fix(desktop): run Windows terminals on the bundled ConPTY, which keeps output in order

The ConPTY built into Windows passes OSC sequences through ahead of text it
has not rendered yet, so the finish marker of a quick command often arrived
before its output and the agent got an empty result (a quarter to two
fifths of runs on Windows Server 2025). node-pty's bundled conpty.dll and
OpenConsole keep the stream in order. The Windows suite now runs a dozen
commands per shell so a lost output shows.

Author: waleedlatif1

apps/desktop/src/main/terminal/session.ts

@@ -422,6 +422,9 @@ export class TerminalSession {
       rows: options.rows,
       cwd: options.cwd,
       env: shellEnv,
+      // The ConPTY built into Windows can deliver an OSC marker ahead of the text written before
+      // it, so a command's output lands after its finish marker; the bundled build keeps order.
+      ...(platform === 'win32' ? { useConptyDll: true } : {}),
     })
 
     logger.info('Started terminal session', { shell: shellPath, instrumented: shell !== null })

apps/desktop/src/main/terminal/windows-shells.test.ts

@@ -29,7 +29,8 @@ const WINDOWS_POWERSHELL = join(
 interface ShellUnderTest {
   name: string
   path: string
-  succeed: string
+  /** Prints `sim-ok-<n>`. */
+  succeed(n: number): string
   fail: string
   /** A cmdlet that fails, which sets no exit code of its own. */
   failingCmdlet?: string
@@ -41,7 +42,7 @@ interface ShellUnderTest {
 }
 
 const POWERSHELL: Omit<ShellUnderTest, 'name' | 'path'> = {
-  succeed: 'Write-Output sim-ok',
+  succeed: (n) => `Write-Output sim-ok-${n}`,
   fail: 'cmd /c exit 3',
   failingCmdlet: 'Get-Item -LiteralPath C:\\sim-does-not-exist',
   quote: (path) => `'${path}'`,
@@ -55,7 +56,7 @@ const SHELLS: ShellUnderTest[] = [
   {
     name: 'Git Bash',
     path: 'C:\\Program Files\\Git\\bin\\bash.exe',
-    succeed: 'echo sim-ok',
+    succeed: (n) => `echo sim-ok-${n}`,
     fail: "sh -c 'exit 3'",
     quote: (path) => `'${path.replaceAll('\\', '/')}'`,
     cd: 'cd',
@@ -70,6 +71,9 @@ Set-Content -LiteralPath $PidFile -Value $PID
 Start-Sleep -Seconds 120
 `
 
+/** Enough runs that output lost to a reordering ConPTY (a quarter of runs or more) shows. */
+const OUTPUT_RUNS = 12
+
 /** A shell's first start on a cold CI runner, then the Stop escalation, take a while. */
 const SHELL_TEST_TIMEOUT_MS = 90_000
 
@@ -115,12 +119,16 @@ describe.runIf(process.platform === 'win32')('the agent terminal on Windows', ()
     const installed = existsSync(shell.path)
 
     it.runIf(installed)(
-      "reports a successful command's output and exit code",
+      "reports each successful command's output and exit code",
       async () => {
-        const result = await run(open(shell), 'call-ok', shell.succeed)
-
-        expect(result).toMatchObject({ status: 'completed', exitCode: 0 })
-        expect(result.output).toContain('sim-ok')
+        const terminal = open(shell)
+        // Repeated: a ConPTY that reorders output past the finish marker loses it only some runs.
+        for (let n = 0; n < OUTPUT_RUNS; n++) {
+          const result = await run(terminal, `call-ok-${n}`, shell.succeed(n))
+
+          expect(result).toMatchObject({ status: 'completed', exitCode: 0 })
+          expect(result.output).toContain(`sim-ok-${n}`)
+        }
       },
       SHELL_TEST_TIMEOUT_MS
     )
@@ -190,7 +198,7 @@ describe.runIf(process.platform === 'win32')('the agent terminal on Windows', ()
         await expect(terminal.cancelTool('call-stubborn')).resolves.toBe(true)
         expect(await running).toMatchObject({ status: 'completed' })
         expect(isAlive(pid)).toBe(false)
-        expect(await run(terminal, 'call-after', shell.succeed)).toMatchObject({ exitCode: 0 })
+        expect(await run(terminal, 'call-after', shell.succeed(0))).toMatchObject({ exitCode: 0 })
       },
       SHELL_TEST_TIMEOUT_MS
     )