> A better approach is to optimistically merge most changes as soon as not-rocket-science allows it, and then later review the code in situ, in the main branch. And instead of adding comments in web ui, just changint the code in-place, sending a new PR ccing the original author.<p>This may work for small projects or open source projects that generally receive high-quality PRs, but this sounds truly infeasible for large teams or organizations.<p>There are a couple reasons why the code review process occurs before merging. First, it helps keep the canonical version of the codebase in a "correct" state. This is, ironically, an extension of the "not rocket science" principle that the article mentions. Without this invariant, any checkout of the codebase might contain what is essentially "WIP" code that will waste other engineers' time if they have to interact with it. Second, the social pressure of blocking someone else's work is an important and necessary force for motivating code review. The idea that folks will go back and review code that's already been merged is akin to "will add tests in a followup PR". It's a pleasant lie we tell ourselves, but it rarely comes true.<p>Relatedly, the article also seems to suggest that the code REVIEWER should be providing the changes necessary after code review. This is problematic for so many reasons: first, it places an even higher burden on code reviewers' time which no one wants; second, it encourages poor code hygiene if someone else is just going to fix up your crappy work later on; third, it robs junior engineers of what is perhaps their most valuable learning experience: getting feedback from more experienced engineers and acting upon it.<p>(there are other issues here; the general thesis of the article leans heavily towards making code easier to _write_ rather than _read_, which again I think it just not appropriate for a codebase with a significant lifetime or number of contributors)