For a while I worked on a project where I would write some code, make a pull request, approve my own merge, let the CI process run, then promote my own code through 4 other environments to get to prod… all in rapid succession. No other person looked at it or tested it. I just had to work inside of this system someone designed for massive dev teams, but I was the old dev. There were about 15 people listed as approvers for the PR, but none of them would look at my code. I once put one out there, posted in a chat that it was out there and needed review, followed up, mentioned it on meetings… months went by. Finally some other stuff went to prod that needed my changes and everything broke, so I just approved it myself and promoted it all in the middle of the day, and no one seemed to care. That was the last time I ever asked anyone to review a PR.
We’v known this for twenty years and had the data ta back it up for ten. Github flow is one of the most damaging things to ever happen to software teams
Got asked about this twice so I’m cut/pasting my answer, but happy to discuss further
Check out the dora reports and the data Nicole Forsgren lays out in her book Accelerate. DORA reborts are free to access. She has found clear links between trunk based (no branching) development and a whole host of positive metrics. There is some suggestion that PRs are not too bad if always done at high quality and within the same day, but its weaker.
Omfg yes! Have nothing to add, but an upvote was not enough to express my hate for gitflow. So fucking stupid. And you’ll show stats from Jez Humble etc about trunk based, and my boss was still “eh not convinced”
Gitflow is far more complex and unnessaray for most places. You do not need a dev, main, and release branches. Github flow is far closer to trunk based dev - create a branch of master, PR back into master when done. If you keep your PRs small it gives you most of the benefits of trunk based dev with a CI check before you merge to the mainline.
Github flow has the same issues, in practice. Branching is the root cause, not the kind of branching. Even anonymous branches. Its the frequency of integration that matters.
what data? just curios because there are so many ways to do PRs properly… like for everything, if it’s done badly better not do it. does not mean it is inherently bad
Check out the dora reports and the data Nicole Forsgren lays out in her book Accelerate. DORA reborts are free to access. She has found clear links between trunk based (no branching) development and a whole host of positive metrics. There is some suggestion that PRs are not too bad if always done at high quality and within the same day, but its weaker.
depends on the company/team culture. are other people gonna have to fix or extend code you wrote? are you the sole engineer working on entire modules? do you hate feedback?
You must trust your team’s abilities more than I trust my own. How often does your team merge bugs into main? We use CI primarily for running the full test suite, including integration tests and e2e tests that would be very difficult to run locally due to them using specific credentials to access testing resources.
I trust my team. I don’t trust my or their ability to write and merge perfect code each time. I’m not sure how we’d fix that problem aside from becoming programming gods.
My point is, having PRs desn’t result in perfect code either. They might help sometimes, maybe 10% of the time if I’m being generous, but the rest of the time they are a hindrance. The problems people tend to try and solve with them, validation and indoctrination are better solved with a good CD pipeline, and pairing sessions.
The solution to that is pairing - spending a few hours collaborating with, and teaching this person will get them up to speed much faster than asynchronously nitpicking their code.
Mandatory pull requests + approvals within a team are a waste of everyone’s time.
For a while I worked on a project where I would write some code, make a pull request, approve my own merge, let the CI process run, then promote my own code through 4 other environments to get to prod… all in rapid succession. No other person looked at it or tested it. I just had to work inside of this system someone designed for massive dev teams, but I was the old dev. There were about 15 people listed as approvers for the PR, but none of them would look at my code. I once put one out there, posted in a chat that it was out there and needed review, followed up, mentioned it on meetings… months went by. Finally some other stuff went to prod that needed my changes and everything broke, so I just approved it myself and promoted it all in the middle of the day, and no one seemed to care. That was the last time I ever asked anyone to review a PR.
sounds like your company sucks. I’m sorry, must be lonely
Big hot take to me; especially in an organization with a large size and code high standard
We’v known this for twenty years and had the data ta back it up for ten. Github flow is one of the most damaging things to ever happen to software teams
Source?
Got asked about this twice so I’m cut/pasting my answer, but happy to discuss further
Check out the dora reports and the data Nicole Forsgren lays out in her book Accelerate. DORA reborts are free to access. She has found clear links between trunk based (no branching) development and a whole host of positive metrics. There is some suggestion that PRs are not too bad if always done at high quality and within the same day, but its weaker.
Omfg yes! Have nothing to add, but an upvote was not enough to express my hate for gitflow. So fucking stupid. And you’ll show stats from Jez Humble etc about trunk based, and my boss was still “eh not convinced”
gitflow != github flow
Gitflow is far more complex and unnessaray for most places. You do not need a dev, main, and release branches. Github flow is far closer to trunk based dev - create a branch of master, PR back into master when done. If you keep your PRs small it gives you most of the benefits of trunk based dev with a CI check before you merge to the mainline.
Git Flow and GitHub Flow are entirely different branching strategies.
Gitflow is has the same issues
Git Flow is awful I absolutely agree. On the other hand I like GitHub Flow.
Github flow has the same issues, in practice. Branching is the root cause, not the kind of branching. Even anonymous branches. Its the frequency of integration that matters.
what data? just curios because there are so many ways to do PRs properly… like for everything, if it’s done badly better not do it. does not mean it is inherently bad
Check out the dora reports and the data Nicole Forsgren lays out in her book Accelerate. DORA reborts are free to access. She has found clear links between trunk based (no branching) development and a whole host of positive metrics. There is some suggestion that PRs are not too bad if always done at high quality and within the same day, but its weaker.
depends on the company/team culture. are other people gonna have to fix or extend code you wrote? are you the sole engineer working on entire modules? do you hate feedback?
You must trust your team’s abilities more than I trust my own. How often does your team merge bugs into main? We use CI primarily for running the full test suite, including integration tests and e2e tests that would be very difficult to run locally due to them using specific credentials to access testing resources.
If your team consists of people you don’t trust, that’s the problem to fix first.
I trust my team. I don’t trust my or their ability to write and merge perfect code each time. I’m not sure how we’d fix that problem aside from becoming programming gods.
My point is, having PRs desn’t result in perfect code either. They might help sometimes, maybe 10% of the time if I’m being generous, but the rest of the time they are a hindrance. The problems people tend to try and solve with them, validation and indoctrination are better solved with a good CD pipeline, and pairing sessions.
Depends how good you are at what you’re doing. I’d argue that humans err and it saves a bunch of time to catch bugs before debugging in the wild
Unless you have people that are known to delivery garbage code. Which is more common than you would imagine, lol.
The solution to that is pairing - spending a few hours collaborating with, and teaching this person will get them up to speed much faster than asynchronously nitpicking their code.