fix(cursor): keep the visual selection whole over characters outside the BMP (#2683)

* fix(cursor): keep the visual selection whole over characters outside the BMP

The inclusive Nvim end column was turned into an exclusive VSCode one by
adding a single UTF-16 unit. A character outside the BMP is two of them, so
the selection ended in the middle of a surrogate pair and drifted away from
where Nvim had the cursor: moving right over it took two presses and the
selected region no longer matched.

Step over the whole character instead, using its byte length.

Closes #2535

* test(cursor): add visual selection tests for non-BMP characters

* test(cursor): expect the astral pair to be whole in the mixed-character selection

---------

Co-authored-by: wongxy <xiyao.wong@foxmail.com>

src/test/integ/multi-byte-chars.test.ts

@@ -1,4 +1,5 @@
 import { NeovimClient } from "neovim";
+import vscode from "vscode";
 
 import {
     attachTestNvimClient,
@@ -286,4 +287,54 @@ describe("Multi-width characters", () => {
             client,
         );
     });
+
+    it("Works - visual mode over chars outside the BMP", async () => {
+        // 𝕜 is U+1D55C: one Nvim character, but two UTF-16 units for VSCode
+        await openTextDocument({ content: ["a𝕜b"].join("\n") });
+
+        await assertContent({ content: ["a𝕜b"], cursor: [0, 0] }, client);
+
+        await sendVSCodeKeys("v");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 1)] }, client);
+
+        await sendVSCodeKeys("l");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 3)] }, client);
+
+        await sendVSCodeKeys("l");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 4)] }, client);
+
+        await sendEscapeKey();
+    });
+
+    it("Works - visual mode over consecutive non-BMP chars", async () => {
+        await openTextDocument({ content: ["a𝕜𝕝b"].join("\n") });
+
+        await sendVSCodeKeys("v");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 1)] }, client);
+
+        await sendVSCodeKeys("ll");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 5)] }, client);
+
+        await sendVSCodeKeys("l");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 6)] }, client);
+
+        await sendEscapeKey();
+    });
+
+    it("Works - visual mode across mixed BMP and non-BMP chars", async () => {
+        await openTextDocument({ content: ["aᵩ𝕜b"].join("\n") });
+
+        await sendVSCodeKeys("v");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 1)] }, client);
+
+        // the cursor sits on the astral char, whose two UTF-16 units both belong to
+        // the selection: a, the modifier letter and the pair take offsets 0 to 4
+        await sendVSCodeKeys("ll");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 4)] }, client);
+
+        await sendVSCodeKeys("l");
+        await assertContent({ vsCodeSelections: [new vscode.Selection(0, 0, 0, 5)] }, client);
+
+        await sendEscapeKey();
+    });
 });