Dr. Michaela Greiler Makes Code Reviews Your Team's Superpower

Kent C. Dodds:
Hey friends, this is your friend, Kent C. Dodds and I'm joined by my friend, Dr. Michaela Greiler. I said it wrong. Can you pronounce it right for us?

Michaela Greiler:
Yeah, my name is Michaela Greiler, and because it's very complicated I'm also known as Dr. Michaela, which is a little bit easier.

Kent C. Dodds:
Yeah. Well, Dr. Michaela, I've developed a good relationship with her over Twitter over the last few months, and it's just been such a pleasure to get to know you and become your friend, and I'm excited for all my friends to get to know you. So could you introduce yourself to us so we can get to know you a little bit?

Michaela Greiler:
Yeah, sure. I'm also really happy that I'm here. I've been in IT for quite some time now, and just recently started my own company. I think that's also how we met a little bit more closely. I'm focusing right now on helping teams to make code reviews their super power, that's what I call it. At the end of last year I started to do that full time. Before that I was working as a software engineer and as a researcher at Microsoft, and before that I was working as a researcher in the Netherlands, also in the software engineering area.

Kent C. Dodds:
Yeah, and that's where you got your doctor title, right? Can you tell us a little bit about what it's like to become a doctor of software?

Michaela Greiler:
Um ... a bad idea. It took me around four or five years to do that, from 2008 until the beginning of 2013. I was doing that in The Netherlands, I was at the Delft University of Technology, and I was studying static and dynamic analysis techniques to help people understand software systems, in particular test systems.

Michaela Greiler:
What does it mean? It means that you have to get used to a very specific culture, people are writing papers, you have to understand how to write papers, how to conduct research. I think it was a little bit different from what I anticipated, a little bit less free, I actually thought that a researcher has a little bit more freedom.

Michaela Greiler:
So that's probably also why I left academia at the end, because I didn't feel that you have enough freedom to do whatever you want. I feel that, now that I'm on my own, or even at companies, you can deviate a little bit from the structured ways of doing things, the right or the wrong way. But in general, I enjoyed my time doing a PhD.

Kent C. Dodds:
Well, now you get to put "doctor" in front of your name, which is awesome. I have a master's degree in information systems, and I wish it was more socially accepted to put "master" in front of my name, so it's Master Kent C. Dodds. But it's not really a culturally accepted thing.

Michaela Greiler:
Yeah, I didn't do it for a long time, put the "doctor" in front of it. But at one point it was so painful to get that, I put it everywhere.

Kent C. Dodds:
You know what? I think that it's awesome. You worked outrageously hard, you absolutely deserve to have that there. Great, very cool.

Kent C. Dodds:
It's such a pleasure to chat with you. The things that you've been doing with your own business recently is lots of training around code review, which is interesting because I've never seen or heard of anyone giving training on something that might be considered kind of a niche topic, very specific. So I'm really curious to know what got you interested in code review, and what is it about code review that keeps you busy.

Michaela Greiler:
Yeah. It's not something that I really set out to do, it's more something that I grew into, I would say. So until now, I don't know if it's really a business right now. It pays the bills, but I'm really at the start, and as you said, it's a niche thing.

Michaela Greiler:
So how did the whole thing start? Even during my PhD, I was working on dev tools and developer practices. What are developers doing, how can we help developers understand code, for example, or use, as I said, static analysis or dynamic analysis to understand the software system? A lot of the things that I did were about code comprehension, how can you actually understand your code. Which links to code reviews, right? Because there you have to do the same, you have to look at some piece of code, and understand it, and ask questions about it.

Michaela Greiler:
A lot of the things were about communicating about code. When I joined Microsoft, I was in the Tools for Software Engineering team, that was an internal team that was helping all the other product teams to be better at what they were doing. For example, I worked with Office, I worked with Windows, I worked with Visual Studio, with all the larger product teams. We helped them with their testing, for example.

Michaela Greiler:
One of the things that we did was we analyzed Office and Windows testing suites and help them to make them faster. For example, test reduction, which techniques could you have there, how can you still make sure that you're not slipping any error through or any defects through?

Michaela Greiler:
And part of that was also code review. Our team owned the code review tool called CodeFlow, and so we were running studies because people were spending a lot of time on code reviews. Within Microsoft we have 40 thousand developers, so it's a valid question to ask, are the people spending their time right? If every developer only spends six hours ... and this is an old survey, right? But one of the questions that we asked was, "How long are you spending?" And they were like, "Over six hours." Each developer was spending six hours doing code reviews. So you ask, is that actually time worth spent? And so we were running-

Kent C. Dodds:
And that was six hours a week?

Michaela Greiler:
A week, yes, sorry. Yeah, six hours a week. So we were running studies to really understand, is it a good practice? In which cases is it a good practice? Because we had very diverse code review practices at Microsoft. So somebody who had Office, even within Office, would do it completely different from somebody who had Windows, and so on.

Michaela Greiler:
So we were trying to understand what makes code reviews really a good practice? What are the teams doing that are successfully doing code reviews, and what are the teams doing that say, "Code reviews actually suck, we're not doing them," or something like that? So we tried to really distill that. We did a couple of studies, and from what we learned we were improving our own code review tool. And I think it was a really cool tool at that time, CodeFlow. So this is how I got started.

Michaela Greiler:
And then when I wanted to somehow run my own business, which came because I had kids and things like that and I wanted more freedom, to be honest, I was thinking, "What did I really enjoy the most?" And code reviews was definitely something that I tremendously enjoyed, working with people. It's very sociotechnical, this means that you have the social aspect of it, but you also have the technical aspect, and it's very intertwined.

Michaela Greiler:
Then I started just blogging about it and writing what I did at Microsoft, because most of the things I didn't talk about much when I was at Microsoft. So later on I started talking about that, and then this somehow developed into what I'm doing right now.

Kent C. Dodds:
I love how you spent ...How long were you at Microsoft doing that work?

Michaela Greiler:
I joined in 2013 and I left last year.

Kent C. Dodds:
Yeah, so you were there for almost seven years, six or seven years.

Michaela Greiler:
Six years, yeah.

Kent C. Dodds:
And then the amount of stuff and experience that you had there just accumulated in an enormous amount of experience and knowledge around this idea of code reviews. You saw lots of situations where things worked really well, and things where it could have been better. And you're now able to take that experience and give that as kind of a gift. You're getting paid for it, but a gift that you're offering to other people so they don't have to go through the whole trial and error process where the errors can be pretty bad.

Michaela Greiler:
Yeah, that's true.

Kent C. Dodds:
So that's very awesome, and you know what? I think that it's really cool that we're at a time where people like you can specialize in that kind of thing and make a really positive impact on the rest of us who are working on our day-to-day products or whatever it is. We really don't have the time, or even want to take the time, to experiment with different things and see how different things work. So we can learn a lot from you as you teach from your wealth of experience and knowledge.

Michaela Greiler:
I think that, in general, niching down, it's something that I read about probably one and a half years ago, when I thought about what should I do with my business and things like that. I understood the theoretical concept of what it means, niching down, but I always felt like, "Oh, I need to do more. I need to do more, I need to know more," things like that.

Michaela Greiler:
But the funny thing is that, for example with these code reviews, first of all, I gravitated towards it again and again, over and over again. It was one of those things that I wrote an article, and I could write another one. And then I wrote the other article, and I could write yet another one, there was so much more. Over the years that I worked with code reviews, there's still tons that I can learn more about, and I can dive deeper into it.

Michaela Greiler:
Which is very, very interesting because I also was a little bit skeptical. Could I even make a business out of that? Could I help people enough? Is it not enough if you just go and read one blog post about code reviews? And the funny thing is that that's somehow also how code reviews work right now, people are just expected to know everything about it. I mean, how hard can it be? You're opening some editor, you're looking at some code, you're making some comments, and that's it, right?

Michaela Greiler:
And I think that's exactly why there is this huge variety of experiences with it. It's either really good, and it can be really, really horrible as well. There is not a lot of formal training around it. So yeah, I think there is tons that you can learn. It's about how do we communicate with each other, how do we learn, how do we resolve conflicts. Even those topics alone, you could again write a PhD out of it, right? Or do several PhDs there.

Michaela Greiler:
And there are also the technical aspects, what do you focus on, things like that. The policies, I think policies are also really interesting. How would you design your code review policies to actually really meet your goals that you have with code reviews? Things like that.

Kent C. Dodds:
Absolutely. I was actually going to ask you about that because I imagine some people look at this and think, "How could you possibly make a business out of this? I don't even see how you would need training, it's just open up the editor, type a couple of comments on the code, and then move on with your day." But it's a lot more than that, obviously. So what would you say are some of the things that people are missing when they just automatically assume that people know how to talk about code in a code review setting?

Michaela Greiler:
For example, what I see in my workshops is that junior drafts often don't even know what they really should" look at. "What should I say? What should I focus at? Sometimes it has to do with the fact that they don't feel in a position to give feedback to somebody that seems ahead of them, something like that. Sometimes it has just to do with the fact that they are not taught to do it right. It would give them such a self-esteem boost and confidence boost if they would know, "This is what we're actually looking for, this is how we give code review of feedback, this is how I phrase it."

Michaela Greiler:
But it's not only the junior devs, it's also how should we resolve conflicts. A lot of things are unconscious, so they are not aware of how this feedback can actually come across to the other side. I think also a lot of people ... If developers, companies, teams are doing code reviews, they're really spending a lot of time doing code reviews. And the question always becomes, "Is it really helpful?"

Michaela Greiler:
And I think a lot of people take the status quo, this is how code reviews are, this is how it is. It's just taking a week, and they're not thinking that it can actually change something. There are many ways that you could investigate what's actually happening in your code reviews.

Michaela Greiler:
It could be data driven, I did a lot of data driven studies at Microsoft where we really looked at millions of comments that we analyzed using AI or machine learning, things like that, and then we ran correlations, right? So for example, one study that I did was we were looking at code review comments.

Michaela Greiler:
And first of all, we interviewed developers, so we were just sitting with them, getting data on what is a valuable comment and what's not valuable. So we were sitting with developer over developer and manually marking what are they saying. Then you could use that information to build a decision tree, some classifier, and you can run that over millions of comments, and then you know, millions of comments, whether or not they are valuable or not valuable.

Michaela Greiler:
And then you can look, for example, for characteristics. So what's happened if a review has more than 1,000 lines of code, or 500 lines of code? You can fine tune that and see how that, for example, changes. So we had graphs where you could see the number of lines of code, or the number of files in a review, and on the other hand the density of valuable comments. And we could see that that's definitely going down, so you could see that there's a very clear correlation.

Michaela Greiler:
So, with those insights, you can actually do something about your code review practice, right? You open up and you see how many files are actually in my reviews. And if, let's say, there are 30 files, then it's not only a drag for yourself, but it also means that you're not in a position to give valuable feedback. It's just as easy as that.

Michaela Greiler:
Or if you are a dev and you're on a different team, and you're getting sent over code from somebody else, it's really hard for you to make any comments. We have so much data on that. People that previously looked at comments or are familiar with the code base, they have a very different ability to give valuable comments. So I wouldn't say that you should not include, for example, cross-team reviews, but you should have a very clear understanding of what is your goal with that, what should they look at? Maybe they should not look at the readability of your code base right now, maybe they should just look at the API. And you could also guide them.

Michaela Greiler:
So I think there are many, many things. As you said, if you're spending a lot of time thinking about how code reviews can be improved, and also looking at data, I think this is very important as well, then there are many things that you can do. It's just like many other things, you have to know it, to be curious about it, learn it. That was a very long answer, right?

Kent C. Dodds:
No, that was fascinating. I'm imagining the things that you could create with that kind of machine-learned algorithm. As I'm preparing my pull request, I could get a little editor notification like, "This PR is starting to get kind of big, you might start thinking about chunking this up or something. Or somebody invites me to review their pull request, and I get a little thing that says, "This one's a little bit bigger or more complicated, you might want to stop what you're doing, prepare yourself to review this well." Yeah, it's fascinating. Or maybe as I'm making the comment a little thing pops up and says, "Hey, that's a little bit maybe aggressive, maybe change the way that you're saying that." It's fascinating.

Kent C. Dodds:
I'm learning some of the things that you learned. You must have so much knowledge in your brain about how to go about making good code reviews. What would you say are some of the biggest dangers for a team that is not intentional about having good, positive code review experiences?

Michaela Greiler:
I think there are three levels that I always look at. One is the level of communication. I think you can really burn a lot of bridges, and you can really hurt people with the code review comments, be it intentionally or unintentionally. Even if you're not aware and you're just carefree, you can still hurt the feelings of people and leave negative marks on their lives. I think that's one of the things that I would like to stress, having better communication skills, being more compassionate about the other person. It's something that you can learn, compassion, empathy, all of those things, you can learn.

Michaela Greiler:
For example, nonviolent communication is something that I teach in my code review workshops, and it's really a wonderful tool. It's something also for, I would now say, people like us, this is very generalizing, as engineers, it's like an algorithm, it's something that you can learn. It gives you some recipe where you decipher messages and be more conscious about expressing your own feelings, your own needs, but also understanding the needs and feelings of others. So I'm a big fan of that.

Michaela Greiler:
There are other things like having small reviews, small reviews are really, really important. A large review, it's not possible for a person to look at the large review and give you valuable feedback. That's it. You can argue as much as you want. For example, in one of my workshops, or in some of my workshops, people would say, "But I cannot break it up," I have that over and over, "This is the unit, it cannot be any smaller." And you're looking at it, maybe 5,000 lines of code, and yes, you can make that smaller. I'm definitely telling you, you can make that smaller.

Michaela Greiler:
And I think it has a lot to do with ... It's the same as learning how to make a good software design. It's not easy at the beginning, but it's not hard, right? It's something that you have to train and to learn to understand where's the right abstraction layer, what am I telling you right now? The same you can do with code reviews. You can have smaller code reviews, and the purpose is probably different, and then you can compose those code reviews together into a larger review.

Michaela Greiler:
But then what you're looking through, the focus of your review, will be very different. Also, maybe the person that you are asking to review that code. So yeah, this is definitely something that I would like people to take away, make small code reviews. They are much easier to review for everybody, and you get better feedback from it.

Kent C. Dodds:
You know, when I was at PayPal, or pretty much wherever, if I was going to create a really big review and I knew I was making a big change that would impact many files, refactoring the API for something that's just used everywhere, one thing that I would do to help with that is I would make multiple commits. Here's the commit where I do this thing, and then here's the commit where I apply what I did to everything. And so it all looks very similar, it's just updating this code to change what the name of that parameter is, or something. Have you seen strategies like that work, do you have other strategies or good ways to break down big PRS?

Michaela Greiler:
Yeah, so one of the things that I advertise ... Maybe advertise is not the right word, advocate I think is the right word ... advocate for are stacked commits. I don't know if you've heard of that.

Kent C. Dodds:
No, I haven't.

Michaela Greiler:
I should write a blog post about it, I have a really long list of blog posts that I should write. Stack commits, so instead of just making the commits that you have in one PR, you would have several PRS, and each of those PRS is on their own feature branch. So you branch out each of those commits and make its own pull request.

Michaela Greiler:
And what's easier with that, it depends on your tool, but if you, for example, have GitHub or GitLab, then it's easier to review that because the stacked pull requests, you would start at the very end rather than work your way forward. You just see those diffs from that. So it's very similar to what you just said, but it's a little bit easier to even go about the review process itself. So if you can do that, I would do that.

Michaela Greiler:
Also, get familiar with Git, because in Git itself you can then just add parts of your changes and add them, for example, Git Patch it's called, you can patch parts of your changes in separate commits. I don't know if that makes sense, how I described it. Probably it's a bit better if you have something visual to see it. But it's a very powerful tool within Git. You can then, for example, mark each line that you would like to go into a certain commit, and redo those things.

Michaela Greiler:
For example, as you said, the refactorings, they should go into one specific thing, they shouldn't be merged with something else. Because it's much easier, if you have this mental model of, "Well, now it's about refactorings," then this shouldn't be mixed with any changes to the functionality and things like that. With parts of Git, you can redo them, let's put it that way. And there are other tricks, I should definitely write a blog post about it.

Michaela Greiler:
But yeah, this is something that I do some exercises on them as well, so that people know how to break those up. For example, at Microsoft there was a colleague of mine who was doing a study, and they were automatically detaching the commits and the reviews into coherent packages, and they were experimenting very with it. So it's very important, coherent commits.

Kent C. Dodds:
Good, good. Let's talk about the other side of the coin. So let's say that I am a developer, and I make a pull request, and the person who is reviewing my pull request is not being super nice about their review. What's something that I can do to maybe encourage better reviews, or how do I deal with people who just aren't interested in improving how good they are at doing code reviews?

Michaela Greiler:
I think it really depends on how bad it is, right? Sometimes I think, especially with written communication, there's a lot of research on that as well, written communication is difficult because the receiver has to interpret the message completely. There's no tone, there is no facial expression, there is no body language that you can read. So whatever the person says, you have to interpret. And there are studies that show that we are always on the negative side, people tend to negatively interpret that. So if there are two worlds, two truths or something like that, we tend to be on the negative interpretation side if you're reading something, much more negative than if we are hearing something.

Michaela Greiler:
For example, there was a study by Kruger, it was about two people. It was the same email, an email that was sent to a group of people, and it was coming from two different senders. They had no information about the sender, just a picture. One was a punk, something like that, a wild person, and then the other was a nicely dressed businessman, or something like that.

Michaela Greiler:
They're sending those messages, and the receiver had to rate their perceived intelligence of that person. So they had a Likert scale or something where they could say, "Do you think the person is intelligent or not?" Zero is no, that person isn't intelligent, and let's say four is they are very intelligent. It was an identical email, it was just the picture that was different. The punk guy, I think it was around zero-point-something perceived intelligence points. And the businessman was around 3.9, or something like that.

Michaela Greiler:
And then they had that person, who was the same person, call. No facial, right, they didn't see each other, but calling, so you had the voice. Then the people had to rate again and it was leveling out. So just by having the person talk to you and listening to them, the perceived intelligence almost leveled out for the two people.

Michaela Greiler:
This is just to say that written communication is really tricky. So what would I say? First is I would err on the side of caution, I would say maybe could you interpret it differently. Some people are very, very focused when they are working, so they forget all the niceties that you can do, right? Like saying hi, or thank you, or could you, or please. Maybe it wasn't meant that way.

Michaela Greiler:
Then you could make an attempt to express what you need. So I would probably go back to really nonviolent communication and thinking, "Well, what's the need that I have that isn't met here?" That would be I don't feel that they are valuing me, for example, or I don't feel respected, or whatever is going on here. So really thinking about what's wrong with this message, why are you triggered by it.

Michaela Greiler:
And then thinking what you would need, and thinking if you, in a very actionable, positive request, can make that. Say, "I would really appreciate if you could formulate your feedback in a more friendly way. For example, could you say please." You can be very concrete, it really depends on what's going on with that message, what's happening.

Michaela Greiler:
But if in the end you think, "There's something going on, this is really toxic behavior," then I think you can also ask somebody to help you, maybe having a mediator. This becomes a more tricky situation, I think there is no one answer to it. You would have to see what's going on there, and why it's going on there. But what I also see is that communities are formed by the behavior that we allow. So if you just let it slip through, then you will get more of it.

Kent C. Dodds:
Yeah, that's a very good insight right there. It sounds like, in general, first check your own bias against the assumptions that you're making on the message that they're trying to communicate, or the way they're trying to communicate it. But then especially a good takeaway for the project maintainers is that last thing that you said, the types of behavior that you allow are the types of behavior that will continue, and usually even grow. Yeah, that's a great takeaway.

Kent C. Dodds:
Well, Michaela, it's been such a pleasure to talk with you about code review, and as we wrap up here I wanted to just ask if there was anything else that you wanted to bring up that I just didn't ask a good enough question about, or if there's anything else you wanted to mention right as we wrap up.

Michaela Greiler:
Maybe just adding to the last thing, one of the things that I do recommend is to have a code of conduct for your code reviews. I think it's a good way to have that part of your policy, and really make clear what behavior is allowed and isn't allowed, and what will happen if you're breaching that. I think it's good, and I would advise that. Other than that, no, it was really fun talking to you. The time flew by, I didn't think it was 30 minutes already.

Kent C. Dodds:
Yeah, it's been a good time. Thank you so much for sharing all of your knowledge, it's just amazing the amount of stuff that you know about this, and just thinking about the amount of experience that you've had working with teams and improving code reviews is awesome. You actually do have a talk for people who want to dive in deeper into some specific tips. It's called 10 Tips for Respectful and Constructive Code Review Feedback. It's on YouTube, totally free.

Kent C. Dodds:
So our homework for everybody is to go and watch that, it's not super long, and then take a previous code review that you gave and critique your own review. See if there's anything that you could do to change that, and then maybe apply some of those learnings to your next one. Anything you want to add to that?

Michaela Greiler:
No, I think that's perfect. And if you can make that a habit in your team, to just review your reviews and see how it can improve, I think that goes a long way.

Kent C. Dodds:
Very good. Awesome, well thanks so much for coming. Just one last thing, what is the best way for people to reach out to you and maybe hire you to give them some training on it?

Michaela Greiler:
Okay. Well, the best way, you can find me always on Twitter. You will probably link that there because it's not an easy handle, it's "mgreiler" again. I should maybe change that as well. Also via my website, all my workshops are on there for corporate trainings. Just send me an email and ... Yeah, sounds good.

Kent C. Dodds:
Great. Awesome, thank you so much, it's been such a pleasure, and we'll chat with everybody next time. See ya.

Michaela Greiler:
Sounds good, bye bye.