[graemep]

> The Linux Mint Timeshift tool has an issue open documenting a number of regressions that are currently open on the rsync issues page, that were only introduced post-vibecoding (https://github.com/linuxmint/timeshift/issues/548)

There are four actual regressions there. The commits that introduced two of them have been identified, and neither neither of those mentions Claude (or another LLM).

If you look at the actual commits that credit Claude, they are not huge commits (many are a few lines), most are tests and config. This is not vibe-coding.

> there are multiple reports from users in industrial and government settings that now have to go through whole processes to update this software

If people are handling such critical data with it they will be testing upgrades before deploying to production right?

> I wonder how many more issues have been introduced that we simply won't learn about until a company has a $10 million dollar data loss because they were not testing their backups consistently.

If a backup failure would cause a $10mn loss then its grossly irresponsible to not test your backups.

[dmix]

https://en.wikipedia.org/wiki/Brandolini's_law

I don’t think people really care about rsync or the nuance. They just want to make an insta-reaction, rant about AI, then move on to the next story that raises their blood pressure.

[culi]

I don't know how deeply you read into it, but people pointed out that Claude rewrote the entire testing stack in Python. Worse than that but it rolled its own unique framework. Every test file will randomly redefine its own `_run_and_capture` function

How could we even check if either human or robot code is working properly if we're not even sure the test suite works?

Also, another user[1] compiled a nonexhaustive list of 7 issues they found introduced because of the changes.

[1] https://github.com/RsyncProject/rsync/issues/929#issuecommen...

[graemep]

If the 7 issues three are the same underlying issue. Another two at least relate to the same commit and probably the same underlying code. One is not related to a Claude assisted commit. That leaves three that are.

Adding that to the two in the issues further up, that makes a total of five bugs in AI assisted commits.