I'm working on a pure-Rust rcsfile(5) parser for FreshBSD's CVS backend, so I can get away from shelling out to a hacked-up copy of rcshist(1). Going pretty well!
It is — of course — blazingly fast. Across all 1.3 million, 53.7 GiB-worth of OpenBSD and NetBSD `,v` files I have:
gorcs log: `30:15.06 real, 56:33.42 user, 5:53.63 sys, 2019 MiB`
GNU rlog: `4:48.29 real, 3:35.93 user, 27.650 sys, 357 MiB`
rcshist: `4:46.17 real, 3:47.13 user, 58.865 sys, 9293 MiB`
BSD rlog: `4:21.49 real, 3:51.37 user, 30.075 sys, 392 MiB`
rcsfile-rust: `1:45.68 real, 1:08.45 user, 37.199 sys, 205 MiB`
Note `rcshist` loads all passed files in one go so it can give you a list of revisions in date-order, so it gets a pass on its peak memory use. It also generates full unified diffs instead of a simple line count.
`gorcs` gets a pass because... I dunno, I suppose you'd have to ask Gemini? But it's apparently good enough to drive the next generation of cvs-fast-export.