Hi, rsync has started to allow LLM-generated code in. This has already resulted in user reports that saw subtle changed behaviour. If you look at https://github.com/RsyncProject/rsync/commits/master/ you can see all recent commits being co-authored by claude. Since this is also very questionable copyrightwise (since LLM code cannot be copyrighted), please revert rsync to 3.2.1. best, werdahias
Why should this be a problem? The Linux kernel, LLVM and many other projects have done the same. Debian also allows this. "Changed behavior" is also expected with any code change: otherwise one would not need any code change. Code that cannot be copyrighted would be public domain. This would not be a problem either and indeed be less problematic than highly impractical licenses like AGPL. Ansgar
Hello, rsync has been in dire need of funding and contributors for at least the last couple of years. The biggest regressions started in January 2025 with the CVE fixes preceding AI contributions, the biggest regression concerning symlinks is still pending 1+ year later. Reverting rsync to the pre-LLM version means reintroducing fixed CVEs (a different wave than the ones from January 25), but ultimately it won't solve the lack of funding and contributors. We cannot reintroduce the CVEs unless we are certain the regressions are worse than the CVEs, I'm not there yet. I've started packaging gokr-rsync last year when I noticed alternatives would be needed, but a lack of time blocked me from proceeding. puida@ helped me package the reverse-dependencies and now very little should be pending before the package is ready for upload. However, out of all rsync alternatives, the maximum supported protocol version is v27. https://salsa.debian.org/go-team/packages/gokr-rsync Regards,
On Sat, 30 May 2026 17:16:59 +0200 Ansgar files/lines are public domain and which are licensed otherwise (such as GPL)? Example law that might be relevant: https://www.law.cornell.edu/uscode/text/17/1202 Regards