Files
orca/src/main
ecfc547ece fix(windows): reach the pty job on the teardown path that actually runs (#15886)
W2 gave every ConPTY pane a job object and wired `terminateOwnedTree` into
`local-pty-provider.ts`. Measured in #11047: worktree delete does not execute
there. It runs in the terminal daemon, so on the path that matters the sweep
still fell back to a parent-pid walk -- which a detached, reparented grandchild
is not in. That process is the one that holds the worktree cwd open, so the
delete this was meant to fix could still fail.

Expose the job on `SubprocessHandle` and use it at the three daemon call sites
(`terminal-session-teardown.ts` x2, `session-termination-controller.ts`).

Also: on Windows `forceKill()` returned early after any `kill()`, to avoid
double-closing the ConPTY handle node-pty owns. Correct, but it left force-kill
a permanent no-op -- and a wedged ConPTY never fires `onExit`, so such a session
had no escalation at all (#9854). Terminating the job is that escalation, and it
does not touch node-pty's handle.

A tree-walking guard now fails if any production `killWithDescendantSweep` call
omits `terminateOwnedTree`, since the option is optional by design and its
absence is invisible in review -- exactly how this gap survived W2.

Co-authored-by: OrcaWin <orcawin@users.noreply.github.com>
Co-authored-by: hanbong5938 <hanbong5938@users.noreply.github.com>
2026-08-21 23:58:48 -07:00
..
…