fix(desktop): derive avatar initials by character, not code unit

desktop/src/shared/lib/initials.test.mjs

@@ -20,3 +20,63 @@ describe("getInitials", () => {
     assert.equal(getInitials("()"), "");
   });
 });
+
+describe("getInitials beyond the BMP", () => {
+  it("keeps a whole astral letter instead of half a surrogate pair", () => {
+    // U+20000, CJK Extension B — an ordinary character in some names.
+    const initials = getInitials("\u{20000}明");
+    assert.equal(initials, "\u{20000}");
+    assert.equal([...initials].length, 1);
+  });
+
+  it("keeps both initials whole when both are astral", () => {
+    const initials = getInitials("\u{20000}\u{20001} \u{20002}\u{20003}");
+    assert.equal(initials, "\u{20000}\u{20002}");
+    assert.equal([...initials].length, 2);
+  });
+
+  it("mixes an astral first name with an ordinary surname", () => {
+    assert.equal(getInitials("\u{1D400}da Lovelace"), "\u{1D400}L");
+  });
+});
+
+describe("getInitials with combining marks", () => {
+  it("does not split a word at a vowel sign", () => {
+    // अनिल कुमार — the vowel sign in अनिल used to split the word, so the
+    // second initial came from the middle of the first name. कु is one
+    // cluster: the surname's vowel sign belongs to its consonant.
+    assert.equal(getInitials("अनिल कुमार"), "अकु");
+  });
+
+  it("gives a one-word name one initial", () => {
+    assert.equal(getInitials("नमस्ते"), "न");
+  });
+
+  it("handles a Burmese name the same way", () => {
+    assert.equal(getInitials("မောင်မောင်"), "မေ");
+  });
+
+  it("still strips punctuation that is not a mark", () => {
+    assert.equal(getInitials("B (relay)"), "BR");
+  });
+});
+
+describe("getInitials takes a grapheme cluster, not a code point", () => {
+  it("keeps a decomposed accent with its letter", () => {
+    // NFD: E + U+0301. A code point initial dropped the accent entirely.
+    // The result stays decomposed — the initial is the input's own cluster,
+    // not a renormalized one — so compare against the decomposed form.
+    assert.equal(getInitials("E\u0301lodie Durand"), "E\u0301D");
+    assert.equal(getInitials("E\u0301lodie Durand").normalize("NFC"), "ÉD");
+  });
+
+  it("does not split a cluster joined by a zero-width joiner", () => {
+    // क्‍ष is one cluster; ZWJ is neither a letter nor a mark, so it used to
+    // act as a word separator and produce two initials from one word.
+    assert.equal(getInitials("\u0915\u094D\u200D\u0937 Name"), "क्‍षN");
+  });
+
+  it("still returns nothing for a name with no letters", () => {
+    assert.equal(getInitials("()"), "");
+  });
+});

desktop/src/shared/lib/initials.ts

@@ -1,11 +1,42 @@
-/** Derive up to two uppercase initials from a display name. */
+const graphemeSegmenter =
+  typeof Intl.Segmenter === "function"
+    ? new Intl.Segmenter(undefined, { granularity: "grapheme" })
+    : null;
+
+/** The first user-perceived character of a word, or "" when it has none. */
+function firstGrapheme(word: string): string {
+  if (graphemeSegmenter) {
+    for (const { segment } of graphemeSegmenter.segment(word)) {
+      return segment;
+    }
+    return "";
+  }
+  // Older engines without Intl.Segmenter degrade to a code point, which is
+  // still whole — never half a surrogate pair. Mirrors `MessageLinkPill`.
+  return Array.from(word)[0] ?? "";
+}
+
+/**
+ * Derive up to two uppercase initials from a display name.
+ *
+ * An initial is a grapheme cluster, not a code unit and not a code point.
+ * Taking `word[0]` returned half a surrogate pair for a name outside the
+ * Basic Multilingual Plane; taking one code point returned a bare consonant
+ * for `कुमार` or `မောင်`, and dropped the accent from a decomposed `Élodie`.
+ * Only a cluster keeps a letter together with what belongs to it.
+ *
+ * Word separation keeps combining marks and join controls, which are neither
+ * `\p{L}` nor `\p{N}`: replacing them with a separator cut words apart from
+ * the inside, splitting `अनिल` at its vowel sign and the joined cluster
+ * `क्‍ष` at its ZWJ.
+ */
 export function getInitials(name: string): string {
   return name
-    .replace(/[^\p{L}\p{N}\s]/gu, " ")
+    .replace(/[^\p{L}\p{M}\p{N}\p{Join_Control}\s]/gu, " ")
     .trim()
     .split(/\s+/)
-    .map((part) => part[0] ?? "")
-    .join("")
+    .map(firstGrapheme)
     .slice(0, 2)
+    .join("")
     .toUpperCase();
 }