What

The terminal PTY winsize could lag the rendered xterm grid after a resize, so tmux (and any multiplexer) drew pane content across the dividers. This commits any pending PTY resize on every renderer-pool teardown path so the winsize always converges to the grid.

Why

Closes #981.

The renderer pool refits the grid synchronously on resize but debounces the PTY resize by PTY_RESIZE_DEBOUNCE_MS (256ms). If a tmux leaf is hidden/evicted inside that window, detachSlotFromLeaf cleared the pending timer without flushing it, so the queued winsize update was dropped. releaseSlot then reported the uncommitted grid size as the session size, and a retained fast rebind suppressed the now-size-matched resizePty, so the PTY and grid stayed diverged until some unrelated later resize resynced them. tmux positions panes by absolute coordinates against the size it was told, so the stale winsize smeared content across the pane dividers.

How

• New pure helper pendingPtyResize(grid, committed) in rendererResize.ts: returns the size still owed to the PTY, or null when they already agree, and never emits a degenerate (<= 0) resize.
• rendererPool.ts: flushPty (the debounce timer) and detachSlotFromLeaf (every teardown path) now both route through a single commitPendingResize() that calls the helper and updates slot.lastCols/lastRows. Dropping the queued resize on teardown is no longer possible, so the winsize cannot lag the grid.

Single concern, 3 files, no src-tauri/ changes, no new dependencies.

Testing

• pnpm lint clean
• pnpm check-types clean
• pnpm test clean (264 tests; new rendererResize.test.ts covers match / grow / shrink, the degenerate-dimensions deny path, and the interrupted-debounce regression)
• pnpm build and pnpm size clean (size well within budget)

Verification level is unit test plus a static trace of the resize path (ResizeObserver -> debounced resizePty -> pty_resize -> master.resize), matching the triage note in #981. I have not captured a live before/after tmux recording: the trigger is timing-dependent (hide/show a leaf within the 256ms debounce) and does not reproduce deterministically in a headless run. Glad to add an integration-level test around detachSlotFromLeaf if you prefer that over the pure-helper unit test.

• pnpm lint clean
• pnpm check-types clean
• pnpm test clean
• Manual smoke-test of the affected feature (timing-dependent; see note above)
• (If you touched src-tauri/) cargo clippy - N/A, frontend only
• (If you touched src-tauri/) cargo nextest run - N/A, frontend only
• (If you changed a #[tauri::command] signature) - N/A, no command changes
• (If UI) tested in pnpm tauri dev - N/A, no UI/visual change
• Platforms tested: macOS (the logic is platform-agnostic; the desync is not platform-specific)
• Shells tested (if relevant): N/A, not shell-specific
• pnpm build + pnpm size clean (extra CI gates)

Screenshots / GIFs

N/A - no visual/UI change; this is terminal-to-PTY size plumbing.

Notes for reviewer

• Follows the existing functional-core pattern: the size decision is a pure, tested function and the pool wiring stays thin.
• This is the "minimal, single-concern fix plus a unit test" offered in Terminal: tmux content bleeds across panes after a resize (xterm grid vs PTY winsize desync) #981. detachSlotFromLeaf is the teardown path that dropped the resize; the debounce timer now shares the same commitPendingResize() so there is one source of truth for "what the PTY was last told".
• Flagging that I understand the terminal subsystem is load-bearing and you may prefer to discuss before merging; happy to adjust scope or test shape.

Summary by CodeRabbit

• Bug Fixes
  • Improved synchronization between the terminal’s rendered grid and its underlying PTY size, including during teardown to prevent content bleed.
• Tests
  • Added new test coverage for pending PTY resize behavior, including growth, shrink, interrupted resize debounce, and invalid grid dimensions.