maybe...?

Author: cometkim

src/grapheme.js

@@ -49,20 +49,24 @@ export function* graphemeSegments(input) {
   }
 
   const len = input.length;
-  let cursor = 0;
   let index = 0;
+  let cp = /** @type {number} */ (input.codePointAt(0));
+  let cursor = cp <= 0xFFFF ? 1 : 2;
 
-  /** @type {GraphemeCategoryNum | null} Category of codepoint immediately preceding cursor, if known. */
-  let catBefore = null;
+  /** Memoize the beginnig code point a the segment. */
+  let _hd = cp;
+
+  /** @type {import('./_grapheme_data.js').GraphemeCategoryRange} */
+  const cache = [0, 0, 2 /* GC_Control */];
+
+  /** Category of codepoint immediately preceding cursor, if known. */
+  let catBefore = cat(cp, cache);
 
   /** @type {GraphemeCategoryNum | null} Category of codepoint immediately preceding cursor, if known. */
   let catAfter = null;
 
-  /** @type {GraphemeCategoryNum | null} Beginning category of a segment */
-  let catBegin = null;
-
-  /** @type {import('./_grapheme_data.js').GraphemeCategoryRange} */
-  const cache = [0, 0, 2 /* GC_Control */];
+  /** Beginning category of a segment */
+  let catBegin = catBefore;
 
   /** @type {number} The number of RIS codepoints preceding `cursor`. */
   let risCount = 0;
@@ -79,34 +83,7 @@ export function* graphemeSegments(input) {
   /** InCB=Consonant InCB=Linker x InCB=Consonant */
   let incb = false;
 
-  let cp = /** @type {number} */ (input.codePointAt(cursor));
-
-  /** Memoize the beginnig code point a the segment. */
-  let _hd = cp;
-
   while (cursor < len) {
-    cursor += cp <= 0xFFFF ? 1 : 2;
-
-    // Note: Of course the nullish coalescing is useful here,
-    // but avoid it for aggressive compatibility and perf claim
-    catBefore = catAfter;
-    if (catBefore === null) {
-      catBefore = cat(cp, cache);
-      catBegin = catBefore;
-    }
-
-    if (cursor >= len) {
-      yield {
-        segment: input.slice(index, cursor),
-        index,
-        input,
-        _hd,
-        _catBegin: /** @type {typeof catBefore} */ (catBegin),
-        _catEnd: catBefore,
-      };
-      return;
-    }
-
     // Note: Lazily update `consonant` and `linker` state
     // which is a extra overhead only for Hindi text.
     if (cp >= 2325) {
@@ -156,7 +133,23 @@ export function* graphemeSegments(input) {
       catBegin = catAfter;
       _hd = cp;
     }
+
+    catBefore = catAfter;
+    cursor += cp <= 0xFFFF ? 1 : 2;
   }
+
+  if (index < len) {
+    yield {
+      segment: input.slice(index),
+      index,
+      input,
+      _hd,
+      _catBegin: /** @type {typeof catBefore} */ (catBegin),
+      _catEnd: catBefore,
+    };
+  }
+
+
 }
 
 /**