Clear the dirty region in the CoreGraphics renderer (fixes restricted-region scroll artifacts)

[tavvet]

Problem

Scrolling or editing inside a restricted DECSTBM scroll region (a fixed header and/or footer, as in nano, vim, less, htop) leaves stale rows and a sub-cell "ghost" on the region's last row on the CoreGraphics renderer (Metal disabled). The Metal path masks it by clearing the whole target each frame.

Reproducible: open nano on a file with lines wider than the window and hold PageDown/PageUp; or do vim line edits in a fixed-footer layout.

Root cause — the renderer, not the scroll primitives

The scroll/edit primitives report a correct dirty range. The CG renderer:

1. never clears the dirty region's background before painting — drawTerminalContents fills background per-cell-run only, and only for cells that carry an explicit background; default-background cells rely on transparent backing-store pixels showing the layer's backgroundColor. AppKit clears the backing store only on a full-view redraw, so a partial repaint (a restricted region, IL/DL) keeps stale glyphs/backgrounds.
2. only invalidates the sub-cell bottom remainder when rowEnd == rows-1 — a restricted region ends at scrollBottom < rows-1, so the sliver below it is never repainted → the ghost.

(There is no pixel blit; unchanged pixels persist via AppKit's dirty-rect coalescing over the layer backing store.)

Fix — macOS CG path only; Metal unaffected

In AppleTerminalView:

1. context.clear(dirtyRect) before the per-row paint — clears the invalidated region to transparent (a clear, not a fill, so a translucent background is preserved), giving the per-cell paint a clean slate.
2. Extend a mid-screen region's invalidated region down by one cell, so the bottom remainder is cleared too (previously only rowEnd == rows-1 got this).

This fixes every row-mutating primitive (scroll, reverseIndex, cmdScrollUp/Down, cmdInsertLines/DeleteLines) at the source. With the renderer clearing any dirtied region, the primitives only need to flag the rows they changed — so refreshScrolledRegion flags just [scrollTop, scrollBottom] (keeping the full-screen + scrollback cheap path) instead of inflating to the whole viewport.

The Metal renderer is untouched: it already loadAction = .clears the whole target each frame, and the region extension only feeds the non-Metal setNeedsDisplay.

Testing

Verified in nano / vim / less / htop on the CG renderer — no stale rows or bottom ghost while paging or editing — with a translucent background intact, and normal full-screen output scrolling unaffected (cheap path preserved). The package builds cleanly.

---

This supersedes an earlier per-primitive workaround on this branch (which inflated each scroll op's dirty range to the whole viewport) with the renderer-level root fix.

[migueldeicaza]

Thank you, let me review this - it looks pretty clean.

[migueldeicaza]

This looks good!