Hacker News new | ask | show | jobs
by couchand 1647 days ago
CI should be sufficient review for most code changes. Manually review architecture and design, but automatically verify implementation.
2 comments

But if you push a bad commit to main without making a branch or PR first, then CI tells you about your mistake... after main is already broken and not "deployable any time"!
Just thinking out loud: why in that case we couldn't have automatic revert of the push?
The review should be perfunctory, but a second set of eyes is always needed
That seems inherently contradictory? Particularly if you put some cost on a) not being able to merge and build on the feature in another branch, b) disturbing someone to do something perfunctory
It is an inherent contraction. Having good testing and fast flow is a contradiction. The advantage of a second set of eyes is simply a sanity check... from a second pair of eyes. It's all a balancing act, but it does prevent a large set of failure classes that are basically "This dude went crazy"