fix(mentions): show the mention after a link that ends at CJK punctuation

src/features/messages/MessageMarkdown.test.tsx

@@ -721,6 +721,15 @@ describe("Markdown profile mentions", () => {
     },
   );
 
+  it("renders a mention after a bare link that ends at CJK punctuation", () => {
+    const html = render("看这个 https://example.test，@Mic");
+    expect([...html.matchAll(/href="([^"]*)"/g)].map((m) => m[1])).toEqual([
+      "https://example.test/",
+    ]);
+    expect(html).toContain('，<button type="button"');
+    expect(html).toContain('aria-label="View Mic profile"');
+  });
+
   it.each<[string, RenderOptions]>([
     ["no opener", { canOpenLink: undefined }],
     ["a refusing opener", { canOpenLink: () => false }],

src/features/messages/message-edit.test.ts

@@ -1,6 +1,7 @@
 import { assert, describe, expect, it } from "vitest";
 import { shareMessageRows } from "../relay/row-identity";
 import { foldMessages } from "../relay/fold";
+import { scanMarkdown } from "../relay/message-content";
 import { profileTarget } from "../profiles/target";
 import { messageEditText, validateMessageEdit } from "./message-edit";
 import type { EventData } from "../relay/events";
@@ -75,6 +76,27 @@ describe("message edit round trips", () => {
     const row = firstRow([{ ...original, tags: [["h", "room"]] }]);
     expect(messageEditText(row, profiles)).toBe(original.content);
   });
+  it("keeps a reference definition's URL when an edit changes only prose", () => {
+    const definition = "[ref]: https://example.test/a，@Sam";
+    const sent = {
+      ...original,
+      content: `https://example.test，@Sam look at [ref].\n\n> ${definition}\n\n${definition}\n\n${"> ".repeat(101)}${definition}`,
+      tags: [
+        ["h", "room"],
+        ["p", person],
+      ],
+    };
+    const draft = messageEditText(firstRow([sent]), profiles);
+    expect(draft).toBe(
+      sent.content.replace(
+        "@Sam look",
+        `[@Sam](${profileTarget(person)}) look`,
+      ),
+    );
+    expect(
+      scanMarkdown(draft.replace("look", "see")).definitions.get("ref"),
+    ).toBe("https://example.test/a，@Sam");
+  });
   it.each([
     "[ref]: <https://example.com/a @Sam>",
     '[ref]: https://example.com "Ask @Sam"',

src/features/messages/message-link-parts.ts

@@ -1,4 +1,5 @@
 import { parseBuzzLink } from "../navigation/buzz-links";
+import { BARE_URL_REST } from "../relay/message-content";
 
 type LinkPart = { text: string; url?: string; label?: string };
 
@@ -82,6 +83,12 @@ function closingParenthesis(content: string, start: number, escaped: boolean) {
   return undefined;
 }
 
+const linkPattern = new RegExp(
+  /(?<!\\)\[((?:\\[^\r\n]|[^[\]\\\r\n])+)\]\\?\(|<(?:https?|buzz):\/\/[^\s<>"`]+>|(?:https?|buzz):\/\//
+    .source + BARE_URL_REST,
+  "gi",
+);
+
 /** Display-only link syntax. Message storage and navigation keep the original destination. */
 export function messageLinkParts(
   content: string,
@@ -95,10 +102,7 @@ export function messageLinkParts(
 ): LinkPart[] {
   const parts: LinkPart[] = [];
   let offset = 0;
-  // CJK punctuation ends a bare URL, except 。．｡ before a host label (URLs read them as dots).
-  for (const match of content.matchAll(
-    /(?<!\\)\[((?:\\[^\r\n]|[^[\]\\\r\n])+)\]\\?\(|<(?:https?|buzz):\/\/[^\s<>"`]+>|(?:https?|buzz):\/\/(?:[^\s<>"`:/?#，。、．！？；：＂＇（）［］｛｝「」『』《》【】〈〉〔〕〖〗〝〞〟｡｢｣､“”‘’…～〜]|[。．｡](?=[a-z\d-]+(?![a-z\d-]|:\/\/)))*(?:[:/?#][^\s<>"`，。、．！？；：＂＇（）［］｛｝「」『』《》【】〈〉〔〕〖〗〝〞〟｡｢｣､“”‘’…～〜]*)?/gi,
-  )) {
+  for (const match of content.matchAll(linkPattern)) {
     if (match.index < offset) continue;
     if (match[1] !== undefined) {
       const start = match.index + match[0].length;

src/features/messages/profile-mentions.test.ts

@@ -47,9 +47,25 @@ it.each([
   "\\@Mic",
   "@Mic_foo",
   "@Micé",
+  "https://example.test/文档?to=@Mic",
+  "https://example.test，https://example.test/b?to=@Mic",
+  "<https://example.test/a，@Mic>",
+  "[[a]](https://example.test/a，@Mic)",
+  "[a](\nhttps://example.test/a，@Mic)",
+  String.raw`[a]\(https://example.test/a，@Mic\)`,
 ])("leaves code/link/non-token context literal: %s", (content) => {
   expect(parts(content)).toEqual([{ text: content }]);
 });
+it("binds a mention after a bare link that ends at CJK punctuation", () => {
+  for (const content of [
+    "https://example.test，@Mic",
+    "看这个 https://example.test/a。@Mic",
+    "(https://example.test，@Mic)",
+  ])
+    expect(parts(content).filter((part) => part.target)).toEqual([
+      { text: "@Mic", target: profileTarget(mic) },
+    ]);
+});
 it("recognizes prose after a closed code span even when the span contained HTTPS", () => {
   expect(
     parts("`https://example.test @Mic` then @Mic").filter((p) => p.target),

src/features/messages/profile-mentions.ts

@@ -1,5 +1,6 @@
 import type { AgentLibrary } from "../agents/library";
 import type { ChannelMessage, Profile } from "../relay/contracts";
+import { BARE_URL_REST } from "../relay/message-content";
 import { profileTarget } from "../profiles/target";
 
 /** The names a person's signed prose mention binds by: their shown name and,
@@ -95,6 +96,15 @@ export function profileMentionParts(
   return parts.length ? parts : [{ text }];
 }
 
+const urlScheme = /^(?:[a-z][a-z\d+.-]*:\/\/|mailto:|nostr:)/.source;
+/** Bare URLs end where rendered ones do, and another can start right there. A
+ * URL after `<`, or after `](` or a pasted `]\(` and any whitespace, keeps
+ * CJK punctuation. */
+const urlChar = new RegExp(`^(?:${BARE_URL_REST})$`, "i");
+const wrapped = /(?<=<|\]\\?\(\s*)/y;
+const wrappedUrl = new RegExp(urlScheme + /[^\s<>`]+/.source, "i");
+const bareUrl = new RegExp(urlScheme + BARE_URL_REST, "i");
+
 // Evaluate exclusions on the full body BEFORE splitting link/emoji rendering.
 // Deliberately conservative for incomplete code/link syntax; not a Markdown parser.
 function protectedTextEnd(text: string, start: number): number {
@@ -120,7 +130,8 @@ function protectedTextEnd(text: string, start: number): number {
       return match ? match.index + match[0].length : text.length;
     }
   }
-  const boundary = start === 0 || /[\s<([]/.test(text[start - 1] ?? "");
+  const prev = text[start - 1] ?? "";
+  const boundary = start === 0 || /[\s<([]/.test(prev) || !urlChar.test(prev);
   if (char !== "`" && char !== "~" && char !== "[" && char !== "!" && !boundary)
     return start;
   const span = /^(?:`+|~{3,})/.exec(rest)?.[0];
@@ -132,8 +143,9 @@ function protectedTextEnd(text: string, start: number): number {
     const link = /^!?\[[^\]\n]*\](?:\([^\n]*?\)|\[[^\]\n]*\])/.exec(rest)?.[0];
     if (link) return start + link.length;
   }
-  const url = /^(?:[a-z][a-z\d+.-]*:\/\/|mailto:|nostr:)[^\s<>`]+/i.exec(
-    rest,
-  )?.[0];
+  const bare = bareUrl.exec(rest)?.[0];
+  // The lookbehind rescans whitespace, so test it only where a URL starts.
+  wrapped.lastIndex = start;
+  const url = bare && wrapped.test(text) ? wrappedUrl.exec(rest)?.[0] : bare;
   return url ? start + url.length : start;
 }

src/features/relay/message-content.ts

@@ -23,6 +23,12 @@ export function safeMessageUrl(value: string): string | undefined {
   return url?.startsWith("https:") ? url : undefined;
 }
 
+/** A bare URL after `scheme://`. CJK punctuation ends it, except 。．｡ before
+ * a host label (URLs read them as dots). */
+export const BARE_URL_REST =
+  /(?:[^\s<>"`:/?#，。、．！？；：＂＇（）［］｛｝「」『』《》【】〈〉〔〕〖〗〝〞〟｡｢｣､“”‘’…～〜]|[。．｡](?=[a-zA-Z\d-]+(?![a-zA-Z\d-]|:\/\/)))*(?:[:/?#][^\s<>"`，。、．！？；：＂＇（）［］｛｝「」『』《》【】〈〉〔〕〖〗〝〞〟｡｢｣､“”‘’…～〜]*)?/
+    .source;
+
 type MarkdownNode = {
   type: string;
   value?: string;