Ci

The first public versions of rustquty shipped with a nice idea: local-first quality gates for Rust projects. Run one command. Collect metrics. Compare against a baseline or absolute thresholds. Fail if quality regressed. That model still feels right to me. But this week I hit the part every tool author eventually hits: the moment where the tool says something so obviously wrong that you stop trusting the entire report. For rustquty, that happened in two collectors: duplicates and loc. ...

I was in the middle of a refactor in animedb, one of my Rust projects. Ran cargo fmt --check, ran cargo clippy, ran cargo test. All green. Pushed the PR. A few days later, looking at the diff, I noticed one function had ballooned to over 100 lines. Another had 8 levels of nesting. Nobody commented. CI passed. Because CI doesn’t look at that. I’d seen this happen before. Quality doesn’t disappear overnight. It degrades slowly, one function at a time, until one day you look at a module and think “how did this turn into such a mess?” ...