Love it when my coworkers reformat the code style, making it nigh impossible to understand what they actually changed, while greatly inflating their “contribution.”
It also blows away the git blame, making it hard to know who actually changed that one critical line of business logic 3 years ago that you need to understand before trying to fix some obscure bug.
I have one coworker who does this constantly and if you just looked at git blame, you’d think he wrote the entire code base himself.
I’m gonna make you break this up into multiple PRs before reviewing, but honestly, if your refactoring reduced the surface area by 20% I’m a happy man.
I try to keep my changes under 300-350 lines. Seems like a good threshold.
I’m still annoyed that Github doesn’t have good support for stacked diffs. It’s still not possible to say that one PR depends on a different one, and still has no ability to review and land them as a stack.
Making PRs against a feature branch still has the same problem.
Say you’re working on a major new feature, like adding a new log in flow. It’s a good idea to split it into many small changes: create initial log in form, implement user sessions, add SSO support, add “remember me” functionality, etc.
Those changes will likely depend on each other, for example adding the “remember me” checkbox requires the form to actually be built, and you probably don’t want to wait for the reviewers to review one change before continuing your work. This means you naturally end up with PRs that depend on each other in a chain.
Stacked PRs (or stacked diffs, stacked MRs, whatever your company calls it) means that:
The code review tool lets you specify dependencies between the PRs, for example the “remember me” PR depends on the initial login form implementation
It shows the dependencies visually in the UI, like a chain or tree
Changes can’t be landed until the PRs they depend on have been reviewed
There’s a way to land an entire stack
When you review a PR later in the stack, it doesn’t show any of the changes that were made earlier in the stack. Each PR focuses just on the changes in that part.
For each PR, the build steps and linters run for all the changes in the stack up until that point. It helps detect if incompatible changes are made earlier in the stack.
Making all your commits directly to a branch then creating a PR for the whole branch is similar, but reviews are a nightmare since it’s only a single review for the entire branch, which can be thousands of lines of code.
At my workplace, we use feature flags (essentially if statements that can be toggled on or off) rather than feature branches. Everyone works directly from the main branch. We use continuous deployment which means landed code is quickly pushed to production. New features are hidden behind a feature flag until they’re ready to roll out.
You can make a PR against your feature branch and have that reviewed. Then the final PR against your man branch is indeed huge, but all the changes have already been reviewed, so it’s just LGTM and merge that bad boy!
I suppose it is possible to have two PR that have changes that depend on each other. In general this just requires refactoring… typically making a third PR removing the circular dependency.
It sounds like your policy is to keep PR around a long time, maybe? Generally we try to have ours merged within a few days, before bitrot sets in.
Sorry, my comment was unclear. I didn’t mean a circular dependency, just PRs that have a chain of dependencies (e.g. PR 100 that depends on 99, that depends on 98, that depends on 97)
They’re usually not around for a long time, but there can be relatively large chains if someone is quickly adding new features.
also iirc gitlab does offer something like this as a feature now with “merge trains” (though i’ve never really used it, usualy just go for the feature branch out of habit x) )
Please, no, I get flashbacks from my 6-month journey (still ongoing…) of the code review process I caused/did. Keeping PR scope contained and small is hard.
From this experience, I wish GitLab had a “Draft of Draft” to tell the reviewer what the quality of the pushed code is at: “NAK”, “It maybe compiles”, “The logic is broken” and “Missing 50% of the code”, “This should be split into N PRs”. This would allow openly co-develop, discuss, and steer the design, before moving to nitpicking on the naming, formatting, and/or documentation details of the code, which is likely to drastically change. Drafts do work for this, but the discussions can get uncomfortably long and convolute the actual finishing of the review process.
Once both reviewer(s) and the author agree on the code design, the “DraftDraft” could be collapsed into a link in an normal Draft to be mocked next. The scope of such draft would be limited by the earlier “DraftDraft”.
I’m still annoyed that Github doesn’t have good support for stacked diffs. It’s still not possible to say that one PR depends on a different one, and still has no ability to review and land them as a stack.
I do like the idea of mandating git clang-format as the Kate project has.
That way the other devs don’t need to change their own IDE settings to comply.
Pre-commit hooks is a common approach to this, so that whatever is committed gets processed. Another possibility would be to set a bot on the repo to do automated commits after human-made ones, but that can get a little noisy.
Yeah I think that’s what he meant. You don’t want CI editing commits.
I use pre-commit for this. It’s pretty decent. The major flaws I’ve found with it:
Each linter has to be in its own repo (for most linter types). So it’s not really usable for project-specific lints.
Doesn’t really work with e.g. pyright or pylint unless you use no third party dependencies because you need a venv set up with your dependencies installed and pre-commit (fairly reasonably) doesn’t take care of that.
Overall it’s good, with some flaws, but there’s nothing better available so you should definitely use it.
I’ve used pre-commit pretty extensively over the years and I’m confused.
Each linter has to be in its own repo (for most linter types). So it’s not really usable for project-specific lints.
Not sure what you mean by this. I have pre-commit set up to do linting in several different projects, and even have it running multiple differently-configured lint jobs in the same repo.
Doesn’t really work with e.g. pyright or pylint unless you use no third party dependencies because you need a venv set up with your dependencies installed and pre-commit (fairly reasonably) doesn’t take care of that.
Again, I have pre-commit set up on multiple repos running pylint with multiple different plugins. Pre-commit absolutely does take care of setting up venvs with needed dependencies.
Not sure what you mean by this. I have pre-commit set up to do linting in several different projects, and even have it running multiple differently-configured lint jobs in the same repo.
I don’t mean using lints, I mean writing custom ones. Say you have a custom lint you want to use but it only will ever be used for that specific project. You can’t just put the lint code in a subdirectory. It has to go in a separate repo.
Pre-commit absolutely does take care of setting up venvs with needed dependencies.
Again I think you might be misunderstanding. It will install pylint fine, but if your project does e.g. import yaml, it’s not going to set up a venv and install pyyaml for you.
Say you have a custom lint you want to use but it only will ever be used for that specific project. You can’t just put the lint code in a subdirectory. It has to go in a separate repo.
You can run locally defined hooks with pre-commit, just define them in the repo: local section of the .pre-commit-config.yaml, and have it run a bash/python/whatever script or something that invokes your custom linting, wherever it lives in your file structure.
It will install pylint fine, but if your project does e.g. import yaml, it’s not going to set up a venv and install pyyaml for you.
Yeah I misspoke/misremembered there. For Python based stuff, it uses the currently active virtualenv or your global python install, so it relies on you installing your own dependencies. Which isn’t really that big a deal imo, because you need to install those dependencies to run/debug/test locally anyways.
You can run locally defined hooks with pre-commit, just define them in the repo: local section of the .pre-commit-config.yaml
Sounds like you’re just googling it rather than actually speaking from experience. Suppose I have written a Python lint and it’s in my ci/lints/foo folder. How do I tell pre-commit that? (Hint: you can’t)
Which isn’t really that big a deal imo
For small Python projects, maybe not. The project I’m working on has multiple sub-projects and those each have their own venvs, pyproject.tomls, etc.
Sounds like you’re just googling it rather than actually speaking from experience.
Like I said, I’ve used pre-commit for multiple years now. If you can run your lints from a command line, you can configure pre-commit to run them.
The project I’m working on has multiple sub-projects and those each have their own venvs, pyproject.tomls, etc.
Monorepos definitely make things a bit trickier, but again, you absolutely can write a local pre-commit hook that runs a bash command or script that 1.) activates the necessary venv and 2.) runs the lint command. I know this because I’ve done it, multiple times.
If you can run your lints from a command line, you can configure pre-commit to run them.
Yes but the whole point of pre-commit is it takes care of installing the lints. For most supported languages this requires the lint to be in its own repo. That is very annoying for project-specific lints that you would ideally want to just put in a subdirectory. Does that make sense?
can write a local pre-commit hook that runs a bash command or script that 1.) activates the necessary venv and 2.) runs the lint command. I know this because I’ve done it, multiple times.
Yeah there’s not really any point using pre-commit at that point.
Last time somebody did this to me there were a lot of sit downs about how to properly chop up large scale code changes and why we don’t sit on our own branch for two months.
“How long will this take to get in?”
“Well, two weeks for me to initially review it, a week for you to address all the changes, then another week or so for me to re-review it… Then of course we have to merge in all the changes that have been happening in primary…”
Last time I got this PR I was like, “Okay, I’ll do my best, but you asked the guy that has like 30 mins a day to actually focus and look at someone else’s code AND yours isn’t the only PR I’ll have to look at this sprint. Have fun reminding me about this for the next week.”
But part of working as a professional on a team is communicating and achieving consensus. Just trying to make a change like that out of the blue is poor form.
Also consider the opportunity cost: we had planned on getting XYZ done this week, and instead he spent a few hours on this and dragged a few people into a “do we want to change to poetry right now?” conversation
That wasn’t me, but that also used to be me. I learned to pick my battles, especially with complex code bases, and tried to keep scope creep in the name of improvement to like a dozen lines (provided it was fully tested).
I think it’s definitely a thing most people grow out of when they gain experience.
My boss told me about how when he was new he rewrote a whole chunk of the front end. His boss gave him a talking to about how you can’t just go and do that when you’re working with a team.
At an old job I just opened a PR to apply a code formatter to an internal project I wasn’t even a routine contributor to. PR was rejected and I learned a valuable lesson about talking and getting buy-in before making sweeping changes.
Some people, like me, are not built to be developers. I can sculpt code in any language I need for whatever problem I need to solve, but maintaining code over a long period of time, with others, is not my thing.
The drive to do additional changes is just too high and the tendency for typos or obvious logic errors is too common. (There is one little improvement. It’s right there. One line up. Just change it now while you are in there…)
I am not stupid and regard myself as a decent engineer but my brain is just wired in a more chaotic way. For some things that is ok. For developing code on a team, not so much.
Security is the field I am most comfortable with because it allows for creative chaos. Rule breaking is encouraged. “Scripting” is much more applicable and temporary.
Very relatable. Especially when it’s less effort to make the change than it is to try and ignore it. But it’s understandably harder for those who are reviewing your work.
I do. Also, I am old(ish) so I have had many years to come to terms with what I can do well and where I struggle.
In this case, I didn’t want to use it as a crutch or an excuse. After reading the number of awesome replies this morning, I realized I should have probably framed my comment differently.
People here put some real time and effort into giving some solid advice and I didn’t expect that.
Edit: As a pure example, this is the third or fourth edit of this comment. Words are challenging, and programming is very similar in that regard.
Thank you so much for your reply. Your comment was so recognizable to me, and I’m in the process of being diagnosed with ADHD.
Edit: I want to say that I do feel sorry for asking such a personal thing about you. I’m not young either, but just now coming to terms with it and figuring out that this is the reason why everything I do goes as it does. To recognize it in the wild is absolutely weird.
No worries! I am generally very open about it. (Your comment was recognizable to me, actually.) There is a very specific non-malicious bluntness that comes with the condition, actually.
But yeah, you have been practicing dealing with it your entire life. Treatment just helps a ton.
I’m the same way. Chasing code changes through the codebase then fighting with an edit rebase stack trying to explode them into less-interlocked changes.
It doesn’t always work, sometimes I am just committing a giant blob of changes at work on my project I near-solo maintain 💀
I tell my young developers - the primary goal in software engineering is maintainability. Code reuse, encapsulation, abstraction, and myriad other concepts all contribute to ease of maintaining source code over the long term. Maintainability allows for easier, predictable feature addition and removal, with fewer changes, and by different people. You’re also a different person than the one you were months or years ago when it comes to software.
Did I already post in here about how he wrote a custom DSL instead of using the standard widely used ORM we use everywhere? Maintainability nightmare.
He doesn’t work here anymore and now I have to either figure it out or rip it out. So far it looks like “rip it out” because it took less than an hour to swap in the orm, and now there’s just a lot less code needed.
When using git and are working on a feature, and suddenly want to work on something else, you can use git stash so git remembers your changes and is able to restore them when you are done. There is also git add -p this allows you to stage only certain lines of a file, this allows you to keep commits to a single feature if you already did another change that you didn’t commit (this is kind of error prone, since you have to make sure that the commit includes exactly the things that you want it to include, so this solution should be avoided). But the easiest way is when you get the feeling that you have completed a certain task towards your goal and that you can move on to another task, to commit. But if you fail you can also change the history in git, so if you haven’t pushed yet, you can move the commits around or, if you really need to, edit past commits and break them into multiple.
Refactoring in PRs just makes it more difficult to review. “Do these lines belong to the goal nor not?”. Also, we’re human and miss things. Adding more text to review means the chance of missing something increases.
Especially if the refactored code isn’t just refactored but modified, things are very easy to miss. Move an entire block of code from one file to another and make changes within = asking for trouble or a “LGTM” without any actual consideration. It makes code reviews more difficult, error-prone, and annoying.
Code reviews aren’t there to just tick off a box. They are there to ensure what’s on the tin is actually in it and whether it was done well.
In my experience I haven’t had an issue because usually the refactorings are small. If they’re not I just hop on a call with the person who wrote the MR and ask them to walk me through it.
In theory I’d like to have time to dedicate solely to code health, but that’s not quite the situation in basically any team I’ve been in.
I haven’t had any trouble separating refactors PRs from ticket PRs. Make the ticket PR, make a refactor PR on that ticket PR, merge the ticket PR, rebase refactor PR on master, open ticket PR for review, done 🤷
I have a rule about this (not that I don’t break it at times). I only refactor in an unrelated story if it doesn’t delay deliverables and existing tests cover the code.
And you’re generally right about tech that not being prioritized, but you should have a talk with your product manager/owner to strike a deal for some small percentage of your work to include tech debt. We were able to convince ours that it was otherwise affecting our velocity.
I hear you, but they should MVP the ticket, close it, then concisely collar the PM/lead and say “I know how to make this better and am hungry to do it. Let me address some tech debt next sprint. I got this and will keep it timeboxxed. I’ll also ensure my changes pass QA before coming to you”
This does seem like a potential issue if the PR is itself implementing more than one vertical slice of a feature. Then it could have been smaller and there might be wasted effort.
If the patches are small and well-organized then this isn’t necessarily a bad thing. It will take more than one day to review it, but it clearly took much more time to write it.
It’s harder to review. As a reviewer it’s difficult to know which code change is related to which task.
It’s harder to verify. Did you really test every change you made?
You might end up with a “hostage” situation. There might be a few code changes in the PR that looks good and is really wanted, but other code changes in the same PR of lower quality. As a reviewer, should you just let these lower quality code changes slide so you can bring in the code change you really want? Probably not, but you’re going to let it slide either way.
He’d have like a pretty normal PR to do something like change a field to be read only in some API. But then snuck into the pr there’d also be something like “change application to run locally different than prod for entirely selfish reasons” or “lower global coverage requirements” or “disable type checking in this whole folder”
I also worked with a guy like that. Impossible to predict what he was going to do in his next PR. Always a nightmare to review. Also exhausting to argue with, so let some things slide because I was so done dealing with his bs.
This guy was also difficult to argue with. He was always professional, which I guess is worth something.
One time he tried to remove all the types from a bunch of code “because they were all going to change later.” I refused to allow this. We went back and forth in the comments for hours. I think eventually the lead or boss got involved. Thankfully people sided with me.
His “later” project that was allegedly going to change all the things was rejected by the team. The code in question is still used and properly typed a year later.
Right but it’s pretty rare that a tiny PR actually accomplishes a valuable user story.
So my point is just that lines of code is mostly irrelevant as long as it’s organized well and does no more than necessary to accomplish the agreed upon goal.
Add comment