@aleph-alpha/chat-kit 6.1.1 → 6.2.0

package/dist/{useMessageHandler-Tellov3w.js → useMessageHandler-nnsort0-.js}

@@ -1,4 +1,4 @@
-import { g as generateId } from "./generateId-BHf0W4Rq.js";
+import { g as generateId } from "./markdownToPlainText-BOtlQ_BY.js";
 import { reactive, ref, computed, readonly } from "vue";
 function createEmptyTree() {
   return { nodes: {}, rootId: null };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@aleph-alpha/chat-kit",
-  "version": "6.1.1",
+  "version": "6.2.0",
   "description": "Chat kit package for building chat interfaces with Vue",
   "type": "module",
   "main": "./dist/index.js",
@@ -113,9 +113,9 @@
     "vue": "^3.5.27",
     "vue-tsc": "^3.2.4",
     "wait-on": "9.0.3",
+    "@aleph-alpha/prettier-config-frontend": "2.1.1",
     "@aleph-alpha/eslint-config-frontend": "0.7.1",
-    "@aleph-alpha/tsconfig-frontend": "0.7.3",
-    "@aleph-alpha/prettier-config-frontend": "2.1.1"
+    "@aleph-alpha/tsconfig-frontend": "0.7.3"
   },
   "scripts": {
     "build": "vite build && vue-tsc --project tsconfig.build.json --declaration --emitDeclarationOnly --outDir dist",

package/src/helpers/index.ts

@@ -1,5 +1,6 @@
 export * from './conversationTree';
 export * from './formatRelativeTime';
 export * from './generateId';
+export * from './markdownToPlainText';
 export * from './secondsToDate';
 export * from './toError';

package/src/helpers/markdownToPlainText.spec.ts

@@ -0,0 +1,107 @@
+import { describe, expect, it } from 'vitest';
+
+import { markdownToPlainText } from './markdownToPlainText';
+
+describe('markdownToPlainText', () => {
+  it('returns text without markdown unchanged', () => {
+    expect(markdownToPlainText('A plain excerpt from a page.')).toBe(
+      'A plain excerpt from a page.',
+    );
+  });
+
+  it('keeps a link label and drops its url', () => {
+    expect(
+      markdownToPlainText(
+        'See the [annual report](https://example.com/r.pdf).',
+      ),
+    ).toBe('See the annual report.');
+  });
+
+  it('strips emphasis markers', () => {
+    expect(markdownToPlainText('Revenue **rose** by _12%_ and ~~fell~~.')).toBe(
+      'Revenue rose by 12% and fell.',
+    );
+  });
+
+  it('separates a heading from the paragraph that follows it', () => {
+    expect(markdownToPlainText('## Cost\n\n12 EUR per seat.')).toBe(
+      'Cost 12 EUR per seat.',
+    );
+  });
+
+  it('flattens a GFM table without its pipes', () => {
+    const table = ['| Plan | Cost |', '| --- | --- |', '| Pro | 12 |'].join(
+      '\n',
+    );
+    expect(markdownToPlainText(table)).toBe('Plan Cost Pro 12');
+  });
+
+  it('separates list items', () => {
+    expect(markdownToPlainText('- first\n- second\n- third')).toBe(
+      'first second third',
+    );
+  });
+
+  it('keeps code content but not its fence', () => {
+    expect(markdownToPlainText('Run `pnpm build` first.')).toBe(
+      'Run pnpm build first.',
+    );
+  });
+
+  it('drops an image rather than reading its alt text as prose', () => {
+    expect(markdownToPlainText('![Revenue chart](c.png) Revenue rose.')).toBe(
+      'Revenue rose.',
+    );
+  });
+
+  it('removes inline html tags without joining the words around them', () => {
+    expect(markdownToPlainText('Revenue rose.<br>Costs fell.')).toBe(
+      'Revenue rose. Costs fell.',
+    );
+  });
+
+  it('keeps the prose inside a raw html block', () => {
+    expect(
+      markdownToPlainText('<div class="x">Revenue rose.</div>\nCosts fell.'),
+    ).toBe('Revenue rose. Costs fell.');
+  });
+
+  it('keeps a hard line break as a word separator', () => {
+    expect(markdownToPlainText('Revenue rose.  \nCosts fell.')).toBe(
+      'Revenue rose. Costs fell.',
+    );
+  });
+
+  it('leaves a literal comparison inside a raw html block alone', () => {
+    expect(markdownToPlainText('<div>2 < 3 and 4 > 1</div>')).toBe(
+      '2 < 3 and 4 > 1',
+    );
+  });
+
+  it('removes an html comment', () => {
+    expect(markdownToPlainText('<div>Hi</div>\n<!-- hidden -->\nText.')).toBe(
+      'Hi Text.',
+    );
+  });
+
+  it('collapses the newlines a multi-paragraph excerpt carries', () => {
+    expect(markdownToPlainText('First line.\n\n\nSecond line.')).toBe(
+      'First line. Second line.',
+    );
+  });
+
+  it('leaves an unbalanced emphasis marker as literal text', () => {
+    expect(markdownToPlainText('Revenue **rose sharply in')).toBe(
+      'Revenue **rose sharply in',
+    );
+  });
+
+  it('is idempotent', () => {
+    const once = markdownToPlainText('A [linked](https://example.com) claim.');
+    expect(markdownToPlainText(once)).toBe(once);
+  });
+
+  it('returns an empty string for empty input', () => {
+    expect(markdownToPlainText('')).toBe('');
+  });
+});

package/src/helpers/markdownToPlainText.ts

@@ -0,0 +1,69 @@
+import { unified } from 'unified';
+import remarkParse from 'remark-parse';
+import remarkGfm from 'remark-gfm';
+import type { Nodes } from 'mdast';
+
+const parser = unified().use(remarkParse).use(remarkGfm);
+
+const DROPPED_NODES = new Set(['image', 'imageReference']);
+
+// A hard line break is a childless, valueless leaf, so it has to yield its
+// separator explicitly or `first<br>second` flattens to `firstsecond`.
+const SEPARATOR_NODES = new Set(['break']);
+
+// An `html` node cannot be dropped whole: a line starting with a block tag
+// makes markdown's HTML-block rule run to the next blank line, so the node
+// holds the following prose too. Its tags are stripped instead, replaced by a
+// space so `rose.<br>Costs` doesn't come out as one word. A tag has to start
+// like one (`<p`, `</p`, `<!--`) so a literal comparison inside the block —
+// `2 < 3 and 4 > 1` — isn't mistaken for one and deleted. The repetition is
+// bounded because an unbounded `[^>]*` scans quadratically over a run of `<`
+// with no closing `>` (`sonarjs/slow-regex`).
+const HTML_TAG_RE = /<[!/]?[a-zA-Z-][^>]{0,512}>/g;
+
+// Children of these need a separator, or a heading runs straight into the
+// paragraph below it (`## Cost` + `12 EUR` → `Cost12 EUR`).
+const BLOCK_CONTAINERS = new Set([
+  'root',
+  'blockquote',
+  'list',
+  'listItem',
+  'table',
+  'tableRow',
+  'footnoteDefinition',
+]);
+
+function flatten(node: Nodes): string {
+  if (DROPPED_NODES.has(node.type)) return '';
+  if (SEPARATOR_NODES.has(node.type)) return ' ';
+  if ('value' in node) {
+    return node.type === 'html'
+      ? node.value.replace(HTML_TAG_RE, ' ')
+      : node.value;
+  }
+  if (!('children' in node)) return '';
+  return node.children
+    .map(flatten)
+    .join(BLOCK_CONTAINERS.has(node.type) ? ' ' : '');
+}
+
+/**
+ * Flatten markdown to the prose inside it: `[label](url)` becomes `label`,
+ * headings lose their `#`, emphasis markers and table pipes go.
+ *
+ * For excerpt text that is *displayed as plain text* — a citation snippet
+ * preview, say — where the markdown is an artefact of how the source was
+ * extracted rather than something the UI intends to render. Reads the parsed
+ * tree instead of pattern-matching the string, so it is structurally exact for
+ * well-formed markdown where a regex would guess. Idempotent on text carrying
+ * no markdown, so it is a no-op once a producer starts sending plain text.
+ *
+ * Two artefacts survive. A truncated excerpt can end mid-syntax (`**bold` with
+ * no closing pair), which the parser reads as literal text — the correct
+ * reading, at the cost of a visible `**`. And markdown inside a raw HTML block
+ * is never parsed as markdown, so it is flattened as the literal text it is.
+ */
+export function markdownToPlainText(markdown: string): string {
+  if (!markdown) return '';
+  return flatten(parser.parse(markdown)).replace(/\s+/g, ' ').trim();
+}

package/src/index.ts

@@ -148,7 +148,7 @@ export type {
 } from './components';
 
 // Helpers
-export { formatRelativeTime } from './helpers';
+export { formatRelativeTime, markdownToPlainText } from './helpers';
 
 // Shared Types
 export type {

package/dist/generateId-BHf0W4Rq.js

@@ -1,6 +0,0 @@
-function generateId() {
-  return typeof crypto !== "undefined" && "randomUUID" in crypto ? crypto.randomUUID() : String(Date.now());
-}
-export {
-  generateId as g
-};