The biggest benefits are:
- Breaking the framing that a review should only address code that has changed in a PR, and encouraging a more holistic view of the codebase. You don't have to review every merge every time, just review code at a granularity that makes sense when it makes sense to do it.
- Removing a delaying step between code being functionally complete and being delivered, where value invested is sitting on the shelf waiting for a reviewer.
- Forcing a high test pipeline confidence. If you're relying on code reviews to catch functional problems rather than stylistic or structural ones, yeah, you won't have the maturity to do this. Saying "we'll review after merge" is saying "we have enough confidence in our automated quality gates not to rely on a human before then."
To address your worries:
- Cost of context switching is removed, not made worse. If you insist on a review before merge, the author has to wait for a review, and unless you have a culture of reviewers context switching to do reviews immediately when they are requested, they will probably pick up something else. Now when the review comes back, what do they do? Do they leave the review waiting, increasing the likelihood of a merge conflict as it gathers dust? Or do they context switch to deal with it and realize the value invested? Contrast this to review-after-merge: the value is delivered, and there's no merge conflict potential; the reviewer can get to it when convenient; and the result of the review can just be another set of tickets on the board to be picked up in the normal sequence. No context switch required.
- Disconnect the review from the PR and the problem of conflicts goes away. Instead review a module at a time (where "module" can be whatever scope makes sense: file, import, header, whatever), so that you're always looking at the whole context rather than the few lines that happen to have changed. That also minimises the human tendency to find as many issues in a 50 line diff as in a 2000 line diff.
- The tooling issue is to some extent an availability bias. Just because tooling for a specific process is well-known does not make that process good. It can nail on harmful practices and make them hard to change or even to see how harmful they are. And yes, if you're doing something close to trunk-based development then either PRs get very small or they go away entirely - branch management is orthogonal to when reviews happen, but you can't usefully move in that direction without addressing the delays inherent in review-before-merge.
- Learning and understanding would only decrease if reviews stopped entirely, rather than moving when they happen. The reviewer always has `git blame` to direct feedback to the right place, and by expanding the scope of a review from just-the-diff to all the code around it, the reviewer has more of a learning opportunity, not less.
It's possible you do have a healthy review culture, but the question I would have is this: how long do PRs sit on the shelf waiting for a review, or for review feedback to be addressed? Do you track that number? And are you relying on humans to catch problems that should be embedded in automated quality gates? Moving the review out of the commit delivery cycle removes a potential process bottleneck, and it's a very easy one to be completely blind to.