diff options
| author | Berke Güzel <wenekar1@gmail.com> | 2026-10-15 12:38:32 +0300 |
|---|---|---|
| committer | Berke Güzel <wenekar1@gmail.com> | 2026-10-15 12:38:32 +0300 |
| commit | 4c72f7b02fbafd15d0cf843edece106714bcc5f7 (patch) | |
| tree | ad5bdc869fd00cdb05d62eda4404586403382ce8 /tests/extern/kitty/screen.py | |
| parent | 5f03033b0e211f9ed13e7a1351f9a8f9011ab665 (diff) | |
Keep the snapshot unchanged when an update fails
term_snapshot_update converted the viewport in place, so an allocation
that failed midway left the snapshot empty and a renderer painted a blank
frame. Split the update into two phases. Phase 1 makes every allocation
the update needs and counts the arena words the converted rows add, then
grows the arena once. Phase 2 writes the view, cells, keys, clusters, and
selection, and cannot fail. A failed update now keeps the previous
content, and a retry is an ordinary incremental update.
invalidate() is removed. The term_snapshot_update comment and the
correctness audit now say the snapshot keeps its previous content.
The failure test fails each allocation in turn from the same starting
state. After every failed call it compares the view, cells, row keys,
selection spans, and cluster code points with the state before the call.
A retry must succeed and match a fresh capture. The cases are a first
update, a 32 by 16 screen of ZWJ families and regional-indicator flags
(arena growth), a resize, a 300-row screen (scratch arrays), and a long
scroll run.
Co-Authored-By: Claude Haiku 5.5 <noreply@anthropic.com>
Diffstat (limited to 'tests/extern/kitty/screen.py')
0 files changed, 0 insertions, 0 deletions
