fix: guard TIOCGWINSZ before winpty_set_size to avoid assert on non-tty stdin

[davidcoleman007]

Summary

When the unix-adapter is invoked with stdin that is not a tty (e.g. piped from a parent process such as an editor's terminal harness, a CI runner, or any non-interactive tool that wraps bash), ioctl(STDIN_FILENO, TIOCGWINSZ, ...) fails with ENOTTY. Both call sites in src/unix-adapter/main.cc ignored the return value, leaving the local winsize variable uninitialised. In the resize loop, the subsequent memcmp against the prior size fires spuriously and winpty_set_size(wp, sz.ws_col, sz.ws_row, NULL) ends up being called with cols == 0 && rows == 0, which trips the assertion in libwinpty:

Assertion failed: ASSERT_CONDITION("wp != nullptr && cols > 0 && rows > 0"),
file src/libwinpty/winpty.cc, line 924

This causes a stream of assertion banners on stderr from any wrapper that pipes bash through winpty without a real console.

Fix

Guard both ioctl(TIOCGWINSZ) call sites: only act on the result when the call succeeds and ws_col/ws_row are positive. The libwinpty invariant in winpty_set_size is correct as-is — the bug was in the caller passing zero dimensions.

• Hunk A (initial-size, line 595) — defensive: re-fall back to the 80x25 default if the ioctl returned an error or zero-sized struct. The pre-init to 80x25 already covered most cases, so this is mostly symmetry with Hunk B.
• Hunk B (resize loop, lines 691-694) — the actual crash path. Skip the winpty_set_size call entirely if the ioctl failed or reported a degenerate size.

Test plan

• Built locally on a Git for Windows SDK / MSYS2 MinGW64 toolchain (gcc 15.2.0).
• Replaced the system winpty binaries; reproduced previous behaviour with the unpatched version (assert spam) and confirmed it is silent with the patched version.
• Verified normal interactive use (winpty cmd.exe /c echo hello, winpty --version) still works in mintty.

[davidcoleman007]

@rprichard thanks for the work on this package!

I don't know if this is actively developed these days, but i have been having the above issue consistently when using claude in git bash and windows git sdk. I patched it, tested with both sdk and git bash.

Please let me know if there is some additional process to certify this fix?

[rprichard]

Sorry, I haven't been maintaining this project since 2017 or so. Software that uses winpty ought to switch to using the official Windows ConPTY API instead, but ConPTY wasn't added until (some version of) Windows 10.

…

On Thu, May 21, 2026 at 1:53 PM Dave Coleman ***@***.***> wrote: *davidcoleman007* left a comment (rprichard/winpty#212) <#212 (comment)> @rprichard <https://github.com/rprichard> thanks for the work on this package! I don't know if this is actively developed these days, but i have been having the above issue consistently when using claude in git bash and windows git sdk. I patched it, tested with both sdk and git bash. Please let me know if there is some additional process to certify this fix? — Reply to this email directly, view it on GitHub <#212 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAL7753TLDCHJCBF7J5AJ7D435GCZAVCNFSM6AAAAACZHALPMCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHM2DKMJRG44TIMBRG4> . Triage notifications on the go with GitHub Mobile for iOS <https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675> or Android <https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[davidcoleman007]

Sorry, I haven't been maintaining this project since 2017 or so. Software that uses winpty ought to switch to using the official Windows ConPTY API instead, but ConPTY wasn't added until (some version of) Windows 10.

That's a real shame. Interestingly the current git bash distribution shipped by git/Microsoft today is actively including this package, and NOT ConPTY.