Added
scripts/benchmark.sh: an automated, transparent CPU/RAM/responsiveness benchmark that drives the real release binary headlessly via tmux across seven scenarios, from an empty notebook up to 100,000 notes in one folder, 200 levels of nested folders, and a single 300,000-line note — samples real /proc/<pid>/stat/status numbers (CPU ticks, VmRSS, wall-clock time to first rendered frame) rather than estimating anything, so it doubles as a freeze/hang check.
Changed
- PREVIEW's note view and folder-peek now cache their formatted output (
App::note_preview_cache/ folder_preview_cache) instead of re-formatting on every ~100ms draw tick, and borrow rather than clone that cached text into the Paragraph (render::borrow_lines). Previously, a selected note's entire body was reformatted via markdown_to_lines on every redraw regardless of size, and a selected-but-not-entered folder re-listed the directory and re-parsed every note's frontmatter on every redraw regardless of folder size — both scaled with content size at ~10Hz whether or not anything had actually changed. Verified via scripts/benchmark.sh's aggressive scenarios: a 100,000-note folder now costs ~4.7% idle CPU (found via the same benchmark to be in the double digits before formatted-output caching was added) and a 300,000-line note ~9.9%, both with a sub-second first frame and zero measured RSS drift.
