Rework TextEdit arrow navigation to handle Unicode graphemes

Cargo.lock

@@ -1263,6 +1263,7 @@ dependencies = [
  "profiling",
  "ron",
  "serde",
+ "unicode-segmentation",
 ]
 
 [[package]]

Cargo.toml

@@ -103,6 +103,7 @@ web-time = "1.1.0" # Timekeeping for native and web
 wgpu = { version = "24.0.0", default-features = false }
 windows-sys = "0.59"
 winit = { version = "0.30.7", default-features = false }
+unicode-segmentation = "1.12.0"
 
 [workspace.lints.rust]
 unsafe_code = "deny"

crates/egui/Cargo.toml

@@ -87,6 +87,7 @@ ahash.workspace = true
 bitflags.workspace = true
 nohash-hasher.workspace = true
 profiling.workspace = true
+unicode-segmentation.workspace = true
 
 #! ### Optional dependencies
 accesskit = { version = "0.17.0", optional = true }

crates/egui/src/text_selection/text_cursor_state.rs

@@ -4,8 +4,9 @@ use epaint::text::{
     cursor::{CCursor, Cursor},
     Galley,
 };
+use unicode_segmentation::UnicodeSegmentation;
 
-use crate::{epaint, NumExt, Rect, Response, Ui};
+use crate::{epaint, NumExt, Rect, Response, TextBuffer, Ui};
 
 use super::{CCursorRange, CursorRange};
 
@@ -224,7 +225,7 @@ fn select_line_at(text: &str, ccursor: CCursor) -> CCursorRange {
 
 pub fn ccursor_next_word(text: &str, ccursor: CCursor) -> CCursor {
     CCursor {
-        index: next_word_boundary_char_index(text.chars(), ccursor.index),
+        index: next_word_boundary_char_index(text, ccursor.index),
         prefer_next_row: false,
     }
 }
@@ -238,9 +239,10 @@ fn ccursor_next_line(text: &str, ccursor: CCursor) -> CCursor {
 
 pub fn ccursor_previous_word(text: &str, ccursor: CCursor) -> CCursor {
     let num_chars = text.chars().count();
+    let reversed: String = text.graphemes(true).rev().collect();
     CCursor {
         index: num_chars
-            - next_word_boundary_char_index(text.chars().rev(), num_chars - ccursor.index),
+            - next_word_boundary_char_index(&reversed, num_chars - ccursor.index).min(num_chars),
         prefer_next_row: true,
     }
 }
@@ -254,22 +256,28 @@ fn ccursor_previous_line(text: &str, ccursor: CCursor) -> CCursor {
     }
 }
 
-fn next_word_boundary_char_index(it: impl Iterator<Item = char>, mut index: usize) -> usize {
-    let mut it = it.skip(index);
-    if let Some(_first) = it.next() {
-        index += 1;
-
-        if let Some(second) = it.next() {
-            index += 1;
-            for next in it {
-                if is_word_char(next) != is_word_char(second) {
-                    break;
-                }
-                index += 1;
-            }
+fn next_word_boundary_char_index(text: &str, index: usize) -> usize {
+    let start_byte_index = text.byte_index_from_char_index(index);
+    let words = text.split_word_bound_indices().collect::<Vec<_>>();
+
+    for i in 0..words.len() {
+        let bi = words[i].0;
+        // Splitting considers contiguous whitespace as one word, such words must be skipped
+        // such that the cursor jumps right after the next word (this is consistent with text
+        // editors or browsers), the naive approach would jump after the whitespace following
+        // the next word.
+        if bi > start_byte_index && i > 0 && skip_word(words[i - 1].1) {
+            return char_index_from_byte_index(text, bi);
         }
     }
-    index
+
+    char_index_from_byte_index(text, text.len())
+}
+
+fn skip_word(text: &str) -> bool {
+    // skip words that contain anything other than alphanumeric characters and underscore
+    // (i.e. whitespace, dashes, etc.)
+    !text.chars().any(|c| !(c.is_alphanumeric() || c == '_'))
 }
 
 fn next_line_boundary_char_index(it: impl Iterator<Item = char>, mut index: usize) -> usize {
@@ -328,6 +336,19 @@ pub fn byte_index_from_char_index(s: &str, char_index: usize) -> usize {
     s.len()
 }
 
+pub fn char_index_from_byte_index(input: &str, byte_index: usize) -> usize {
+    for (ci, (bi, _)) in input.char_indices().enumerate() {
+        if bi == byte_index {
+            return ci;
+        }
+    }
+
+    input
+        .char_indices()
+        .last()
+        .map_or(0, |(i, c)| i + c.len_utf8())
+}
+
 pub fn slice_char_range(s: &str, char_range: std::ops::Range<usize>) -> &str {
     assert!(char_range.start <= char_range.end);
     let start_byte = byte_index_from_char_index(s, char_range.start);

crates/egui/src/widgets/text_edit/text_buffer.rs

@@ -10,8 +10,8 @@ use epaint::{
 
 use crate::text_selection::{
     text_cursor_state::{
-        byte_index_from_char_index, ccursor_next_word, ccursor_previous_word, find_line_start,
-        slice_char_range,
+        byte_index_from_char_index, ccursor_next_word, ccursor_previous_word,
+        char_index_from_byte_index, find_line_start, slice_char_range,
     },
     CursorRange,
 };
@@ -51,6 +51,10 @@ pub trait TextBuffer {
         byte_index_from_char_index(self.as_str(), char_index)
     }
 
+    fn char_index_from_byte_index(&self, char_index: usize) -> usize {
+        char_index_from_byte_index(self.as_str(), char_index)
+    }
+
     /// Clears all characters in this buffer
     fn clear(&mut self) {
         self.delete_char_range(0..self.as_str().len());