feat: per-block direction detection in Auto mode #2
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,71 @@ | ||
| # Auto Mode: Per-Block Direction Detection | ||
|
|
||
| **Date:** 2026-08-04 | ||
| **Status:** Approved | ||
| **Scope:** `src/content.ts` only (Auto mode CSS + JS) | ||
|
|
||
| ## Problem | ||
|
|
||
| In Auto mode, once a bubble is tagged `.YBYrtl`, response paragraphs get | ||
| `unicode-bidi: plaintext`, which sets each paragraph's base direction from its | ||
| **first strong character**. In a Hebrew conversation this causes: | ||
|
|
||
| 1. Paragraphs that start with an English word flip to LTR and align left. | ||
| 2. List markers (bullets/dots) follow that direction and land on the wrong side. | ||
| 3. Mixed Hebrew/English lines scramble punctuation and word order, hard to read. | ||
|
|
||
| ## Decision | ||
|
|
||
| Direction is decided **per block element** by *presence* of RTL characters, not | ||
| by first character: | ||
|
|
||
| - Block **contains** Hebrew/Arabic/Persian → `dir="rtl"` → right-aligned, | ||
| bullet on right, embedded English isolated inline. | ||
| - Block has **no** RTL characters (pure English heading, free-standing code-ish | ||
| line) → `dir="ltr"` → left-aligned, natural. | ||
|
|
||
| User-approved trade-off: a lone English word on its own line goes left; a Hebrew | ||
| paragraph containing lots of English still goes right. | ||
|
|
||
| ## Mechanism | ||
|
|
||
| ### JS (extends the existing Auto-mode observer in `RTL_AUTO_JS_CODE`) | ||
|
|
||
| - For each `.YBYrtl` bubble, walk block elements — `p`, `li`, `h1`–`h6`, | ||
| `blockquote` — **only inside markdown containers** (`[class*="root_"]`); | ||
| tool/thinking/todo UI reuses the same tags and must not be tagged. | ||
| - Additionally skip anything inside every container the LTR overrides protect: | ||
| `pre`, `code`, `codeBlockWrapper_`, `thinking_`/`thinkingContent_`, | ||
| `toolUse_`/`toolSummary_`/`toolBody_`/`toolResult_`/`toolReference_`, | ||
| `todoList_`/`todoListContainer_` — a native `dir` attribute on a child is not | ||
| neutralized by direction rules on its container. | ||
| - Set `dir="rtl"` or `dir="ltr"` per the RTL-char test | ||
| (`/[--ۿݐ-ݿﭐ-﷿ﹰ-]/`). | ||
| - Re-scan on mutations, debounced — same pattern as the BiDi stripper — so a | ||
| streamed line that starts in English flips right once Hebrew arrives. | ||
|
|
||
| ### CSS (in `AUTO_RTL_RULES`) | ||
|
|
||
| - Remove `unicode-bidi: plaintext` and the blanket `text-align: right` from | ||
| response-paragraph rules. | ||
| - Add: | ||
| - `[dir="rtl"]` blocks → `direction: rtl; text-align: right; unicode-bidi: isolate`. | ||
| - `[dir="ltr"]` blocks → `direction: ltr; text-align: left`. | ||
| - Bubble-level `direction: rtl` layout rules stay (bubble alignment unchanged). | ||
| - Prompt input rules stay as-is (first-char live detection is correct while | ||
| typing). | ||
|
|
||
| ## Out of Scope | ||
|
|
||
| - Active / Always / LTR-Always modes (their semantics are explicit, not | ||
| detected). | ||
| - Plan Preview. | ||
| - Input field. | ||
|
|
||
| ## Verification | ||
|
|
||
| - `npm run build` clean; `npm test` (concurrency) still passes. | ||
| - Manual: in Auto mode with a mixed Hebrew/English response confirm — | ||
| English-first mixed line right-aligned and readable, bullets on the right, | ||
| pure-English heading stays left, code blocks untouched, streaming reply | ||
| settles correctly. |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,92 @@ | ||
| /** | ||
| * Regression guard for the Auto-mode per-block direction walker | ||
| * (Codex round-1 P2, PR #2): the walker must only tag blocks inside | ||
| * markdown containers, and its skip selector must cover every container | ||
| * the LTR overrides protect — a native dir attribute on a child is not | ||
| * neutralized by direction rules on its container. | ||
| */ | ||
| const fs = require('fs'); | ||
| const path = require('path'); | ||
|
|
||
| const src = fs.readFileSync(path.join(__dirname, '..', 'src', 'content.ts'), 'utf8'); | ||
|
|
||
| const walker = src.match(/\/\* Per-Block Direction[\s\S]*?\n\}\)\(\);/); | ||
| if (!walker) { | ||
| console.error('FAIL — per-block direction walker not found in content.ts'); | ||
| process.exit(1); | ||
| } | ||
| const w = walker[0]; | ||
|
|
||
| const failures = []; | ||
|
|
||
| // 1. Walker must query blocks from markdown roots, never bubble-wide. | ||
| if (!/querySelectorAll\('\[class\*="root_"\]'\)/.test(w)) { | ||
| failures.push('walker no longer scopes its block query under [class*="root_"] markdown containers'); | ||
| } | ||
|
|
||
| // 2. Skip selector must include every protected container class. | ||
| const skipSel = (w.match(/var SKIP_SEL = '([^']+)'/) || [])[1] || ''; | ||
| for (const cls of [ | ||
| 'codeBlockWrapper_', 'thinkingContent_', 'thinking_', | ||
| 'toolUse_', 'toolSummary_', 'toolBody_', 'toolResult_', 'toolReference_', | ||
| 'todoList_', 'todoListContainer_', | ||
| ]) { | ||
| if (!skipSel.includes(cls)) failures.push(`SKIP_SEL is missing protected container "${cls}"`); | ||
| } | ||
| for (const tag of ['pre', 'code']) { | ||
| if (!new RegExp(`(^|,)${tag}(,|$)`).test(skipSel)) failures.push(`SKIP_SEL is missing "${tag}"`); | ||
| } | ||
|
|
||
| // 3. CSS dir rules must keep the thinking-block guard (Copilot round-2, PR #2): | ||
| // per-block [dir] styling must never apply under thinkingContent_. | ||
| for (const dir of ['rtl', 'ltr']) { | ||
| const re = new RegExp( | ||
| `\\[class\\*="root_"\\]:not\\(\\[class\\*="thinkingContent_"\\] \\[class\\*="root_"\\]\\) :is\\([^)]*\\)\\[dir="${dir}"\\]` | ||
| ); | ||
| if (!re.test(src)) { | ||
| failures.push(`CSS [dir="${dir}"] block rule lost its thinkingContent_ guard`); | ||
| } | ||
| } | ||
|
|
||
| // 4. Direction detection must ignore text inside skipped containers | ||
| // (Codex round-4 P2, PR #2): a code block quoting Hebrew inside an | ||
| // English list item must not flip the item to RTL. | ||
| if (!/function hasOwnRtl/.test(w) || !/createTreeWalker/.test(w)) { | ||
| failures.push('walker direction detection no longer excludes skipped-container text (hasOwnRtl/TreeWalker missing)'); | ||
| } | ||
| if (!/hasOwnRtl\(el\)/.test(w)) { | ||
| failures.push('tagBlocks no longer uses skip-aware hasOwnRtl for direction detection'); | ||
| } | ||
|
|
||
| // 5. Independently nested blocks' RTL text must not flip the parent, while | ||
| // loose-list paragraphs still count toward their li (Codex round-5/6 P2s): | ||
| // ownership stops at nested li/blockquote, not at p/headings. | ||
| if (!/INDEPENDENT_SEL = 'li,blockquote'/.test(w) || !/function ownsText/.test(w)) { | ||
| failures.push('hasOwnRtl lost the independent-nested-block ownership rule (ownsText/INDEPENDENT_SEL)'); | ||
| } | ||
| if (!/ownsText\(el, p\)/.test(w)) { | ||
| failures.push('hasOwnRtl no longer consults ownsText for direction detection'); | ||
| } | ||
|
|
||
| // 6. The observer must never full-scan on unrelated mutations (Codex round-5 | ||
| // P2): pure-English streaming outside bubbles must not re-walk history. | ||
| if (/pendingFull/.test(w) || /scanAll\(\)/.test(w.split('Debounced watcher')[1] || '')) { | ||
| failures.push('observer regained a full-scan fallback (pendingFull/scanAll in mutation path)'); | ||
| } | ||
| if (!/addedNodes/.test(w)) { | ||
| failures.push('observer no longer inspects addedNodes for new bubbles'); | ||
| } | ||
|
|
||
| // 7. Anchors need their own inline bidi run (Codex round-3 P2, PR #2): | ||
| // unicode-bidi is not inherited, so block-level isolation alone leaves | ||
| // URL punctuation reorderable by the surrounding RTL context. | ||
| if (!/\[class\*="root_"\]:not\(\[class\*="thinkingContent_"\] \[class\*="root_"\]\) a \{\s*\n\s*unicode-bidi: plaintext;/.test(src)) { | ||
| failures.push('anchor unicode-bidi rule missing from Auto-mode CSS'); | ||
| } | ||
|
|
||
| if (failures.length) { | ||
| console.error('FAIL — dir-walker scope regression:'); | ||
| for (const f of failures) console.error(' - ' + f); | ||
| process.exit(1); | ||
| } | ||
| console.log('PASS — dir walker scoped to markdown roots, skip list covers all protected containers'); |
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Bumping the extension to 0.5.1 here leaves both version fields in
package-lock.jsonat 0.5.0. This makes the committed package metadata inconsistent and causes tooling that reads the lockfile to report the previous release; regenerate or update the lockfile so its top-level andpackages[""]versions are also 0.5.1.Useful? React with 👍 / 👎.