maxLengthSingleLineRendering truncation can split grapheme clusters (emoji/ZWJ)

Start in lib/src/_code_paragraph.dart at the maxLengthSingleLineRendering path and its trucate(TextSpan, int) helper. Reproduce with wordWrap enabled, a small limit, and the listed emoji sequences, then inspect how both TextSpan.text and plainText are truncated. Done means truncation stops at grapheme-cluster boundaries consistently in both paths without broken rendering.

Problem

When using CodeEditor.maxLengthSingleLineRendering, Re-Editor truncates a long single line by doing substring(0, N) on TextSpan.text / plainText. This counts UTF-16 code units and may cut in the middle of a grapheme cluster (surrogate pairs, ZWJ sequences, variation selectors).

This can lead to broken rendering (e.g. replacement character "�"), and potentially inconsistent caret/selection hit-testing near the truncation boundary.

Where

lib/src/_code_paragraph.dart:

// current implementation
if (renderingLength != null && plainText.length > renderingLength) {
  impl = _build(trucate(span, renderingLength), plainText.substring(0, renderingLength), true);
}

TextSpan trucate(TextSpan span, int maxLength) {
  // ...
  if (text.length > remainingLength) {
    text = text.substring(0, remainingLength);
  }
  // ...
}

Repro

1. Enable wordWrap: true
2. Set maxLengthSingleLineRendering to a small number
3. Type/paste an emoji sequence around the boundary, e.g.:
   • Surrogate pair: 😀
   • ZWJ sequence: 👨‍👩‍👧‍👦 or 👩‍❤️‍💋‍👩
4. If truncation cuts inside the sequence, rendering becomes broken.

Suggestion

Make truncation grapheme-safe:

• Use package:characters (grapheme cluster iterator) to compute a safe prefix length <= maxLength (code units) and truncate at cluster boundaries.
• Apply the same safe length to both the span truncation and the plainText substring.

This keeps the performance guard while avoiding invalid Unicode sequences.

Thanks!