Ctrl+H (delete previous character) is misinterpreted as Ctrl+Backspace (delete word) under WSL2 due to WT_SESSION leaking from Windows Terminal
Personne n'a encore pris cette issue.
- Langage dominant
- Shell
- Étoiles
- 11.2k
- Forks
- 1.9k
- Merge moyen
- 14 h 16 min
- PR mergées (30 j)
- 6
Description
Description
/help documents ctrl+h as "delete previous character", but under WSL2 it instead deletes the whole previous word (i.e. behaves like ctrl+w/Ctrl+Backspace).
Environment
- Copilot CLI version: 1.0.78-2
- Platform: WSL2 (
Linux ... 6.18.33.2-microsoft-standard-WSL2), Windows Terminal as host, plainxterm-256colorinside (issue reproduces both inside and outside tmux) WSLENV=WT_SESSION:WT_PROFILE_ID:(default WSL/Windows Terminal integration), soWT_SESSIONis present in the WSL shell environment even though the CLI is not running as a native Windows process
Root cause
In the bundled app.js, the raw-byte key decoder special-cases byte 0x08 (\b) — which is exactly what Ctrl+H sends — as Ctrl+Backspace when a ctrlHIsCtrlBackspace flag is set:
function ZDe(t, e=!1) {
return t.length!==1 ? null
: t==="\r" ? {code:"return"}
: ...
: t==="\b" && e ? {code:"backspace", ctrl:!0} // <-- Ctrl+H reinterpreted as Ctrl+Backspace
: t==="\b" || t==="\x7F" ? {code:"backspace"}
: ...
}
The flag is computed once at startup:
reader = new ope({ ctrlHIsCtrlBackspace: fSt() })
function Nk() { return !RA() && !process.env.TMUX && !process.env.STY }
function fSt() { return process.platform==="win32" && Nk() || !!process.env.WT_SESSION }
The intent is reasonable on native Windows (Windows Console/Windows Terminal apps can't distinguish a literal Ctrl+H keypress from Ctrl+Backspace, since both produce byte 0x08), so trading away the Ctrl+H binding there makes sense. However, the || !!process.env.WT_SESSION clause has no accompanying process.platform==="win32" check, so it also fires under WSL2, where WT_SESSION is forwarded into the Linux environment by WSLENV for terminal-integration purposes, but the CLI is actually running in a normal Linux pty where Ctrl+H and Ctrl+Backspace are distinguishable. This causes a false positive: the workaround for a Windows-only ambiguity ends up breaking a real, distinguishable keybinding on WSL.
This also affects users in the reverse scenario, if WT_SESSION is otherwise present in the environment (e.g. propagated through SSH or subshells) on non-Windows platforms.
Steps to reproduce
- Launch Copilot CLI inside WSL2 under Windows Terminal (default
WSLENVintegration). - Type some text, then press Ctrl+H.
- Expected: deletes one character before the cursor (per
/help). - Actual: deletes the entire previous word.
Suggested fix
Gate the WT_SESSION check on process.platform === "win32" as well, e.g.:
function fSt() {
return process.platform === "win32" && (Nk() || !!process.env.WT_SESSION);
}
or otherwise detect WSL (e.g. via /proc/version containing "microsoft" or process.platform !== "win32") and skip the ctrlHIsCtrlBackspace heuristic in that case.
Workaround
unset WT_SESSION before launching copilot restores the documented Ctrl+H behavior.
Guide de contribution
Ouvrir le guide de contribution
Par où commencer
- Lisez l'issue en entier, puis le guide de contribution du projet.
- Signalez en commentaire que vous la prenez — cela évite que deux personnes fassent le même travail.
- Forkez le dépôt et travaillez sur une branche.
- Ouvrez une pull request qui référence le numéro de l'issue.
Piste de recherche
Commencez dans l’app.js incluse, au niveau de fSt() et de son initialisation du lecteur, puis vérifiez comment WT_SESSION et process.platform affectent ctrlHIsCtrlBackspace avant de vérifier ZDe(). Reproduisez le problème sous WSL2 avec WT_SESSION défini et non défini ; c’est terminé lorsque Ctrl+H supprime un caractère tandis que le comportement natif de Windows reste inchangé.
Rédigé par le modèle d'indexation à partir du texte de l'issue.
Évaluation
- Stack technique
- javascript, linux, node.js
- Domaine
- cli, operating-systems
- Type d'issue
- Bug
- Difficulté
- 2/5
- Temps estimé
- 1-3 heures
- Activité
- Active
- Clarté
- Clairement spécifiée
- Accessibilité débutants
- 78/100