[BUG] "gemni --resume" crushed

[mini2s]

What happened?

Running gemini --resume to restore a previous session crashes the CLI with the following critical error:

ERROR  ioctl(2) failed, EBADF

Full stack trace:

UnixTerminal.resize (/node_modules/@lydell/node-pty/unixTerminal.js:243:13)
_ShellExecutionService.resizePty (bundle/chunk-FLP3MI27.js:359821:30)
commitHookEffectListMount (bundle/chunk-DQXMK6DN.js:5266:30)
...

What did you expect to happen?

--resume should restore the session normally without crashing due to a PTY resize operation.

Root Cause Analysis

The error handler in ShellExecutionService.resizePty() currently only catches two recoverable error types:

1. ESRCH — the target process has already exited, causing ioctl to fail
2. Windows-specific error — "Cannot resize a pty that has already exited"

In the resume scenario, however, the PTY file descriptor from the previous session has already been closed when the original process exited. On resume, ExecutionLifecycleService.attachExecution re-attaches the execution, causing isPtyActive() to return true, but the underlying PTY fd is stale. The ptyProcess.resize() call then invokes ioctl(fd, TIOCSWINSZ, ...) on the invalid fd, which returns EBADF (bad file descriptor) rather than ESRCH.

Since EBADF is not handled by the existing catch block, the exception propagates upward and crashes the process.

Proposed Fix

Add EBADF to the list of safely ignorable error codes in the resizePty() catch block, treating it the same as ESRCH.

Location: packages/core/src/services/shellExecutionService.ts around line 1554

         const isEsrch = err.code === 'ESRCH';
+        const isEbadf = err.code === 'EBADF';
         const isWindowsPtyError = err.message?.includes(

-        if (isEsrch || isWindowsPtyError) {
+        if (isEsrch || isEbadf || isWindowsPtyError) {

Client information

• Platform: Linux
• Version: Gemini CLI (development branch)
• Repro command: gemini --resume

Login information

No response

Anything else we need to know?

No response

[mini2s]

Related PR:
#27371

[gemini-cli]

Found possible duplicate issues:

• Crash: interactive PTY resize race throws ioctl EBADF and exits #20792
• Critical Crash: ioctl(2) failed, EBADF during terminal resize or idle state #10523
• ioctl error while running gemeni cli #21835
• from time to time gemini cli crashes #10617

If you believe this is not a duplicate, please remove the status/possible-duplicate label.

[mini2s]

/assign

[gemini-cli]

👋 @mini2s, thanks for your interest in this issue! We're reserving self-assignment for issues that have been marked with the help wanted label. Feel free to check out our list of issues that need attention.

[gemini-cli]

Effort Analysis:
The crash occurs in packages/core/src/services/shellExecutionService.ts within the resizePty method (L1538). During a session resume (gemini --resume), ExecutionLifecycleService.attachExecution re-registers the previous session's PID. If the PTY file descriptor is stale (closed), calling ptyProcess.resize(cols, rows) (L1546) triggers an ioctl(2) failure with error code EBADF. The current error handler only catches ESRCH and Windows-specific exit messages, allowing EBADF to propagate and crash the CLI. According to the guidelines, any issue involving node-pty and platform-specific complexities (like ioctl and EBADF) is categorized as effort/large due to the significant overhead of reproducing and verifying low-level terminal behavior across different OS environments.

[mini2s]

/assign