Optimize large segments (#80)

Co-Authored-By: Claude <[email protected]>

Author: cometkim

.changeset/many-days-melt.md

@@ -0,0 +1,8 @@
+---
+"unicode-segmenter": patch
+---
+
+Optimize grapheme segmenter.
+
+By eliminating unnecessary string concatenation, it significantly improved performance when creating large segments. (e.g. Demonic, Hindi, Flags, Skin tones)
+Also reduced the memory footprint by internal segment buffer.

README.md

@@ -256,7 +256,7 @@ Since [Hermes doesn't support the `Intl.Segmenter` API](https://github.com/faceb
 
 | Name                         | Unicode® | ESM? |   Size    | Size (min) | Size (min+gzip) | Size (min+br) |
 |------------------------------|----------|------|----------:|-----------:|----------------:|--------------:|
-| `unicode-segmenter/grapheme` |   16.0.0 |    ✔️ |    16,031 |     12,149 |           5,063 |         3,792 |
+| `unicode-segmenter/grapheme` |   16.0.0 |    ✔️ |    16,003 |     12,153 |           5,065 |         3,775 |
 | `graphemer`                  |   15.0.0 |    ✖️ ️|   410,435 |     95,104 |          15,752 |        10,660 |
 | `grapheme-splitter`          |   10.0.0 |    ✖️ |   122,252 |     23,680 |           7,852 |         4,841 |
 | `@formatjs/intl-segmenter`*  |   15.0.0 |    ✖️ |   603,285 |    369,560 |          72,218 |        49,416 |
@@ -272,7 +272,7 @@ Since [Hermes doesn't support the `Intl.Segmenter` API](https://github.com/faceb
 
 | Name                         | Bytecode size | Bytecode size (gzip)* |
 |------------------------------|--------------:|----------------------:|
-| `unicode-segmenter/grapheme` |        22,097 |                11,558 |
+| `unicode-segmenter/grapheme` |        22,087 |                11,548 |
 | `graphemer`                  |       133,978 |                31,713 |
 | `grapheme-splitter`          |        63,835 |                19,137 |
 

src/grapheme.js

@@ -88,13 +88,10 @@ export function* graphemeSegments(input) {
   let _hd = cp;
 
   let index = 0;
-  let segment = '';
 
   while (true) {
-    segment += input[cursor++];
-    if (!isBMP(cp)) {
-      segment += input[cursor++];
-    }
+    let chSize = isBMP(cp) ? 1 : 2;
+    cursor += chSize;
 
     // Note: Of course the nullish coalescing is useful here,
     // but avoid it for aggressive compatibility and perf claim
@@ -120,7 +117,7 @@ export function* graphemeSegments(input) {
       catAfter = cat(cp, cache);
     } else {
       yield {
-        segment,
+        segment: input.slice(index, cursor),
         index,
         input,
         _hd,
@@ -150,7 +147,7 @@ export function* graphemeSegments(input) {
 
     if (isBoundary(catBefore, catAfter, risCount, emoji, incb)) {
       yield {
-        segment,
+        segment: input.slice(index, cursor),
         index,
         input,
         _hd,
@@ -160,7 +157,6 @@ export function* graphemeSegments(input) {
 
       // flush
       index = cursor;
-      segment = '';
       emoji = false;
       incb = false;
       catBegin = catAfter;