Proposal: deterministic encoding for child-process I/O

packages/core/package.json

@@ -54,7 +54,6 @@
     "@xterm/headless": "5.5.0",
     "ajv": "^8.17.1",
     "ajv-formats": "^3.0.0",
-    "chardet": "^2.1.0",
     "chokidar": "^5.0.0",
     "command-exists": "^1.2.9",
     "diff": "^8.0.3",

packages/core/src/index.ts

@@ -103,7 +103,6 @@ export {
 export * from './utils/tool-utils.js';
 export * from './utils/tool-visibility.js';
 export * from './utils/terminalSerializer.js';
-export * from './utils/systemEncoding.js';
 export * from './utils/textUtils.js';
 export * from './utils/formatters.js';
 export * from './utils/generateContentResponseUtilities.js';

packages/core/src/services/shellExecutionService.test.ts

@@ -120,10 +120,6 @@ vi.mock('../utils/terminalSerializer.js', () => ({
   convertColorToHex: () => '#000000',
   ColorMode: { DEFAULT: 0, PALETTE: 1, RGB: 2 },
 }));
-vi.mock('../utils/systemEncoding.js', () => ({
-  getCachedEncodingForBuffer: vi.fn().mockReturnValue('utf-8'),
-}));
-
 const mockProcessKill = vi
   .spyOn(process, 'kill')
   .mockImplementation(() => true);
@@ -1030,15 +1026,15 @@ describe('ShellExecutionService', () => {
   });
 
   describe('Platform-Specific Behavior', () => {
-    it('should use powershell.exe on Windows', async () => {
+    it('should use powershell.exe on Windows and prefix the command with chcp 65001 for the PTY session', async () => {
       mockPlatform.mockReturnValue('win32');
       await simulateExecution('dir "foo bar"', (pty) =>
         pty.onExit.mock.calls[0][0]({ exitCode: 0, signal: null }),
       );
 
       expect(mockPtySpawn).toHaveBeenCalledWith(
         'powershell.exe',
-        ['-NoProfile', '-Command', 'dir "foo bar"'],
+        ['-NoProfile', '-Command', 'chcp 65001 >$null;dir "foo bar"'],
         expect.any(Object),
       );
     });

packages/core/src/services/shellExecutionService.ts

@@ -13,7 +13,6 @@ import os from 'node:os';
 import fs, { mkdirSync } from 'node:fs';
 import path from 'node:path';
 import type { IPty } from '@lydell/node-pty';
-import { getCachedEncodingForBuffer } from '../utils/systemEncoding.js';
 import {
   getShellConfiguration,
   resolveExecutable,
@@ -81,6 +80,40 @@ function ensurePromptvarsDisabled(command: string, shell: ShellType): string {
   return `${BASH_SHOPT_GUARD} ${command}`;
 }
 
+// On Windows, a new ConPTY session inherits its codepage from the system
+// OEMCP (microsoft/terminal `src/host/settings.cpp:41` defaults
+// `_uCodePage` to `Globals.uiOEMCP`, set from `GetOEMCP()` in
+// `srvinit.cpp:44`). On locales without "Beta: Use Unicode UTF-8 for
+// worldwide language support" the OEMCP is a legacy codepage (e.g. 850,
+// 866, 936, 932), and conhost converts every byte from the child via
+// `MultiByteToWideChar(gci.OutputCP, ...)` in `_stream.cpp:341-343`,
+// turning UTF-8 output from child processes (perl, python, node, ...)
+// into mojibake.
+//
+// `CreatePseudoConsole` does not accept a codepage argument
+// (microsoft/terminal#9174 — open as a feature request). The only way
+// to set the ConPTY codepage is from inside the new session via
+// `SetConsoleOutputCP` (intercepted by conhost in `getset.cpp:1144`).
+// Prefix the command with `chcp 65001` so the first thing the new
+// session does is switch its codepage to UTF-8.
+function injectUtf8CodepageForPty(
+  command: string,
+  shell: ShellType,
+  isWindows: boolean,
+  usingPty: boolean,
+): string {
+  if (!isWindows || !usingPty) {
+    return command;
+  }
+  if (shell === 'powershell') {
+    return `chcp 65001 >$null;${command}`;
+  }
+  if (shell === 'cmd') {
+    return `chcp 65001>nul&${command}`;
+  }
+  return command;
+}
+
 /** A structured result from a shell command execution. */
 export type ShellExecutionResult = ExecutionResult;
 
@@ -389,6 +422,7 @@ export class ShellExecutionService {
     cwd: string,
     shellExecutionConfig: ShellExecutionConfig,
     isInteractive: boolean,
+    usingPty: boolean,
   ): Promise<{
     program: string;
     args: string[];
@@ -417,7 +451,13 @@ export class ShellExecutionService {
     const resolvedExecutable = resolveExecutable(executable) ?? executable;
 
     const guardedCommand = ensurePromptvarsDisabled(commandToExecute, shell);
-    const spawnArgs = [...argsPrefix, guardedCommand];
+    const finalCommand = injectUtf8CodepageForPty(
+      guardedCommand,
+      shell,
+      isWindows,
+      usingPty,
+    );
+    const spawnArgs = [...argsPrefix, finalCommand];
 
     // 2. Prepare Environment
     const gitConfigKeys: string[] = [];
@@ -520,6 +560,7 @@ export class ShellExecutionService {
         cwd,
         shellExecutionConfig,
         isInteractive,
+        false,
       );
       cmdCleanup = prepared.cleanup;
 
@@ -620,14 +661,8 @@ export class ShellExecutionService {
 
       const handleOutput = (data: Buffer, stream: 'stdout' | 'stderr') => {
         if (!stdoutDecoder || !stderrDecoder) {
-          const encoding = getCachedEncodingForBuffer(data);
-          try {
-            stdoutDecoder = new TextDecoder(encoding);
-            stderrDecoder = new TextDecoder(encoding);
-          } catch {
-            stdoutDecoder = new TextDecoder('utf-8');
-            stderrDecoder = new TextDecoder('utf-8');
-          }
+          stdoutDecoder = new TextDecoder('utf-8');
+          stderrDecoder = new TextDecoder('utf-8');
         }
 
         if (isStreamingRawContent && sniffedBytes < MAX_SNIFF_SIZE) {
@@ -900,6 +935,7 @@ export class ShellExecutionService {
         cwd,
         shellExecutionConfig,
         true,
+        true,
       );
       cmdCleanup = prepared.cleanup;
 
@@ -1115,12 +1151,7 @@ export class ShellExecutionService {
           () =>
             new Promise<void>((resolveChunk) => {
               if (!decoder) {
-                const encoding = getCachedEncodingForBuffer(data);
-                try {
-                  decoder = new TextDecoder(encoding);
-                } catch {
-                  decoder = new TextDecoder('utf-8');
-                }
+                decoder = new TextDecoder('utf-8');
               }
 
               if (isStreamingRawContent && sniffedBytes < MAX_SNIFF_SIZE) {