Highlights
Up to 4× Faster Folding
The smarter grouping introduced in 0.4.4 shipped with a hidden cost: on real production logs it ran 2–4× slower than 0.4.3. This release removes that cost without changing a single byte of output — same groups, same counts, same markers, just faster. Folding now meets or beats 0.4.3 on every corpus we measure; a 7-day journalctl capture (655k lines) drops from 24 seconds to 6.
Under the hood: each line's tokens are computed once and cached instead of re-derived on every comparison, obviously-different lines are rejected by cheap exact bounds before the expensive comparison runs, and repeated lines find their group by hash in O(1) instead of scanning.
0.4.5 (2026-06-10)
Performance
- up to 4x faster folding with byte-identical output (c7efaea)
Full changelog: v0.4.4...v0.4.5