diff options
| author | Gabriel Schneider <[email protected]> | 2026-11-05 11:02:29 -0300 |
|---|---|---|
| committer | Gabriel Schneider <[email protected]> | 2026-11-05 13:14:28 -0300 |
| commit | 8640107e29a940d50a5eef0dc7d12689a515d856 (patch) | |
| tree | 13eb927f31e34d8b50366b3cabc8491c04debdcb /src/Store.zig | |
| parent | 3f1533813b6692a736251e4d467a7f108229ddee (diff) | |
| download | pardes-8640107e29a940d50a5eef0dc7d12689a515d856.tar.gz pardes-8640107e29a940d50a5eef0dc7d12689a515d856.zip | |
text journal, the primitive: File.replace(off, len, bytes) puts an edit in the text's own buffer (its tail moved with @memmove), the buffer grown by an eighth when it must be (State.cap), and moves its line index in place from the edit on (shiftIndex, State.line_cap), so a key, a backspace and a 9P write cost their bytes and what follows them, never a copy of the text (on the 50k-line file, Debug, medians against the base run beside it: edit-char 1105 -> 420 us, edit-twin 930 -> 433, body-2m 8.6 -> 4.7 ms, body-appends 173 -> 79 ms); setContent is the same edit with the caller's whole new text taken as the buffer; a replica's core (its allocator an epoch's, rollback.isEpoch) and a text a join still sends (memory.isPinned) keep their bytes, the edit putting a new buffer in their place as every edit did, so the rollback copy, its pins and the join's blobs work as they were; a pane's event readers hear an edit's span found from the edit (events.editSpan), the one diffSpan found from two texts; tests: 'edited line index equals a rebuilt one' becomes 'a line index shifted in place equals a rebuilt one' (editedLineStarts goes), 'an edit's span is where the texts before and after it differ' is new, and the panes test of history storage counts one allocation less for a redo (the index moves in place)
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_016Y9MNAtZenFv4weoMD3cge
Diffstat (limited to 'src/Store.zig')
0 files changed, 0 insertions, 0 deletions
