mirror of
https://github.com/stablyai/orca.git
synced 2026-10-20 00:03:45 +00:00
Both were added by this PR onto the gate whose false positives silently kill a pane's keyboard, so a wrong mental model here is expensive. `tab.ptyId` does not name the pane that bound last: `updateTabPtyId` keeps the first binder (`currentTabPtyId ?? nextPtyIds[0]`, "later split-pane spawns must not steal it") and only `clearTabPtyId` demotes it to the last survivor. And a keystroke on an unbound pane does not arm recovery against `tab.ptyId` — `requestTerminalPaneRecovery` targets `ptyIdsByTabId[tabId][0]`, which diverges from `tab.ptyId` in split tabs. The reasoning the fixes rest on is unchanged: `tab.ptyId` holds *some sibling's* id rather than this pane's, which is what makes it wrong to gate a bound pane on. Comments only — no behavior change.