@godxjp/ui 32.4.2 → 32.5.0

package/agent/START-HERE.md

@@ -3,7 +3,7 @@
 You are about to write code against a design system you did not author. This file is the whole
 contract. Read it before you write JSX.
 
-**This catalog describes `@godxjp/ui` 32.4.2.** If the project you are editing has a different
+**This catalog describes `@godxjp/ui` 32.5.0.** If the project you are editing has a different
 version in its `package.json`, read the pinned catalog for THAT version instead
 (`…/v<their-version>/agent/…`). A catalog newer than the installed package describes props that do
 not exist yet; older, and it hides props that do. Neither failure announces itself.

package/agent/components/Text.json

@@ -59,9 +59,9 @@
     },
     {
       "defaultValue": "\"normal\"",
-      "description": "Where an over-long unbroken token may break. THE prop for a machine identifier — an email address, a coupon code, a login id, a template key — in a DataTable/TableCell or a Descriptions value: `break=\"anywhere\"` emits `overflow-wrap: anywhere` (and releases a table cell's inherited `nowrap`), so the column shrinks to the viewport instead of scrolling the page. Use it INSTEAD of `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`. NOT `whitespace=\"pre-wrap\"`: that is for typed line breaks, and its `break-word` does not lower min-content, so in a table cell it breaks nothing (measured: a 71-char token holds a 320px table at 630px with pre-wrap, 320px with break=\"anywhere\"). Independent of `whitespace` (both may be set); `truncate` wins (dev builds warn); `clamp` composes.",
+      "description": "Where an over-long unbroken token may break. THE prop for a machine identifier — an email address, a coupon code, a login id, a template key — in a DataTable/TableCell or a Descriptions value: `break=\"anywhere\"` emits `overflow-wrap: anywhere` (and releases a table cell's inherited `nowrap`), so the column shrinks to the viewport instead of scrolling the page. Use it INSTEAD of `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`. NOT `whitespace=\"pre-wrap\"`: that is for typed line breaks, and its `break-word` does not lower min-content, so in a table cell it breaks nothing (measured: a 71-char token holds a 320px table at 630px with pre-wrap, 320px with break=\"anywhere\"). `break=\"phrase\"` is THE prop for Japanese PROSE in a narrow measure — a trainee's katakana name, an organisation, a status in a DataTable cell: it breaks between phrases (`word-break: auto-phrase`, Chrome 119+) instead of between any two characters, never starts a line with a small kana or 「ー」 (`line-break: strict`), adds `text-wrap: pretty`, and releases the cell's `nowrap` (measured in a 974px 8-column table: 「チャン・ヴァ｜ン・ミン」 → 「チャン・｜ヴァン・｜ミン」, 4 mid-katakana breaks → 0). It keeps every phrase whole, so the column is as wide as its longest phrase: a 20+ character katakana compound scrolls the table rather than being cut. Browsers without `auto-phrase` keep their own CJK breaking. Independent of `whitespace` (both may be set); `truncate` wins (dev builds warn); `clamp` composes.",
       "name": "break",
-      "type": "\"normal\" | \"anywhere\""
+      "type": "\"normal\" | \"anywhere\" | \"phrase\""
     },
     {
       "description": "Tabular figures for aligned numbers.",
@@ -195,6 +195,7 @@
   "usage": [
     "DO use `<Text>` for ALL body / inline / caption text instead of a styled `<span>`/`<p>`. Pick `size` from the scale; never write `text-[13px]`/`text-[11px]` or `font-semibold` by hand.",
     "DO use `tone` for colour (`muted`/`primary`/semantic), `tabular` for numbers, `mono` for codes — not `className=\"text-muted-foreground font-mono tabular-nums\"`.",
+    "DO use `break=\"phrase\"` for Japanese prose (a name, an organisation, a status) in a table cell — never `className=\"[word-break:auto-phrase]\"`; it is not for a machine id, which is `break=\"anywhere\"`.",
     "DO use `break=\"anywhere\"` for an email, code or id in a table cell or Descriptions value — never `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`, and never `whitespace=\"pre-wrap\"`, which does not shrink a table column.",
     "DO use `clamp={n}` for a description limited to n lines (card grids) and `truncate` for a one-line ellipsis — never `className=\"line-clamp-2\"` (banned utility). They are mutually exclusive; `clamp` wins.",
     "For a heading, use `<Heading level>` instead of a large-size `<Text>`.",
@@ -205,6 +206,7 @@
     "A muted caption under a value: `<Text size=\"xs\" tone=\"muted\">2026年5月度</Text>`.",
     "A monospace id in a list row: `<Text size=\"xs\" mono tone=\"muted\">RC-204881</Text>`.",
     "A staff email in a narrow DataTable cell at 320px, wrapping instead of scrolling the page: `<Text size=\"sm\" break=\"anywhere\">{staff.email}</Text>`.",
+    "A trainee's katakana name in a DataTable cell, breaking between phrases instead of mid-katakana: `render: (row) => <Text break=\"phrase\">{row.name}</Text>` (pair long-heading columns with `headerWrap: true`).",
     "An emphasized inline figure: `<Text weight=\"medium\" tabular>¥1,240,000</Text>`.",
     "A service-card description clamped to 2 lines on a narrow (390px) index: `<Text as=\"p\" size=\"sm\" tone=\"muted\" clamp={2}>{description}</Text>`.",
     "An issue subject linking out of a table cell, wrapping to two lines: `<Text asChild link truncate={false}><Link href={`/view/${key}`}>{subject}</Link></Text>`.",

package/agent/components.json

@@ -2753,9 +2753,9 @@
       },
       {
         "defaultValue": "\"normal\"",
-        "description": "Where an over-long unbroken token may break. THE prop for a machine identifier — an email address, a coupon code, a login id, a template key — in a DataTable/TableCell or a Descriptions value: `break=\"anywhere\"` emits `overflow-wrap: anywhere` (and releases a table cell's inherited `nowrap`), so the column shrinks to the viewport instead of scrolling the page. Use it INSTEAD of `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`. NOT `whitespace=\"pre-wrap\"`: that is for typed line breaks, and its `break-word` does not lower min-content, so in a table cell it breaks nothing (measured: a 71-char token holds a 320px table at 630px with pre-wrap, 320px with break=\"anywhere\"). Independent of `whitespace` (both may be set); `truncate` wins (dev builds warn); `clamp` composes.",
+        "description": "Where an over-long unbroken token may break. THE prop for a machine identifier — an email address, a coupon code, a login id, a template key — in a DataTable/TableCell or a Descriptions value: `break=\"anywhere\"` emits `overflow-wrap: anywhere` (and releases a table cell's inherited `nowrap`), so the column shrinks to the viewport instead of scrolling the page. Use it INSTEAD of `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`. NOT `whitespace=\"pre-wrap\"`: that is for typed line breaks, and its `break-word` does not lower min-content, so in a table cell it breaks nothing (measured: a 71-char token holds a 320px table at 630px with pre-wrap, 320px with break=\"anywhere\"). `break=\"phrase\"` is THE prop for Japanese PROSE in a narrow measure — a trainee's katakana name, an organisation, a status in a DataTable cell: it breaks between phrases (`word-break: auto-phrase`, Chrome 119+) instead of between any two characters, never starts a line with a small kana or 「ー」 (`line-break: strict`), adds `text-wrap: pretty`, and releases the cell's `nowrap` (measured in a 974px 8-column table: 「チャン・ヴァ｜ン・ミン」 → 「チャン・｜ヴァン・｜ミン」, 4 mid-katakana breaks → 0). It keeps every phrase whole, so the column is as wide as its longest phrase: a 20+ character katakana compound scrolls the table rather than being cut. Browsers without `auto-phrase` keep their own CJK breaking. Independent of `whitespace` (both may be set); `truncate` wins (dev builds warn); `clamp` composes.",
         "name": "break",
-        "type": "\"normal\" | \"anywhere\""
+        "type": "\"normal\" | \"anywhere\" | \"phrase\""
       },
       {
         "description": "Tabular figures for aligned numbers.",
@@ -2889,6 +2889,7 @@
     "usage": [
       "DO use `<Text>` for ALL body / inline / caption text instead of a styled `<span>`/`<p>`. Pick `size` from the scale; never write `text-[13px]`/`text-[11px]` or `font-semibold` by hand.",
       "DO use `tone` for colour (`muted`/`primary`/semantic), `tabular` for numbers, `mono` for codes — not `className=\"text-muted-foreground font-mono tabular-nums\"`.",
+      "DO use `break=\"phrase\"` for Japanese prose (a name, an organisation, a status) in a table cell — never `className=\"[word-break:auto-phrase]\"`; it is not for a machine id, which is `break=\"anywhere\"`.",
       "DO use `break=\"anywhere\"` for an email, code or id in a table cell or Descriptions value — never `className=\"[overflow-wrap:anywhere] break-words whitespace-normal\"`, and never `whitespace=\"pre-wrap\"`, which does not shrink a table column.",
       "DO use `clamp={n}` for a description limited to n lines (card grids) and `truncate` for a one-line ellipsis — never `className=\"line-clamp-2\"` (banned utility). They are mutually exclusive; `clamp` wins.",
       "For a heading, use `<Heading level>` instead of a large-size `<Text>`.",
@@ -2899,6 +2900,7 @@
       "A muted caption under a value: `<Text size=\"xs\" tone=\"muted\">2026年5月度</Text>`.",
       "A monospace id in a list row: `<Text size=\"xs\" mono tone=\"muted\">RC-204881</Text>`.",
       "A staff email in a narrow DataTable cell at 320px, wrapping instead of scrolling the page: `<Text size=\"sm\" break=\"anywhere\">{staff.email}</Text>`.",
+      "A trainee's katakana name in a DataTable cell, breaking between phrases instead of mid-katakana: `render: (row) => <Text break=\"phrase\">{row.name}</Text>` (pair long-heading columns with `headerWrap: true`).",
       "An emphasized inline figure: `<Text weight=\"medium\" tabular>¥1,240,000</Text>`.",
       "A service-card description clamped to 2 lines on a narrow (390px) index: `<Text as=\"p\" size=\"sm\" tone=\"muted\" clamp={2}>{description}</Text>`.",
       "An issue subject linking out of a table cell, wrapping to two lines: `<Text asChild link truncate={false}><Link href={`/view/${key}`}>{subject}</Link></Text>`.",

package/agent/index.json

@@ -48,7 +48,7 @@
   "note": "Pin to the tag that matches the @godxjp/ui version you installed. A catalog newer than your package describes props you do not have; older, and it hides props you do.",
   "read": {
     "live": "https://raw.githubusercontent.com/godx-jp/godxjp-ui/main/agent/index.json",
-    "pinned": "https://raw.githubusercontent.com/godx-jp/godxjp-ui/v32.4.2/agent/index.json"
+    "pinned": "https://raw.githubusercontent.com/godx-jp/godxjp-ui/v32.5.0/agent/index.json"
   },
   "source": "mcp/src/data — the same data @godxjp/ui-mcp serves — plus the foundation and semantic token tiers, read from src/tokens/*.css",
   "start": "https://raw.githubusercontent.com/godx-jp/godxjp-ui/main/agent/START-HERE.md",
@@ -62,5 +62,5 @@
     "foundation": "the seeds a consumer is invited to set — --primary, --background, --radius",
     "semantic": "named roles that follow the seeds — --ring, --text-link, --overlay-background"
   },
-  "version": "32.4.2"
+  "version": "32.5.0"
 }

package/agent/llms.txt

@@ -1,10 +1,10 @@
 # @godxjp/ui
 
 > A Japanese-enterprise React design system: 161 components, 2195 design tokens,
-> 43 cardinal rules. This file is the entry point for AI agents. Catalog version 32.4.2.
+> 43 cardinal rules. This file is the entry point for AI agents. Catalog version 32.5.0.
 
 If your client can run a process, do not read these files — run the MCP server instead
-(`npx @godxjp/ui-mcp@32.4.2`). It is searchable and version-locked. These files exist for agents
+(`npx @godxjp/ui-mcp@32.5.0`). It is searchable and version-locked. These files exist for agents
 that can only fetch URLs.
 
 ## Start
@@ -26,7 +26,7 @@ that can only fetch URLs.
 ## Pinning
 
 Every URL above tracks `main`. To pin to the release a project actually installed, swap `main` for
-the tag: `.../godx-jp/godxjp-ui/v32.4.2/agent/...`. A catalog that does not match the installed
+the tag: `.../godx-jp/godxjp-ui/v32.5.0/agent/...`. A catalog that does not match the installed
 package describes props that are absent, or hides props that are present, and says nothing either way.
 
 Pinned catalogs only exist for releases whose tag actually contains `agent/`. If `…/v<version>/agent/index.json` returns 404, that release predates this catalog: read `…/main/…` instead and compare `index.json` → `version` against the package you have, so you at least know which way it drifted.

package/dist/components/data-entry/search-input.js

@@ -102,7 +102,9 @@ function SearchInput({
           className: cn(
             "ui-search-input-control !ps-[var(--search-input-start-padding)]",
             // gh#1250 — while the key cap shows, the text stops before it instead of running under it.
-            showShortcut ? "!pe-[calc(var(--search-input-edge-inset)+6ch)]" : "!pe-[var(--search-input-end-padding)]",
+            // gh#1285 — written FROM --search-input-end-padding (minus the ✕ slot it reserves, plus the
+            // cap), so a theme that retunes that token still moves this padding (default: edge inset + 6ch, as before).
+            showShortcut ? "!pe-[calc(var(--search-input-end-padding)_-_var(--control-icon-size)_-_var(--control-gap)_+_6ch)]" : "!pe-[var(--search-input-end-padding)]",
             inputClassName
           ),
           disabled

package/dist/components/general/typography.js

@@ -416,9 +416,9 @@ const TextBase = React.forwardRef((props, ref) => {
         'Text: `truncate` and `whitespace="pre-wrap"` are mutually exclusive \u2014 `truncate` takes precedence (one line, one ellipsis); use `clamp` to keep preserved line breaks and still bound the height.'
       );
     }
-    if (breakMode === "anywhere" && truncating) {
+    if ((breakMode === "anywhere" || breakMode === "phrase") && truncating) {
       console.warn(
-        'Text: `truncate` and `break="anywhere"` are mutually exclusive \u2014 `truncate` takes precedence (one line has nowhere to break to); use `clamp` to split the token and still bound the height.'
+        `Text: \`truncate\` and \`break="${breakMode}"\` are mutually exclusive \u2014 \`truncate\` takes precedence (one line has nowhere to break to); use \`clamp\` to keep the breaking and still bound the height.`
       );
     }
     if (ellipsis && (truncate || clamp !== void 0)) {
@@ -450,8 +450,8 @@ const TextBase = React.forwardRef((props, ref) => {
     // Text inherits, e.g. from a table cell. `truncate` still wins by emitting nothing.
     "data-whitespace": whitespace && !truncating ? whitespace : void 0,
     // Same shape and the same precedence as `whitespace`: `normal` is inert, and one line has
-    // nowhere to break to, so `truncate` simply does not emit it (gh#927).
-    "data-break": breakMode === "anywhere" && !truncating ? "anywhere" : void 0,
+    // nowhere to break to, so `truncate` simply does not emit it (gh#927, and `phrase` likewise).
+    "data-break": breakMode && breakMode !== "normal" && !truncating ? breakMode : void 0,
     style: clampStyle,
     "data-tabular": tabular ? "" : void 0,
     "data-decoration": decoration,

package/dist/contracts/measurement.json

@@ -1,6 +1,6 @@
 {
   "$comment": "AUTO-GENERATED by scripts/gen-measurement-contract.mjs — do not edit. Read this instead of guessing: docs/MEASUREMENT-CONTRACT.md.",
-  "version": "32.4.2",
+  "version": "32.5.0",
   "targetSize": {
     "standard": "WCAG 2.2 SC 2.5.8 Target Size (Minimum), level AA — 24×24 CSS px",
     "min": 24,

package/dist/props/components/general.prop.d.ts

@@ -75,6 +75,12 @@ export type TextProp = Omit<React.HTMLAttributes<HTMLElement>, "color"> & Omit<T
      * also releases a `nowrap` it inherited (a table cell's default) back to `normal` — newlines
      * still collapse, because an identifier's whitespace is not content.
      *
+     * `phrase` is for Japanese PROSE in a narrow measure — a name, an organisation, a status in a
+     * table cell: it breaks between phrases (`word-break: auto-phrase`) instead of mid-word, never
+     * starts a line with a small kana or 「ー」 (`line-break: strict`), adds `text-wrap: pretty`, and
+     * releases an inherited `nowrap` too. Each phrase stays whole, so a column is as wide as its
+     * longest phrase. Browsers without `auto-phrase` keep their own CJK breaking.
+     *
      * Independent of `whitespace`: `whitespace="pre-wrap" break="anywhere"` keeps typed line breaks
      * AND splits a long token. `truncate` WINS (one line has nowhere to break to), so the attribute
      * is not emitted under it; `clamp` composes.

package/dist/props/registry.d.ts

@@ -477,7 +477,7 @@ export declare const VOCABULARY_REGISTRY: {
     readonly TextBreakProp: {
         readonly file: "vocabulary/interaction.prop.ts";
         readonly category: "interaction";
-        readonly description: "Where an over-long unbroken token may break normal | anywhere: `anywhere` (overflow-wrap: anywhere) lets a machine identifier (email, code, login id) split to fit a narrow table cell; independent of TextWhitespaceProp, and `truncate` outranks it";
+        readonly description: "Where an over-long unbroken token may break normal | anywhere | phrase: `anywhere` (overflow-wrap: anywhere) lets a machine identifier (email, code, login id) split to fit a narrow table cell; `phrase` (word-break: auto-phrase + line-break: strict + text-wrap: pretty) breaks Japanese prose between phrases, never mid-katakana, keeping each phrase whole (the column is as wide as its longest phrase); independent of TextWhitespaceProp, and `truncate` outranks both";
     };
     readonly TypographyTypeProp: {
         readonly file: "vocabulary/interaction.prop.ts";

package/dist/props/registry.js

@@ -476,7 +476,7 @@ const VOCABULARY_REGISTRY = {
   TextBreakProp: {
     file: "vocabulary/interaction.prop.ts",
     category: "interaction",
-    description: "Where an over-long unbroken token may break normal | anywhere: `anywhere` (overflow-wrap: anywhere) lets a machine identifier (email, code, login id) split to fit a narrow table cell; independent of TextWhitespaceProp, and `truncate` outranks it"
+    description: "Where an over-long unbroken token may break normal | anywhere | phrase: `anywhere` (overflow-wrap: anywhere) lets a machine identifier (email, code, login id) split to fit a narrow table cell; `phrase` (word-break: auto-phrase + line-break: strict + text-wrap: pretty) breaks Japanese prose between phrases, never mid-katakana, keeping each phrase whole (the column is as wide as its longest phrase); independent of TextWhitespaceProp, and `truncate` outranks both"
   },
   TypographyTypeProp: {
     file: "vocabulary/interaction.prop.ts",

package/dist/props/vocabulary/interaction.prop.d.ts

@@ -108,8 +108,16 @@ export type TextWhitespaceProp = "normal" | "pre-wrap";
  * typed are content, `break` says whether a machine identifier may be split to fit. Only `anywhere`
  * — not `break-word` — lowers the box's min-content size (CSS Text 3 §5.5), which is what a table
  * cell sizes its column from; `break-word` breaks in a block and still holds a `<td>` open.
+ *
+ * `phrase` (gh#1292) is for PROSE in a narrow measure — a Japanese name, an organisation, a status label in a
+ * table cell: lines break between phrases instead of between any two
+ * characters (`word-break: auto-phrase`), small kana and the long-vowel mark never start a line
+ * (`line-break: strict`), and a paragraph does not end on one short word (`text-wrap: pretty`).
+ * Like `anywhere` it releases an inherited `nowrap`. Unlike `anywhere` it does NOT lower min-content:
+ * a column is as wide as its longest phrase, so a very long compound scrolls the table instead of
+ * being cut. Browsers without `auto-phrase` keep their own CJK breaking.
  */
-export type TextBreakProp = "normal" | "anywhere";
+export type TextBreakProp = "normal" | "anywhere" | "phrase";
 /**
  * antd `Typography` `type` — its four-value emphasis axis (`BaseType`).
  *

package/dist/styles/layers.json

@@ -1,5 +1,5 @@
 {
-  "version": "32.4.2",
+  "version": "32.5.0",
   "generatedBy": "scripts/gen-style-layers.mjs — do not edit by hand; run `pnpm gen:style-layers`",
   "base": "base.css",
   "fonts": "fonts.css",

package/dist/styles/text-layout.css

@@ -169,10 +169,23 @@
     overflow-wrap: anywhere;
     min-inline-size: 0;
   }
-  [data-slot="text"][data-break="anywhere"]:not([data-whitespace]) {
+  [data-slot="text"][data-break="anywhere"]:not([data-whitespace]),
+  [data-slot="text"][data-break="phrase"]:not([data-whitespace]) {
     white-space: normal;
   }
 
+  [data-slot="text"][data-break="phrase"] {
+    word-break: auto-phrase;
+    line-break: strict;
+    text-wrap: pretty;
+    min-inline-size: 0;
+  }
+  @supports not (word-break: auto-phrase) {
+    [data-slot="text"][data-break="phrase"] {
+      overflow-wrap: anywhere;
+    }
+  }
+
   [data-slot="text"][data-clamp] {
     display: -webkit-box;
     -webkit-box-orient: vertical;