Hacker News Reader

Comment by Waterluvian | Hacker News Reader

Makes me think of Theseus' ship. Change piece by piece out over time. Is it still the same repo people starred?

This feels a lot like what happens when a brand is sold. Is it really the same product?

Stars generally "work" because, I think, we assume repos converge to something. The older it is, the less likely changes are to be drastic. And maybe that's true. But it clearly doesn't have to be.

Of course we shouldn't rely on star count as any sort of quick and dirty metric of quality... But I think many of us do.

View on HN

> This feels a lot like what happens when a brand is sold. Is it really the same product?

It is why brands are sold. Someone buying a trusted brand wants to give their product starting levels of trust that it doesn't deserve. It's an effective way of cheating customers.

> It is why brands are sold. Someone buying a trusted brand wants to give their product starting levels of trust that it doesn't deserve. It's an effective way of cheating customers.

This seems excessively strong. Another way to view it is as a way of signalling intent (and proving skin in the game). Buying a brand is similar to those expensive and structurally pointless columns on old bank buildings: it proves the company has a put a lot of money on the line and doesn't intend to close up shop a week later.

I'm still hoping for a system that can evaluate the quality of a codebase given certain parameters and assign an average rank automatically.

The best way I know right now is to follow authors people trust like TJ or Sindre Sorhus, or to check on lists like https://bestofjs.org/

Which is a shame and shouldn't be happening, since newer authors (such as myself) are heavily disadvantaged from this method.

I'm extemely skeptical about automated code quality scoring tools. Most of them focus on superficial metrics and completely miss the important stuff.

One of the most important metrics for project quality is how easy it is to implement new features. Good code is easy to extend and build on top. Developers who haven't worked on the project before should be able to easily jump in and start contributing good PRs. I don't think that this kind of scoring is something that can be automated... At least not for a long time.

Actually this is somewhat already done in a quite clever way by the product CodeScene by Empear (https://empear.com/). This tool looks at the history of files in a repo and detects a lot of things such as:

* Correlation between files (we always seem to change b.cs when we change a.cs) * We have lost knowledge in the team, file a.cs was 85% created by developer X who has now quit. * A certain file is often modified (indicating too many responsibilities, and thus "bad code").

and so on.

More details: https://empear.com/blog/software-revolution-part1/

Reply on news.ycombinator.com