CJK / Unicode
Failure Corpus

Pressing Enter to confirm a kana to kanji conversion in an n-dynamic-tags input creates a tag from the in-progress text instead of just finishing the conversion.

1. Render <n-dynamic-tags> and focus its input.
2. Switch to a Japanese IME, type "とうきょう", press Space to convert to 東京.
3. Press Enter to pick the candidate.
4. A tag is added from the unconfirmed text before the conversion commits.

Skip tag creation while e.isComposing is true; only act on the Enter that fires after compositionend.

Confirming a Japanese IME candidate with Enter in the chat box sends the message instead of committing the conversion.

1. Open the chat input.
2. With a CJK IME, type a phrase and press Space to get conversion candidates.
3. Press Enter to choose a candidate.
4. The message sends with the unconfirmed text.

Return early when e.nativeEvent.isComposing is true. React's synthetic KeyboardEvent has no isComposing field, so read it off nativeEvent (the DOM event), not e.isComposing.

Typing a CJK query in the command palette and pressing Enter to confirm the IME runs the highlighted command instead of finishing the conversion.

1. Open the command palette.
2. Compose a Japanese search term with the IME.
3. Press Enter to commit the conversion.
4. The first/highlighted command fires prematurely.

Add `if (e.isComposing) return` at the top of the palette input keydown handler. Svelte exposes the native KeyboardEvent, so e.isComposing works directly.

On Safari/WebKit, composing an @-mention name in CJK and pressing Enter to confirm the IME replaces the text with whatever mention is highlighted.

1. In Safari, type @ then compose a Japanese name with the IME.
2. Press Enter to confirm the kanji conversion.
3. The composing text is replaced by the highlighted mention option.

In the ENTER branch, return early on event.nativeEvent.isComposing before preventDefault. WebKit reports the commit keydown as which===Enter + isComposing:true; Chromium reports keyCode 229 and never enters this branch.

The Enter used to commit an IME composition also selects the currently highlighted option in a searchable Select.

1. Open a searchable <Select>.
2. Type a CJK query with the IME and press Enter to confirm the conversion.
3. The active option is selected with the unfinished query.

Track composition state (compositionstart/compositionend) and skip option selection while composing.

The time-picker's key handler fires while an IME composition is active, so keystrokes meant for the IME mutate the time value.

1. Focus the time-picker input.
2. Begin an IME composition.
3. Composition keystrokes are intercepted by the picker's key handler instead of the IME.

Skip the picker's keydown handling while the composition is active (isComposing guard).

VAutocomplete's keydown listener triggers during IME composition, so confirming a CJK query navigates or selects options.

1. Open a VAutocomplete.
2. Compose a Japanese query, press Enter/arrow to confirm.
3. The keydown listener acts on the composing keys.

Bail out of the keydown listener when e.isComposing is true.

Single-key shortcuts registered with defineShortcuts fire while composing text, so romaji keystrokes trigger app shortcuts mid-composition.

1. Register a single-key shortcut (e.g. 'g').
2. In an input, compose Japanese text whose romaji includes that key.
3. The shortcut fires during composition.

Ignore key events when e.isComposing (or keyCode 229) before dispatching shortcuts.

A color-picker channel input runs its Enter handler while an IME composition is active, committing before the composition ends.

1. Focus a color channel input.
2. Trigger an IME composition and press Enter.
3. The channel commit fires before compositionend.

Guard the channel input's Enter handler with isComposing.

Renaming a file inline and pressing Enter to confirm a CJK name submits the rename mid-composition.

1. Start an inline file rename.
2. Compose a Japanese filename with the IME.
3. Press Enter to confirm the conversion; the rename submits with the unfinished name.

Skip the rename submit while the IME is composing (isComposing guard).

The Angular chat input submits the message on the Enter that confirms an IME composition.

1. Type a CJK message in the Angular chat input.
2. Press Enter to confirm the IME conversion; the message sends early.

Check event.isComposing (Angular passes the native KeyboardEvent) before submitting.

The Composer sends the message on the Enter used to confirm an IME composition.

1. Compose CJK text in the Composer.
2. Press Enter to pick a candidate; the message sends instead of committing.

Skip send when e.nativeEvent.isComposing is true.

Pressing Enter to confirm CJK composition in the agent session rename input also triggers submit/rename, discarding the composed text or submitting prematurely.

Use Japanese/Chinese/Korean IME in siyuan's session rename field; press Enter to confirm composition; rename fires before composition is complete.

Check event.isComposing (and event.keyCode===229 for legacy browsers) before acting on keydown/keyup Enter. Ignore Enter events during composition.

CJK users pressing Enter to accept an IME candidate also submits the chat message, sending an empty or partial message.

Type Japanese in llmchat chat input using IME; press Enter to select a candidate; message sends immediately.

Guard Enter keydown handler with if (event.isComposing || event.nativeEvent?.isComposing) return.

Confirming CJK IME candidate with Enter in AIaW chat input simultaneously submits the message.

Japanese IME → type → Enter to confirm → message sent before input complete.

Ignore Enter keydown when isComposing is true.

Obsidian Smart Composer chat submits messages on IME Enter key in macOS/Windows with CJK input methods.

Use Japanese IME in Obsidian Smart Composer chat; confirm composition with Enter; chat sends.

Check isComposing flag before processing Enter keydown.

ListFieldInput in Onyx web accepts a list entry on the IME confirmation Enter, adding an empty or partial item.

ListFieldInput with CJK IME → Enter to confirm → list item added prematurely.

Skip Enter action when event.nativeEvent.isComposing is true.

RSuite InputPicker selects the focused item when CJK IME confirmation Enter is pressed, skipping intended input.

RSuite InputPicker with CJK IME → search → Enter to confirm composition → list item selected prematurely.

Guard item selection Enter handler with isComposing check.

Dify tag input adds a tag on IME confirmation Enter, creating malformed tags for CJK users.

Dify web tag input → Japanese IME → Enter to confirm → tag added prematurely.

Check isComposing before handling Enter in tag input.

Multiple Flowise text inputs fire on IME Enter, interrupting CJK text composition.

Flowise text inputs with CJK IME → Enter to confirm → premature action fired.

Add isComposing guard to all remaining Enter keydown handlers.

Langflow inline rename inputs commit the rename on IME Enter, saving partial CJK names.

Langflow node rename → CJK IME → Enter to confirm composition → rename saved with partial text.

Guard Enter handler in inline rename with event.nativeEvent?.isComposing check.

Plane select dropdowns pick an item on IME confirmation Enter, making it impossible for CJK users to search.

Plane dropdown search → Japanese IME → Enter to confirm → item picked mid-composition.

Skip item selection when event.nativeEvent.isComposing is true.

Twenty CRM chat-thread send and attachment rename both fire on IME Enter, breaking CJK workflows.

Twenty CRM chat or attachment rename → CJK IME → Enter to confirm → send/rename triggered.

Check event.nativeEvent?.isComposing before handling Enter.

Excalidraw search navigation (arrow keys, Enter) fires during CJK IME composition, moving selection mid-composition.

Excalidraw search → CJK IME → type → Arrow/Enter during composition → result navigation fires.

Guard navigation key handlers with event.isComposing check.

Payload CMS SearchInput submits on IME confirmation Enter, bypassing the intended search query.

Payload admin search → CJK IME → Enter to confirm → search fires with partial query.

Ignore Enter in SearchInput when event.nativeEvent?.isComposing is true.

Trilium kanban card and column title editing commits on IME Enter, saving partial CJK text.

Trilium board → edit card title → CJK IME → Enter to confirm → title saved with partial text.

Check isComposing before committing title edits on Enter.

big-AGI custom instruction field fires on IME Enter, submitting partial CJK instructions.

big-AGI custom instruction → CJK IME → Enter to confirm → instruction submitted prematurely.

Skip Enter action when event.nativeEvent?.isComposing is true.

LibreChat prompt name, label, and tag inputs all trigger on IME Enter, causing broken CJK input across multiple components.

LibreChat prompt editor → tag or name field → CJK IME → Enter to confirm → submit fires.

isComposing guard across prompt name, label, and dynamic-tag Enter handlers.

Jan AI assistant rename and project dialogs accept partial CJK names when Enter confirms IME composition.

Jan rename dialog → CJK IME → Enter to confirm → dialog committed with partial name.

Guard Enter submit with event.nativeEvent?.isComposing check.

In a controlled <input>/<textarea>, onChange fires for the intermediate keystrokes of an IME composition, so any onChange-driven search or filter runs on half-finished CJK text.

Type Chinese/Japanese through an IME into a controlled input that filters on onChange; the handler runs on every uncommitted composition keystroke.

Track compositionstart/compositionend and suppress onChange handling while a composition is active.

On Windows, Entry.Completed is raised when the user presses Enter inside the IME conversion candidate window, so completion fires before the CJK conversion is committed.

Focus a MAUI Entry on Windows, use a Chinese/Japanese IME, press Enter to pick a candidate; Completed fires mid-conversion.

Do not raise Completed while the IME conversion window is open; fire only once the composition is fully committed.

The spec orders input before compositionend (Chrome and Safari follow it), but Firefox and Edge fire input after compositionend, so isComposing-based guards behave differently per browser.

Log the input and compositionend events while committing a CJK composition in Chrome versus Firefox; their relative order differs.

Define a canonical order in the spec; frameworks should not assume input fires before compositionend.

Every input event during a composition, including the one that commits it on Enter, carries isComposing=true and inputType insertCompositionText, so code cannot detect the commit without also listening for compositionend.

Type s, i, Space to convert to a kanji, change the candidate, then press Enter to commit; the input events for each step are indistinguishable.

Assign a distinct inputType (or property) to the input event that commits the composition.

With Vim mode and a CJK IME, the jk insert-mode escape only fires while the IME is composing in Chinese mode; in direct English input it inserts a literal j and k instead of escaping.

Enable Vim mode, map jk to escape, use a CJK IME such as macOS Pinyin; jk escapes while composing in Chinese mode but not in English mode.

Detect the jk key sequence consistently regardless of the IME composition state.

While composing Chinese text with an IME on Windows, the line of text jumps vertically as the composition updates, which disrupts reading and editing.

Open Zed (Vim mode) on Windows 11 and type Chinese characters through an IME; the text shifts vertically during composition.

Keep the line baseline stable while updating the IME marked-text region during composition.

Warp does not render IME marked (preedit) text, so dead keys and CJK composition show nothing until committed and the user composes blind.

In Warp on macOS, press Option+E (or start a CJK composition); no marked text is shown to indicate the in-progress input.

Render marked text inline via the platform IME API (NSTextInputClient on macOS) so the preedit is visible at the cursor.

Katakana ン before a vowel or Y is romanized without the apostrophe, unlike hiragana ん. シンヨウ becomes SHINYOU (should be SHIN'YOU), so it collides with シニョウ.

const { fromKana } = require('hepburn')
fromKana('しんよう') // SHIN'YOU
fromKana('シンヨウ') // SHINYOU  <- apostrophe dropped

Map katakana ン to N' (matching hiragana) and add the ンー long-vowel digraph as N', so N is never a katakana map key and toKatakana still round-trips (PAN to パン).

The geminated d + small i sequence っでぃ (ddi) has no Hepburn or Kunrei entry, so loanword spellings that use it romanize incorrectly.

Convert a word containing っでぃ; the doubled d before the small i is not produced.

Add the missing っでぃ (ddi) sokuon entry to both the Hepburn and Kunrei tables.

The historical kana ゐ (wi) and ゑ (we) are absent from the mapping, so text containing them is dropped or left unconverted.

Convert text containing ゐ or ゑ (or katakana ヰ/ヱ); the characters pass through unmapped.

Add ゐ/ゑ and their katakana counterparts to the kana mapping.

kana2alphabet does not handle the small katakana ヵ/ヶ (small ka/ke), so counters like 一ヶ月 are mis-romanized.

import jaconv
jaconv.kana2alphabet('ヶ')  # small ke not handled

Add the small ka/ke mappings to kana2alphabet.

The ヲ to ヺ voiced-mark mapping is reversed, so adding or stripping a dakuten on ヲ produces the wrong character.

Add a voiced mark to ヲ; expected ヺ, got the wrong glyph because the mapping is reversed.

Correct the reversed ヲ/ヺ entry in both addVoicedMarks and combineSoundMarks.

kana-romaji library drops the 'z' consonant when romanizing づ, outputting 'u' instead of 'zu'.

romanize('づ') returns 'u' instead of 'zu'.

づ should romanize as 'zu' (Hepburn). Map づ→zu in the kana table.

pykakasi does not romanize half-width katakana correctly, particularly when a half-width voiced or semi-voiced mark (U+FF9E / U+FF9F) follows the base kana.

Convert half-width katakana such as a half-width ka followed by a half-width dakuten; the output is wrong instead of 'ga'.

NFKC-normalize half-width katakana and combining voiced marks to their full-width equivalents before romanization.

Python unidecode transliterates half-width katakana carrying dakuten/handakuten incorrectly, producing artifacts, while hiragana and full-width katakana romanize correctly.

unidecode on half-width ba bi bu be bo returns a wrong string instead of 'babibubebo'.

Pre-compose half-width katakana plus combining voiced marks (NFKC) before the transliteration lookup.

During normalization, the chōonpu (ー) is expanded with the wrong vowel after katakana ヒ and ビ.

Normalize ヒー / ビー; the long-vowel expansion produces the wrong vowel for these two kana.

Fix the ー-expansion table entries for ヒ and ビ.

Full-width/half-width and kana range conversions skip the boundary characters of each range (！ ～ ぁ ゖ ...), so edge code points like ！ (U+FF01) are not converted.

moji('！～').convert('ZE', 'HE').toString()  // boundary chars at range edges are skipped

Convert the first and last character of each range, not just the interior.

Meilisearch's charabia tokenizer incorrectly classifies halfwidth katakana (U+FF65-U+FF9F) and some fullwidth forms as wrong scripts, causing them to be processed by the wrong tokenizer and failing Japanese search.

Index halfwidth katakana text in Meilisearch; search for the same terms; results missing because script detection classifies halfwidth katakana as non-Japanese.

Extend script detection to recognize U+FF65–U+FF9F (halfwidth katakana) as Japanese script.

tabled's text-wrapping splits combining marks (diacritics, dakuten, etc.) away from their base grapheme when calculating width for terminal table cells.

Create a tabled table with text containing combining marks (e.g., 'が' as 'か' + combining voiced mark U+3099); wrapping separates the base and combining mark onto different lines.

Use a grapheme cluster iterator (Unicode UAX #29) when wrapping, never splitting within a grapheme cluster.

In Zed editor, the block cursor over Unicode ambiguous-width characters (East Asian Width 'A' category, e.g., some symbols, box-drawing) is misaligned — the cursor glyph is not centered in the cell.

Open Zed with a CJK font; position cursor on an ambiguous-width character; block cursor appears shifted or misaligned.

Center the block cursor glyph using the rendered cell width rather than the glyph's intrinsic width for ambiguous-width characters.

cli-table3 truncates text by byte/code-unit count rather than code-point count, splitting surrogate pairs in emoji or supplementary CJK characters, producing mojibake in terminal table cells.

Create a cli-table3 table with a column containing emoji (e.g., 🎉) or supplementary CJK characters; set a column width that truncates mid-emoji; output shows garbled characters.

Use a Unicode-aware splitter (spread operator or Array.from) to iterate code points rather than code units when truncating.

Clerk UI's truncateWithEndVisible function uses substring/slice on raw code units in its short-width fallback, splitting surrogate pairs in emoji or non-BMP characters.

A Clerk UI component displaying an email/name containing emoji in the short-width fallback path; the truncated string ends mid-surrogate-pair, showing '?' or garbled chars.

Use Array.from() or spread to split by code points, or use Intl.Segmenter, before truncating.

opentype.js does not clamp cmap format 12/13 character codes to U+10FFFF; malformed fonts with out-of-range codes cause incorrect glyph lookups for supplementary characters.

Load a font with a cmap subtable containing entries beyond U+10FFFF; glyph lookup for supplementary characters (emoji, CJK Extension B+) returns wrong glyph.

Clamp all format 12/13 startCharCode/endCharCode values to 0x10FFFF during parsing.

markdown-it's smart quotes replacement does not handle non-BMP punctuation and symbols (U+10000+); surrounding text with supplementary characters causes wrong quote pairing or no conversion.

markdown-it smartquotes on text adjacent to emoji or supplementary Unicode symbols (e.g., '𝕳ello'); smart quote pairing is incorrect.

Use a regex that is Unicode-aware for the 'whitespace' and 'punctuation' character class checks, or use Array.from for code-point iteration.

Slate rich-text editor does not implement Unicode UAX #29 GB9c rule, splitting Indic conjunct clusters (consonant + virama + consonant sequences) across grapheme boundaries, causing incorrect cursor positioning and deletion in Hindi, Bengali, Tamil, etc.

Type a conjunct consonant in Hindi (e.g., 'क्ष') in Slate; press Backspace; only one codepoint is deleted instead of the full cluster.

Apply Unicode GB9c rule: treat <Indic_Conjunct_Break=Linker> sequences as a single grapheme cluster.

wenmode's is_punctuation function treats Unicode combining marks (Mn category) and format characters (Cf/ZWJ) as punctuation per the CommonMark spec, incorrectly suppressing valid emphasis around CJK text with diacritics or ZWJ sequences.

Markdown text using emphasis (* or _) adjacent to a combining mark or ZWJ character; emphasis fails to render because combining marks are classified as punctuation.

Exclude Unicode General Categories Mn, Mc, Cf from the punctuation classification; add ASCII fast-path to avoid performance regression.

grapheme-splitter breaks ZWJ-joined emoji into parts instead of one grapheme cluster: the rainbow flag splits into its component glyphs, and skin-tone sequences come apart.

new GraphemeSplitter().splitGraphemes('🏳️‍🌈') returns two elements instead of one.

Implement the Unicode emoji ZWJ sequence rules (UTS #51) so a ZWJ-joined emoji stays a single cluster.

_.toArray splits an emoji built from a tag sequence (a subdivision flag) into its component code points instead of returning it as one element.

_.toArray for the England flag emoji (a base flag plus tag characters) returns seven pieces instead of one.

Use a Unicode-aware iterator such as Intl.Segmenter that handles tag sequences when converting a string to an array.

Windows Terminal ignores the text-presentation variation selector U+FE0E, rendering the color-emoji form even when text style is explicitly requested.

Print a text-style sequence such as U+23CF followed by U+FE0E; it renders as a color emoji instead of the text glyph.

Honor VS-15 (U+FE0E) for text presentation and VS-16 (U+FE0F) for emoji presentation, per the Unicode emoji variation sequences.

emoji-regex matches a base character even when it is followed by U+FE0E (the text variation selector), so text-presentation characters are wrongly classified as emoji.

emojiRegex().test('\u2757\uFE0E') returns true even though U+FE0E requests text presentation.

Exclude a match followed by VS-15 (U+FE0E); treat only a trailing VS-16 (U+FE0F) or no selector as emoji.

compileStyledText builds charStyleMap keyed by UTF-16 code-unit length, but formatText later applies the styles by grapheme index (via runes()). The two indexings match for ASCII, but drift apart after any character longer than one code unit: an emoji, a ZWJ sequence, or an astral-plane CJK ideograph (CJK Extension B, e.g. names written with 𠮷). Every style after such a character lands on the wrong grapheme or is dropped.

In styled text, "😀[c]x[/c]" keys the colour style at code unit 2, but runes("😀x") puts x at grapheme index 1, so the style is lost.

Make compileStyledText walk grapheme clusters with the same runes() helper formatText already uses, keying charStyleMap by grapheme index. Normalize the input to NFC up front and consume a whole grapheme per escape so the slice lengths stay consistent.

Onboarding decides whether a spoken answer has enough content with len(transcript.split()) >= 2. str.split() returns 1 for CJK text that has no spaces, so a full answer like 東京に住んでいます is counted as a single word, never reaches the LLM check, and the question stays marked unanswered for Japanese, Chinese, and Korean speakers.

len('東京に住んでいます'.split())  # 1, so word_count >= 2 is False and the answer is rejected

Use the existing CJK-aware _word_count helper (already used by should_discard_conversation) instead of plain split(); it falls back to split() for non-CJK text, so English input is unchanged.

The editor footer's word count and reading time come from TipTap's CharacterCount, whose default wordCounter splits on spaces. CJK scripts have no spaces between words, so a long Japanese or Chinese draft shows "1 word" and "1 min read" even though the published page, which already counts CJK separately, reports the correct time.

Type a CJK paragraph into the editor; CharacterCount's default text.split(' ') returns 1, so the footer reports 1 word and 1 min read.

Configure CharacterCount with a wordCounter that counts CJK characters individually, and derive reading time from the text using the same word/CJK split and rates (200 words/min, 500 CJK chars/min) as the published reading-time util, so the editor and the rendered page agree.

kanji2number cannot parse 萬, the daiji (大字) traditional form of 万 (10,000), so legal and financial documents that use 大字 numerals fail to convert.

from kanjize import kanji2number
kanji2number('萬')  # not recognized as 10000

Map 萬 to 万 (10,000) in kanji2number.

formatjs intl-relativetimeformat ignores the numberingSystem locale option (e.g., 'jpan', 'arab'), always producing Latin numerals in relative time strings.

new Intl.RelativeTimeFormat('ja-JP-u-nu-jpan', {}).format(-1, 'day') returns '1日前' with Latin '1' instead of Japanese numeral '一日前'.

Apply the numberingSystem extension from the locale tag when formatting relative time numbers.

The ja-JP labels for Typography's expand/collapse control were incorrect, so Japanese users saw the wrong 展開/折りたたみ text.

1. Set ConfigProvider locale to ja_JP.
2. Render <Typography.Paragraph ellipsis={{ expandable: true }}>.
3. The expand/collapse label is wrong.

Correct the ja-JP expand/collapse label strings.

weekPlaceholder was missing from the Japanese locale, so the week date-picker showed an English placeholder.

Use the date picker in week mode with the ja locale; the placeholder renders in English.

Add the Japanese weekPlaceholder translation.

Future timestamps in the ja locale used 以内 (within) instead of 後 (later), so '3 minutes from now' rendered as 3分以内 (within 3 minutes), which means the opposite.

timeago a future timestamp with the ja locale; it renders '3分以内' instead of '3分後'.

Use 後 for future time strings in the ja locale.

The Japanese description for a day-of-month step was not scoped to a month, producing a mistranslated cron description.

cronstrue.toString('0 0 1-31/3 * *', { locale: 'ja' }); the description is not scoped to the month.

Scope the ja day-of-month step description to a month.

PrimeLocale Japanese (ja) locale uses 成約 (meaning 'conclusion of a contract') for 'filterConstraint' aria label, which should be 制約 ('constraint/restriction').

PrimeFaces component with aria-label for filter constraint in Japanese reads '成約' (contract) instead of '制約' (constraint).

Change filterConstraint value from '成約' to '制約' in the ja.json locale file.

Vant component library Japanese (ja-JP) locale contains incorrect translations for multiple UI strings.

Vant components with ja-JP locale display mistranslated strings for picker, datetime, and other components.

Correct multiple mistranslated strings in vant/src/locale/lang/ja-JP.ts.

MUI X Japanese (ja-JP) locale has incomplete or incorrect translations for data grid, date picker, and time picker components.

MUI X components with jaJP locale show untranslated English strings or incorrect Japanese.

Improve Japanese translations across multiple MUI X locale strings.

Semi Design's Japanese locale is missing translation strings for the Upload component's crop modal UI.

Semi Design Upload component with crop modal in Japanese locale shows untranslated (English) strings in the crop modal.

Add missing Japanese translation keys for Upload crop modal to ja-JP locale.

NG-ZORRO Angular Japanese (ja_JP) locale is missing quarter placeholder strings, causing runtime errors or blank UI for date range pickers.

NG-ZORRO DatePicker quarter mode with ja_JP locale throws error or shows blank quarter placeholders.

Add missing quarter i18n keys to ja_JP locale file.

FilePond Japanese locale uses '読込中' (loading/reading) for the file processing label, which should be 'アップロード中' (uploading) to accurately describe the action.

FilePond with Japanese locale shows '読込中' during file upload, misleading users into thinking a read is occurring.

Change labelFileProcessing from '読込中' to 'アップロード中' in the Japanese locale.

Arco Design Japanese (ja-JP) locale contains multiple translation errors across component strings.

Arco Design components with ja-JP locale show incorrect Japanese translations.

Correct translation errors in arco-design/arco-design/components/_utils/locale/ja-JP.ts.

jsoneditor's Japanese locale is incomplete; many UI strings remain in English when ja locale is selected.

jsoneditor with locale:'ja' shows English strings for missing Japanese translations.

Fill all missing Japanese translation keys in the ja locale file.

Quasar Framework Japanese locale has mistranslations in editor toolbar labels and tree component strings.

Quasar editor toolbar or tree component in Japanese shows incorrect label text.

Correct mistranslated Japanese strings in quasar/lang/ja.js.

Uppy's Japanese (ja_JP) locale has a broken smart_count plural placeholder in the 'folderAdded' string, causing pluralization to fail and show a raw placeholder.

Uppy file picker in Japanese displays 'folderAdded' with a visible placeholder token instead of correct plural text.

Fix the smart_count placeholder syntax in ja_JP's folderAdded locale string.

Video.js Japanese (ja) locale is missing the translation for the 'Playing in Picture-in-Picture' accessibility string, falling back to English.

Video.js in Japanese locale; PiP mode accessible label shows 'Playing in Picture-in-Picture' in English.

Add 'ピクチャーインピクチャーで再生中' or equivalent to the ja.json locale.

Memos Japanese (ja) locale has 24 missing translation keys; these fall back to English in the UI.

Memos with Japanese locale selected; 24 UI strings display in English.

Fill all 24 missing Japanese translation keys in locales/ja.json.

Appwrite Japanese email templates have 48 missing translation keys, causing emails to be sent in English to Japanese users.

Appwrite with Japanese locale; email notifications contain 48 English fallback strings.

Translate all 48 missing keys in app/config/locale/translations/ja.json email templates.

Medusa dashboard Japanese (ja) locale has 511 missing translation keys, leaving most of the admin UI in English for Japanese users.

Medusa admin dashboard with Japanese locale; majority of UI strings display in English.

Translate all 511 missing keys in packages/admin-ui/ui/src/i18n/translations/ja.json.

Clerk's oauthConsent UI section lacks Japanese translations, showing English strings to Japanese users during OAuth authorization.

Clerk OAuth consent page with Japanese locale; consent section displays in English.

Add Japanese translations for all oauthConsent locale keys.

jp-prefectures.js has the English name of Aichi prefecture (愛知県) set to 'ehime' (which is Ehime prefecture / 愛媛県), causing incorrect prefecture mapping.

jpPrefectures.findByCode(23).enName returns 'ehime' instead of 'aichi'.

Change enName for 愛知県 (code 23) from 'ehime' to 'aichi'.

In the gl (Galician) locale the June parse pattern is /^xun/i. It matches the abbreviation "xun" but not the wide form "xuño", because the third character is ñ, not n. So format then parse round-trips fail for June; the locale's own snapshot already records Invalid Date for June while the other eleven months parse.

const s = format(new Date(2021, 5, 1), 'MMMM', { locale: gl }); // 'xuño'
parse(s, 'MMMM', new Date(), { locale: gl });               // Invalid Date

Widen the June pattern to /^xu[nñ]/i so it matches both "xun" and "xuño" while staying distinct from July ("xul"), mirroring locales such as Catalan that already fold diacritics into their patterns (e.g. /^març/i).

The zh-CN locale is missing the MuiPagination block, so the pagination aria-label and getItemAriaLabel text fall back to English even though every other component in the file is localized. Among all locales only the three Chinese ones omit it; ja-JP and ko-KR already localize this block.

Render a MUI <Pagination> under the zhCN locale and inspect it with a screen reader; the navigation aria-label and per-page item labels are announced in English.

Add the MuiPagination block to zhCN, reusing the noun phrases already in this file's MuiTablePagination.getItemAriaLabel (第一页 / 最后一页 / 下一页 / 上一页) plus 转到 ("Go to") to match the English source.

The Japanese (ja) locale is missing the removeItem and search keys, so Japanese users get the English fallback for two ARIA labels: the per-item remove button (used in selection/multiple.js) and the search field (used in selection/search.js and dropdown/search.js). Both are part of the canonical set in en.js.

Use a multi-select Select2 under the ja locale with a screen reader; the remove-item button and the search field announce their English fallback labels.

Add removeItem (アイテムを削除) and search (検索) to the ja locale, following the existing removeAllItems wording (すべてのアイテムを削除).

weekPlaceholder was left as the English fallback "Select Week" in the Korean (koKR) and Traditional Chinese (zhTW) DatePicker locales, so the week-mode date picker showed an English placeholder. zhCN and jaJP were already translated; these were the remaining CJK locales still showing the fallback.

Use the date picker in week mode under the koKR or zhTW locale; the placeholder renders "Select Week" in English.

Translate weekPlaceholder to 주 선택 (koKR, following the existing 월 선택 / 년 선택 pattern) and 選擇週 (zhTW, the Traditional form of zhCN's 选择周), propagating the jaJP fix from #8114 to the remaining CJK locales.

isAlpha('el-GR') uses the Greek range [Α-ώ] (U+0391–U+03CE) but isAlphanumeric('el-GR') still ends at ω ([0-9Α-ω], U+03C9). So isAlphanumeric rejects ό, ύ, ώ and the uppercase Ό/Ύ/Ώ even though isAlpha accepts them, and common words like νερό or πρώτα pass isAlpha but fail isAlphanumeric.

isAlpha('νερό', 'el-GR')        // true
isAlphanumeric('νερό', 'el-GR') // false  <- ό (U+03CC) sits past the range end ω

Bump the alphanumeric Greek range to [0-9Α-ώ] so it matches the alpha class.

Optimizing or regenerating a character class can move a literal ^ to the front and emit it unescaped, flipping the meaning: [a^] round-trips to [^a] (a negated class).

const rt = require('regexp-tree')
rt.optimize('/[a^]/').toString() // '/[^a]/'  <- now matches the negation

In the generator, escape a leading literal ^ in a non-negative character class so generate(parse(x)) preserves meaning.

Enum member names containing special characters (including non-ASCII / CJK) are emitted unescaped, so the generated enum does not compile.

Generate types from a schema whose enum values contain quotes or special characters; the emitted enum has invalid member identifiers and tsc errors.

Escape enum member names so special characters produce valid output.

The formatter over-escapes a # in the middle of a line because the heading branch of the escape regex is not anchored to line start: 'C# is a language' becomes 'C\# is a language'.

format(parse('C# is a language')) // 'C\# is a language' - the '#' is escaped mid-sentence

Anchor the heading alternative: change #+\s to ^#+\s, mirroring the already-anchored ^* and ^> cases.

The synchronous parse() strips a leading UTF-8 byte-order mark (U+FEFF) but CsvParseStream does not. When a CSV begins with a BOM, common output from Excel and other Windows tools, the first field name arrives as "\uFEFFname" instead of "name", silently corrupting header-based lookups.

const src = ReadableStream.from(['\uFEFFname,age\n', 'Alice,34\n']);
await Array.fromAsync(src.pipeThrough(new CsvParseStream({ skipFirstRow: true })));
// [{ '\uFEFFname': 'Alice', age: '34' }]  <- BOM leaked into the key

Strip the BOM from the first line read by StreamLineReader, matching what parse() already does via its BYTE_ORDER_MARK constant.