fix: count positions as utf-16 code units

resolves #22

crates/typos-lsp/Cargo.toml

@@ -19,7 +19,6 @@ serde = { version = "1.0", features = ["derive"] }
 ignore = "0.4.20"
 matchit = "0.7.1"
 shellexpand = "3.1.0"
-unicode-segmentation = "1.10.1"
 regex = "1.10.2"
 once_cell = "1.19.0"
 

crates/typos-lsp/src/lsp.rs

@@ -192,6 +192,8 @@ impl LanguageServer for Backend<'static, 'static> {
 
         Ok(InitializeResult {
             capabilities: ServerCapabilities {
+                // only support UTF-16 positions for now, which is the default when unspecified
+                position_encoding: Some(PositionEncodingKind::UTF16),
                 text_document_sync: Some(TextDocumentSyncCapability::Kind(
                     // TODO: should we support incremental?
                     TextDocumentSyncKind::FULL,
@@ -459,9 +461,11 @@ impl AccumulatePosition {
             .unwrap_or(0);
 
         let before_typo = String::from_utf8_lossy(&buffer[line_start..byte_offset]);
-        let line_pos =
-            unicode_segmentation::UnicodeSegmentation::graphemes(before_typo.as_ref(), true)
-                .count();
+
+        // count UTF-16 code units as per
+        // https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#textDocuments
+        // UTF-16 is the only position encoding we support for now
+        let line_pos = before_typo.chars().map(char::len_utf16).sum();
 
         self.line_num = line_num;
         self.line_pos = line_pos;

crates/typos-lsp/tests/integration_test.rs

@@ -20,6 +20,7 @@ async fn test_initialize_e2e() {
                   "codeActionKinds": ["quickfix"],
                   "workDoneProgress": false
                 },
+                "positionEncoding": "utf-16",
                 "textDocumentSync": 1,
                 "workspace": {
                   "workspaceFolders": { "changeNotifications": true, "supported": true }
@@ -218,17 +219,23 @@ async fn test_custom_config_file() {
 }
 
 #[test_log::test(tokio::test)]
-async fn test_unicode_diagnostics() {
-    let did_open = &did_open("¿Qué hace él?");
-
+async fn test_position_with_unicode_text() {
     let mut server = TestServer::new();
     let _ = server.request(&initialize()).await;
 
-    // start position should count graphemes with multiple code points as one visible character
+    // ¿ and é are two-byte code points in utf-8
+    let unicode_text = &did_open("¿Qué hace él?");
     similar_asserts::assert_eq!(
-        server.request(&did_open).await,
+        server.request(&unicode_text).await,
         publish_diagnostics(&[diag("`hace` should be `have`", 0, 5, 9)])
     );
+
+    // ẽ has two code points U+0065 U+0303 (latin small letter e, combining tilde)
+    let unicode_text = &did_open("ẽ hace");
+    similar_asserts::assert_eq!(
+        server.request(&unicode_text).await,
+        publish_diagnostics(&[diag("`hace` should be `have`", 0, 3, 7)])
+    );
 }
 
 fn initialize() -> String {