Author
Owner
Copy link

#903 per-PR gates at 3a634dd1: BLOCKED on first-paint JS (+182 B brotli)

Head = 505064a7 + merge of origin/dev b5e8dcb2 (no conflicts). Everything else passes:

gate result
lint / svelte-check / vitest pass / 2 errors, 16 warnings = dev's / 107 + 55 pass
Chromium 1440+390 and WebKit 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block Chromium 96 passed, 0 failed; WebKit 88 passed, 0 failed
perf/web.sh --bundle-only --gate rc 0 against its own perf/budgets.json, but first-paint JS is bigger than dev (see below)
check-ui-primitives not on dev yet; the train found styles/code-block.css:89 (local :active { scale }) in this PR: must become the shared press (after #941)

First paint, same machine, built from each head (brotli): JS 232.862 → 233.040 KB (dev → #903), the index chunk +182 B (lazyCode ViewPlugin in Editor.svelte, lib/lazy-code.ts); first CSS 18.571 = 18.571. The new parser chunks are lazy (total JS 363.4 → 444.5 KB).
Two things for the orchestrator: (1) the +182 B needs an equal cut in the first chunk, or an explicit OK; (2) perf/budgets.json raises web.bundle.total_js_brotli 404.2 → 488.3 and adds 7 lazy-chunk lines: that is a budget raise, not approved by me.

## #903 per-PR gates at `3a634dd1`: BLOCKED on first-paint JS (+182 B brotli) Head = `505064a7` + merge of origin/dev `b5e8dcb2` (no conflicts). Everything else passes: | gate | result | |---|---| | lint / svelte-check / vitest | pass / 2 errors, 16 warnings = dev's / 107 + 55 pass | | Chromium 1440+390 and WebKit 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block | Chromium 96 passed, 0 failed; WebKit 88 passed, 0 failed | | `perf/web.sh --bundle-only --gate` | rc 0 against its own `perf/budgets.json`, **but first-paint JS is bigger than dev** (see below) | | `check-ui-primitives` | not on dev yet; the train found `styles/code-block.css:89` (local `:active { scale }`) in this PR: must become the shared press (after #941) | First paint, same machine, built from each head (brotli): JS **232.862 → 233.040 KB (dev → #903)**, the `index` chunk +182 B (`lazyCode` ViewPlugin in `Editor.svelte`, `lib/lazy-code.ts`); first CSS 18.571 = 18.571. The new parser chunks are lazy (total JS 363.4 → 444.5 KB). Two things for the orchestrator: (1) the +182 B needs an equal cut in the first chunk, or an explicit OK; (2) `perf/budgets.json` raises `web.bundle.total_js_brotli` 404.2 → 488.3 and adds 7 lazy-chunk lines: that is a budget raise, not approved by me.
kayg commented 2026-11-05 20:41:57 +00:00
Author
Owner
Copy link

#903 after #941: Pill/press done, specs green, first-paint bytes NOT yet ≤ dev

Head 25374a8b0, dev 850c23a9 merged in. The language button is bounce (shared press); the local :active rule is gone (its transition lists scale). perf/budgets.json total 488.3 is kayg-approved.

gate result
lint / svelte-check / vitest pass / 2 errors, 16 warnings = dev's / pass
check-ui-primitives clean
Chromium (dev-01) + WebKit (dev-02), 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block, pill-press all pass (Chromium 102 passed, WebKit 94 passed, 0 failed)
first-paint JS brotli (dev 232.565 KB, first CSS 17.952 = dev) 232.794: +229 B

Not GREEN until the first-paint bytes are cut by the amount above.

## #903 after #941: Pill/press done, specs green, first-paint bytes NOT yet ≤ dev Head `25374a8b0`, dev `850c23a9` merged in. The language button is `bounce` (shared press); the local `:active` rule is gone (its transition lists `scale`). `perf/budgets.json` total 488.3 is kayg-approved. | gate | result | |---|---| | lint / svelte-check / vitest | pass / 2 errors, 16 warnings = dev's / pass | | `check-ui-primitives` | **clean** | | Chromium (dev-01) + WebKit (dev-02), 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block, pill-press | all pass (Chromium 102 passed, WebKit 94 passed, 0 failed) | | first-paint JS brotli (dev 232.565 KB, first CSS 17.952 = dev) | **232.794: **+229 B**** | Not GREEN until the first-paint bytes are cut by the amount above.
kayg commented 2026-11-06 11:39:33 +00:00
Author
Owner
Copy link

#903 re-gated on dev c808ef67: GREEN at 571ebdd1

Merged dev (the vite.config.ts chunk groups of #975 and #903 both kept; the shared-press fix is kept). total_js_brotli 488.3 is the kayg-approved one.

gate result
lint / svelte-check / vitest web / ui primitives / check-ui-primitives pass / 0 errors, 16 warnings (= dev) / 109 / 59 / clean
first-paint JS brotli (this PR vs dev c808ef67 225.407 KB vs prod 71827028 233.040 KB) 225.708 KB: +301 B vs dev, −7.33 KB vs prod (first CSS 17.952 = dev, prod 18.376). Option A: dev after this merge stays ≤ prod.
perf/web.sh --bundle-only --gate rc 0. idle-screens 51.147; total JS 448.3 (code chunks are lazy)
Chromium (dev-01) 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block, pill-press, picker 106 passed, 0 failed
WebKit (dev-02) same 98 passed, 0 failed
Rust n/a
## #903 re-gated on dev `c808ef67`: GREEN at `571ebdd1` Merged dev (the `vite.config.ts` chunk groups of #975 and #903 both kept; the shared-press fix is kept). `total_js_brotli` 488.3 is the kayg-approved one. | gate | result | |---|---| | lint / svelte-check / vitest web / ui primitives / `check-ui-primitives` | pass / 0 errors, 16 warnings (= dev) / 109 / 59 / clean | | **first-paint JS brotli (this PR vs dev `c808ef67` 225.407 KB vs prod `71827028` 233.040 KB)** | **225.708 KB: +301 B vs dev, −7.33 KB vs prod** (first CSS 17.952 = dev, prod 18.376). Option A: dev after this merge stays ≤ prod. | | `perf/web.sh --bundle-only --gate` | rc 0. idle-screens 51.147; total JS 448.3 (code chunks are lazy) | | Chromium (dev-01) 1440+390: code-block, editor, lazy-ui, lazy-offline, lazy-styled, preload, block-no-text-sel, vim-block, pill-press, picker | 106 passed, 0 failed | | WebKit (dev-02) same | 98 passed, 0 failed | | Rust | n/a |
kayg commented 2026-11-06 11:41:27 +00:00
Author
Owner
Copy link

Merged into dev by the orchestrator (U1 GREEN, credit bank).

Merged into dev by the orchestrator (U1 GREEN, credit bank).
kayg closed this pull request 2026-11-06 11:41:28 +00:00
Some checks failed
CI / server (pull_request) Has been cancelled
Details
CI / rust (pull_request) Has been cancelled
Details
CI / web (pull_request) Has been cancelled
Details
Markdown extraction contract / markdown-native-contract (pull_request) Has been cancelled
Details
Markdown extraction contract / markdown-editor-contract (pull_request) Has been cancelled
Details

Pull request closed

This pull request cannot be reopened because the branch was deleted.
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels   
backend
Rust server

  
design
Mockups, UI and TASTE

  
docs
ADRs, CONTEXT, TASTE

  
idea
Parked idea, never worked until promoted

  
infra
Deploy, CI, VMs

  
perf
Performance and budgets

  
reliability
Data integrity, corruption traps

  
web
Web client

No labels
backend
design
docs
idea
infra
perf
reliability
web
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
kayg/calternal!903
No description provided.