Fixed error rendering Burmese text

Iterate over graphemes instead of words, looking for word boundaries to use as line breaking opportunities. This eliminates the possibility of word-wrapping in the middle of a grapheme cluster, which is valid in some writing systems such as Thai, but mitigates the risk of an invalid section index in Burmese, because the word segmenter considers some modifiers to be “words”.

src/symbol/shaping.ts

@@ -532,28 +532,26 @@ export function determineLineBreaks(
     const potentialLineBreaks = [];
     const targetWidth = determineAverageLineWidth(logicalInput, spacing, maxWidth, glyphMap, imagePositions, layoutTextSize);
 
+    const graphemes = splitByGraphemeCluster(logicalInput.text);
+    const words = wordSegmenter.segment(logicalInput.text);
     let currentX = 0;
-    let graphemeIndex = 0;
-    for (const {index: wordIndex, segment: word} of wordSegmenter.segment(logicalInput.text)) {
-        const graphemes = splitByGraphemeCluster(word);
-        for (const {segment: grapheme} of graphemes) {
-            const section = logicalInput.getSection(graphemeIndex);
-            if (grapheme.trim()) {
-                currentX += getGlyphAdvance(grapheme, section, glyphMap, imagePositions, spacing, layoutTextSize);
-            }
-            graphemeIndex++;
+    for (const [i, grapheme] of graphemes.entries()) {
+        // Check whether the grapheme cluster immediately follows a word boundary.
+        const prevWord = words.containing(grapheme.index - 1);
+        const word = words.containing(grapheme.index);
+        if (prevWord && prevWord.index !== word.index) {
+            // Score the line breaking opportunity based on the characters immediately before and after the word boundary.
+            const prevCodePoint = logicalInput.text.codePointAt(grapheme.index - 1);
+            const firstCodePoint = grapheme.segment.codePointAt(0);
+            const penalty = calculatePenalty(prevCodePoint, firstCodePoint);
+            const lineBreak = evaluateBreak(i, currentX, targetWidth, potentialLineBreaks, penalty, false);
+            potentialLineBreaks.push(lineBreak);
         }
 
-        const nextWordIndex = wordIndex + word.length;
-        const lastCodePoint = graphemes.at(-1).segment.codePointAt(0);
-        const nextWordCodePoint = logicalInput.text.codePointAt(nextWordIndex);
-        if (!nextWordCodePoint) {
-            continue;
+        const section = logicalInput.getSection(i);
+        if (grapheme.segment.trim()) {
+            currentX += getGlyphAdvance(grapheme.segment, section, glyphMap, imagePositions, spacing, layoutTextSize);
         }
-
-        const penalty = calculatePenalty(lastCodePoint, nextWordCodePoint);
-        const lineBreak = evaluateBreak(graphemeIndex, currentX, targetWidth, potentialLineBreaks, penalty, false);
-        potentialLineBreaks.push(lineBreak);
     }
 
     return leastBadBreaks(