A study of Google's code review tooling (Critique)

Same. Nor do I consider code review comments personal attacks, but anything that must be changed for "reasons," must be questioned. Not because it's personal, but because I legitimately want to know and I won't give up until I get a good answer.

> Code review feedback such as the following is all objective, beneficial to the reader, and not worth pushing back against

I'll agree that none of this is probably worth pushing back against, but it isn't objective. Very little is objective, in our industry.

> This logic should live in [other component]

Why? DDD will say one thing, MVC will say something different, though usually they are compatible or can be made to be compatible. You can also subscribe to hexagonal architectures and that might say something different...

There's nothing objective about where code should live, except on a hard drive or some other storage medium. I would argue that code probably shouldn't live on paper. I imagine most of us would agree with that.

> Our RPCs are named as GetFoo, not DetermineFoo

This is a nit. I'd honestly probably ignore it.

> That function name does not make it obvious what this code does, please change it

I'd probably ask for a suggestion because it probably looks obvious to me after staring at the code for so long. That being said, it might be a valid suggestion, especially if the code was refactored, but wasn't renamed.

> This needs a test

I hope nobody ever says this on my code reviews. However, I don't write tests for "obviously correct" code (code where the test implements the logic to test the logic): such as a function like:

    function returnTrue(): true { return true; }

> This is untestable and needs to be refactored to support X

Do you know there is a such thing as legitimately untestable code (or at least, it shouldn't be tested in traditional unit tests)? Usually at the edges of two systems. For example, an API integration can't be tested fully, only the known contract from the other system. Then you start running into Postel's Law ... things get weird. Only if you have some kinds of guarantees with the other system (not usually), would I recommend traditional tests.

I wasn't sure if I was going to reply to this part or not. But you took a comment out of context here.

I'm incredibly touchy about code reviews. I used to have the popular opinion here where "the code belongs to the team!" and all that. And then I was stuck on a team with a very toxic guy who weaponized this stuff. His code was "perfect" and he'd approve your PR to turn around and rewrite it just to show how much better it could be (and gloat about it -- not realizing if he hadn't deleted half the tests, they'd be failing). He'd constantly point out all the things that were "wrong" and with the justification of "code quality" without actually being able to point out why it was better. After working with them for two years, the entire team's philosophy turned into the one I've shared here ... because it can't be weaponized. It can't be used to tyrannically rule the code. It can't be used to bully people around "for the sake of quality!"

His "code quality tyranny" did more damage to the code than any actually bad code could have done.

So yeah, I will go into a war of attrition over certain kinds of comments that are spoken authoritatively yet are entirely opinionated ... because I've seen what happens when it gets out of hand. I have no desire to go through that ever again. Ever. If that means someone like you fires me for it, it's better than going through that shit again. Sorry, not sorry.

Do they not have PR's where you are from?

If someone is that annoyed by it, just open a PR. Just because I don't find it annoying, doesn't mean someone else won't. And vice-versa.

The code is mutable. It doesn't have to be perfect on the first iteration.

I will say this though: if you want me to do your nits, you also have to do mine. It's a two-way street when it comes to nits. If you don't do my nits, then I won't do yours. It's pretty simple. If you start withholding approvals because I stop doing your nits, then yay ... politics, I guess.

Once you land RPC DescribeFoo, I have to:

1) Define a second parallel RPC GetFoo

2) Wait for it to roll out to prod

3) Update all callers to call GetFoo, probably controlled with an experiment for safe rollover

4) Wait for all of those to roll out

5) Clean up all the callers

6) Clean up the DescribeFoo RPC

.....or you could just fix it in code review with half an hour of work.