Fire-Side Chat with Kent C. Dodds



Hello. Hey Kent, how are you doing? I'm doing great. It's a beautiful morning here in Utah, kind of rainy, but my wife likes the rain, so I'm happy. Happy wife, happy life. And of course, happy kids with managing four kids. Like I said, crazy managing that with your career. I know that you are really happy with your kids, and I'm also happy with my kids, of course, but how are you managing now with the pandemic? They're probably more at home and managing your career. Is that difficult or is it okay now? You know, I was already doing a lot of remote stuff before the pandemic hit, lots of remote workshops and things. And so the pandemic didn't really impact my career a whole lot in that regard. I mean, everybody was impacted and my kids started doing school at home and everything, so it shook everything up. But yeah, I wasn't impacted as much as everybody else. Good to hear that. So as much as I would love to talk about your kids, I also have a son and yeah, like to talk about it. I think our audience is not here for talking about your kids. They want to talk about testing. So a few weeks ago, we've sent out a tweet that people can ask some questions that they want us to discuss. And the first one is from Peter Hozak. You are migrating some of your projects to typescript. What are the biggest pain points for you so far? And what can you recommend to the people or teams who are not yet decided whether to start the migration in their projects? How time consuming would you say it will be? So it will be relative to the size of your project and the experience of your team. So I can't really tell you exactly how time consuming it would be. Migrations are always really tough. You have to take that with a grain of salt. I actually have a blog post about engineering and business alignment and how important it is that you do what's best for the business, not just what you want to do. It's actually a pretty good blog post. You can go take a look at it at As far as like recommendability, I definitely recommend it. I think it's a great idea. And there are some strategies that you can use. And to be clear, I haven't ever... I'm trying to remember. I think we actually did migrate one of our projects at PayPal to Flow. And that was in process as I was leaving. But here are a couple of tips. So first off, it doesn't have to be perfect, especially if you're just learning typescript. So I would not turn on strict mode. It's actually similar to adding ESLint to a project that's never had Linting. I wouldn't recommend enabling all of the rules and fixing it all in one giant PR. But rather disable most of the rules. And then over time, you add them as warnings. And then you fix things here and there. And then you turn them on as errors or whatever. So in the same way, just make your typescript really loose at the very beginning, especially if you don't know typescript very well. You don't want to just spend a ton of time writing your types and they end up misleading you. And you write it poorly because you're new at typescript. So throw an unknown there. Throw in any here. Whatever you need to do to get it going. And then come back later when you know typescript better. So yeah, I would recommend it. It will take a...I would recommend it iteratively over time rather than just one giant PR. But yeah, I think it's a good idea. Yeah, that's also my experience. Going with a big bang. I mean, besides that it just takes a lot of work from you. You're going to send that PR to someone and he's going to be your archenemy after that, right? So if you want to keep friends with your colleagues, then don't do the big bang. Even if it's a small project, migrating to typescript, even though it's really valuable, it is a lot of work and yeah, don't go with the big bang strategy. Next question is... You're going to make a different story. You're going to big bang prettier all day. Next question is from Bazaat. Nowadays that next.js looks so solid. Should we use next.js instead of Create react App or Create CSR? I'm not sure what CSR is, but yeah, so next.js is solid. Sorry, I'm going to stop you for a little bit. And I want you also to chime in if you're going to next.js on the vendor login. Oh yeah, sure. So yeah, next.js is solid. I used it years ago. I haven't actually used it recently, but a lot of people use it and they love it. And yeah, so I don't think that you can go wrong with using next.js as a framework. I'm really, really excited about remix personally. For the last several months, I've been using it to totally redo my personal website. I'm really excited about a big rewrite that we are working on. So that's what I'm working on, but it is still developer preview. It has features that you won't find anywhere else. And that's what's compelling about it for me. It's not just like a paid version of Next. It's very different from that. Although a lot of people like to say that without any experience with remix. So yeah, but next.js is totally solid, if it looks interesting to you, then go for it. And do you have your remix project in production also already or is it in development still? It's still in heavy development right now, but we're talking like a month or two before it comes out and it'll be pretty sweet. Nice. Well, looking forward to that. Probably you'll announce it on your Twitter and all your channels, of course. Oh yeah, yeah. You won't miss it. If you follow me on Twitter or subscribe to my newsletter or something, you will definitely hear about this. Yeah, I know when you do something new that it's hard to miss. That's a good thing. Because I always like to see what you're doing. So wouldn't want to miss it for the world. Next question is from Carlos. Is the testing library a new mindset for front-end tests? I mean, we will think about user behavior. So maybe it's not to test some component like a button or even hook and test it integrated in a page. So it's more like, yeah, kind of like a cypress test that you're really going through a flow. Yeah. So I don't often test lower level components that don't have very much logic into them because it's so easy to cover those things. Like if you have a little button or something, like you would not find me testing a button unless there was some fancy complicated logic that I wouldn't easily cover in a higher level test. I have a blog post about this as well. You can find it at slash trophy where I describe the different layers of the testing trophy. And one thing that I say in there is actually I'm quoting an interview that's on testing where it's basically testing these different layers of testing are kind of like painting a wall. And the quote is you can throw paint against the wall and eventually you get most of the wall, but until you go up to the wall with a brush, you'll never get the corners. And so, yeah, like your end-to-end test, that's like just taking a bucket of paint and throwing it at the wall. But eventually you need to go with more fine-grained tests and maybe you've got like a painting that's your grandma made it or something and you don't want to even paint with a big brush. You'll take a smaller brush around that. And so that's kind of the way I think about different layers of testing. There's no reason to take a small brush to paint the entire wall. It will take too much time to get full coverage. But yeah, so I focus more of my time on integration tests like that's a paint roller. You cover most things really well with that and certainly end-to-end tests as well. But you can use testing library for all of these. That's the cool thing about testing library. Anywhere a DOM can be found, a testing library can help you to query for elements and interact with those elements. When you were mentioning coverage and I don't think there is a real answer to this, but we can ask it anyway, see what your thought is. What's a good coverage for most projects you would say? Would you always say 100% or? Yeah, I actually would assert that 100% is a bad coverage number. Maybe not as bad as 0%, but yeah, still a bad one for most of the time. Now if we're talking about an open source library, then yeah, sure, 100% that's typically relatively easy to accomplish and there's a high value for that because this is highly reusable code, right? So it's a little bit different. But for applications, if you're shooting for 100% coverage, you typically end up spending a lot of extra time testing things where you'd get a higher return on your investment working on features instead. And so I actually have a blog post about this titled How to Know What to Test. And in that I kind of talk about how code coverage is a useful metric, but a better one would be use case coverage, which is not really reasonably easy to set up as a metric, but it's the way that you should think about your test. So I don't often look at my code coverage report. The only time that I look at that is just to remind myself of use cases that I may not have covered, but I'm always thinking about use cases. What I really like about, well, it's not actually about the coverage, but if you look at that report, I think it's the Jasmine report it's called, where you see your code and then what branches are and are not touched by the code, by the tests. That is a way I really like to go through that and see, oh wait, this is a path that is dangerous that I didn't test it. And then I'd have to ask for that. Yeah, exactly. So that is a good thing. And yeah, like you said, going really for a hundred percent or whatever the percentage is, I've had companies where there was a percentage set and you just know that at a certain point you are going to be writing tests just to satisfy your coverage needs, even though you kind of already know like, this is fine. I don't need to test it, but yeah, my pipeline will fail if I don't test now. So yeah. Yeah. But don't go there. Yeah. Yeah. Next question is from Henrik. What's the best piece of advice that you've received career wise or also in general? What advice would you give now to someone who's in the beginning of their career in tech? Great question Henrik. Yeah, that's an excellent question Henrik. So I think one of the most important things, the first thing I have to say is to be nice. It doesn't matter how good of a coder you are or how good of a, you know, people, well, yeah, you just got to be nice. If people don't want to work with you, then you will not get the work that you want. So if you want to have control over your career and really accomplish whatever goals that you have, you will be able to do that much better if you're a nice person. So then on top of that, I would say communication is a really important career skill because even if you are the best coder in the world, if you're not good at communicating what you've accomplished and how that impacted the business or improved lives for people, then that also will make it difficult for you to get the work that you want. You see this often with people who are really good at something or at least they're the one who doesn't complain about doing something. And so they get all of that work when actually they'd want to do something else. And so you need to be able to communicate what your career goals are and what your accomplishments are. Like you just get really excited about the things that you want to keep doing and you say, hey, look at this cool thing that we did. I'd love to keep doing this. Or hey, this is a cool thing that we did, but I'm not super into that. Could I start doing some more of like the front end side or the back end side or whatever? So communication has been a really, really useful tool for me. Yeah. I love that you're saying that at a tech conference that you don't mention any technology, just be a nice person. And I love that answer. Thank you. I'm going to go look at our Discord channel, what the input is from our audience, the live audience. All right. Oh, I see RTL. I haven't even read the whole question because this is something I just want to know, because what is your experience with testing RTL components? And do you have any tips or any resources you can recommend for people getting into RTL? We've had some talks yesterday also about the internationalization and then I want to get your side, like making it secure. Yeah. Yeah. So Daria gave a great talk yesterday morning on internationalization. So and actually she mentioned my library, RTL css JS, which is pretty handy. So yeah, I've never written a test specifically for RTL. RTL is also an acronym for react testing Library. So maybe that's a little confusing, but right to left is what we're talking about. But yeah, so I've never written a test specifically for that. If I were to, then you'd probably be best. This is like a visual thing. Or I mean, there's obviously there's the actual content side of it as well. Most of my tests or all of my tests use the default language as the content for like querying around for things and entering in texts and stuff. I feel like you just get a marginal improvement in confidence by adding testing for different locales and so that's why I don't bother. But RTL is definitely a feature that you don't want to have break. And so if that is something that's really important for you and your business, then you probably want to go for visual regression testing. And so that's we're talking about tools like Appletools or, where they will use their fancy AI to figure out the right way to do visual regression testing. Because I don't know about anybody else, but visual regression testing is notoriously very finicky flaky. And so these technologies have advanced enough to reduce the flakiness of these types of tests. And that's how I would go about testing RTL if that was a really critical piece of my business. Yeah, well, of course, there are a lot of businesses where it matters if you're like the thing I know about testing RTL is the Netflix blog post. It was like two years old and they shared their approach that even when also doing left to right LTR that we're adding, I think, 20% extra characters in every string just so they kind of know for sure that if there's a language where the string is longer that the interface won't break. So I don't know if you've seen this blog post, but that's something if you want to do RTL testing, I would recommend looking up on the Netflix engineering blog, which is also just a nice blog to follow. is the one I've mentioned. Question from Luke. Does react testing library get tested itself? How do you test your testing library? Yeah, nice. Yeah, yeah, sure. So you think about it, testing code is no different from any code, any other code. In fact, even the test that you're writing, that's just code. And you could write tests for those. You could take your test function block, you could put that in a function, and you could call that function and make assertions on what it does. So like, it's all just code. And when you're writing a bunch of tests, eventually you come up with some utilities that may or may not be useful for anybody else. And that's what react testing library is. It's a bunch of utilities that I wrote for myself that I thought, wow, these are really useful. Let's make a library out of it. And then teach people how to use it. And that's testing So yeah, it is definitely tested itself. It's mostly just like javascript utilities. And so it's not, you know, it's very different from a testing framework. But even the testing framework itself is also tested. That's actually even more interesting, because Jest uses Jest to test itself. And so, yeah, that's where things get a little meta. react testing library doesn't use react testing library to test itself exactly. I mean, we are calling render when we're testing the render function. We are calling, you know, get by text and get by role and all that stuff to make sure that it's doing what it's supposed to. Most of the time we're in general, I recommend that your test should resemble the way that the software is used. And so our tests for react testing library do that as well. And so if you go through the react testing library test, you'll probably find some that are for lower level utilities that are like more lower level unit tests that you wouldn't write. But lots of them are integration tests. And so they'll look very similar to the types of tests that you would write. And if our tests are passing, then we assume that we won't break your tests, hopefully. Hopefully. Well, it sounds kind of scary. Like you said, Jest is using Jest to test their code. Because if Jest breaks, will the test also break? Yeah, that's how do you know that? Yeah, I believe that there are some languages that are written in themselves as well. I don't know how that gets started, but eventually they switch over to their own language to write their language in, which is interesting. That's extra meta. We have a question from Sasha. Do you think integration testing is also important in addition to testing/talks">unit testing for component library? And how do you feel about integration testing in general? Yeah, well, I'm kind of, I have a blog post that's pretty popular on this topic called write tests, not too many, mostly integration, which is a quote from Guillermo Rauch from years ago. And yeah, you can take a look at that. It's slash write dash tests. And so the testing trophy is weighted toward integration tests. I focus on integration tests. For component libraries, the line between integration and unit test kind of gets fuzzy. Some people say unit test is just the unit. You mock everything around it. And if that's how you define a unit test, then I don't write many of those at all. And most of those are like, I have a pure function. It's pretty complicated. Let's just test that by itself. And that would be my form of unit test. So if that's how you define a unit test, then a component library, I pretty much wouldn't use almost any unit test. It would all be integration. I want to render the component, I want to interact with that component. And yeah, so I focus most of my time in the integration space. So let's go a step beyond component libraries. Let's say you have a design system. So just a collection of components in the end. Would you also be adding tests for a design system then? Yeah. So then you would be testing that button. Yeah. Yeah. And actually at PayPal, I was responsible for building that. That was the last thing that I worked on before I left. And yeah, I was testing every component. But there was always logic that I was testing there. And when it is like a design system or component library and you have a button, I would at least have one test that says, this thing renders. Because it's just so low cost to have that sort of a test. And because it's a reusable design system or component library, I want to make sure that I didn't have a syntax error or something. I mean, it was in typescript. So the likelihood of having a problem is pretty low. But then would you say like a snapshot test would be enough? Yeah, snapshot tests. I have a blog post about this called effective snapshot testing. In component libraries, maybe. The thing is that snapshots are just assertions. They're no different from any other assertion. So when you have an assertion, you want to make it so that a failure, when you have a failure, you know why it failed as much as possible. That's your job as an assertion writer. And when you take a snapshot, you could take like a snapshot of the world that could include like where the sun is and like what operating system you're running and like literally everything. And this is the state of the world. And that would not be a useful snapshot test. Of course, we don't do that. Often I will see snapshots that are hundreds of lines long. And so what I'm saying is a snapshot often includes a lot of information that is totally irrelevant to the test. And so I really rarely use snapshots for UI testing. It is very rare that you'll find me doing that. What I do is if I want to start with a snapshot, I will slowly get rid of all the stuff in the snapshot that is irrelevant. So I'll find the element that I'm trying to assert on. And then I'll realize, oh, all I care about is that this attribute is, you know, this certain value. And so instead of doing a snapshot, I'll just make an assertion on that attribute. So yeah, you can read more about that in my blog post about snapshots. Yeah. So if you actually want to, let's say we're going back to a button and you want to test if it accepts a class name, your component for that button, then you can do that with a snapshot. But if the snapshot test fails, yeah, you don't know what you're testing anymore unless you've written a really good description. So it's better to just say it should accept the class name and then, oh, I am failing with my testing library syntax in my head. I was going to write it. Yeah. You just select the button and say to have class name. And then it's very clear. Like I pass it this class name. I expect it to have the class name. If that fails, then whoever's reading it, even if it's me in six months, I know exactly why this failed. I may not, or I know exactly what the object is. I may not know why it happened, but I know what the intended outcome is. Yeah. That's, that's a good point. Sometimes even you're writing tests for yourself. I haven't always been writing tests. I have a quote from a friend of mine, Frank de Jonge, and he once tweeted, I don't write tests. I write proof. And that's always like, I'm proving to myself that this works as I intended. And I really liked that mindset for myself. So that sounds good to me. That's insights. Next, we have a lot of people mentioning, by the way, like go is written in Go, C is written in C, typescript is written in typescript. So really cool to have that interaction. It's a little mind bending. Yeah. It's too meta for me. To follow up, Juan is asking to follow up on the communication tip, what communication resources would you recommend? Nice. Thanks, Juan. Ooh, practice, I guess. I don't have any resources to offer. I'm sure that there are books and things like that. This is something that kind of comes by naturally for me. Maybe I'm a little on the obnoxious side, but I share everything. That's like, that's why, that's often why people think that I get so much done is because I literally share everything that I do. So I may not be getting more done than anybody else, but I just share everything. Yeah, I don't really have any specific advice. Well, I don't know, this may not be totally related, but one podcast that comes to mind is the soft skills Engineering. I really love that podcast. It's hilarious. Just its entertainment value is good all by itself, but they have some really good advice often about communication. All right. So would you mind sharing a link with that in the Discord later on so that Juan can find it and the rest of our audience also, of course. And to follow up from my side, just talk to people. If you want to know if people like you, you'll probably get a fair answer if you just go down for a one-on-one, but it might be scary. So yeah, if you don't like that, I would understand. We have a few, yeah. Let's take a moment. We are reaching the end of our fireside chat. Just to be sure that we don't go over time, let's take a moment to go through the poll results. So Kent has asked what the percent of time you were spending on the different things of testing. And the winner is Hooks, 100%, with 46%. That feels really weird. 100% with 64%, sorry. In Dutch numbers, you say them the other way around. I'm sorry. It's always hard for me. So Kent, is this what you were expecting? Yeah, I've been doing this poll on Twitter over the course of, since Hooks was released, every few months I'll give this poll. And it's just very steadily going in the direction where very few people are not using primarily Hooks. And this is why I totally dropped classes from all of, like class components, from all of my teaching material as soon as Hooks came out. I was like, I'm done teaching that. You could go look at my old stuff if you want to learn about classes. I am 100% Hooks everywhere now. Yeah. Well, that's good. You got to keep up with the modern stuff. And you don't want to write something that people are not using anymore. So that's great. Right. I also remember- You want to stay up to date. I also remember that a few months, or maybe it's over a year ago, you asked if you should write typescript for your material, right? On Twitter. Yeah. What was the result on that? What did people like? Yeah. Well, sometimes I do polls just because I'm interested in what people say, not because I'm going to change what I do. So yeah, I kind of do what I want. And the people who are cool with that will follow me along. But yeah, so I am definitely doing typescript for all of my stuff. I already have a... I'm in the process of obtaining all of Epic react to be in typescript 100%. I'll have to rerecord all 350 videos for that, which will not be fun. I am going to make a typescript course or a series of workshops for people. So if you haven't done typescript before, you can go through that. I also already have a script that'll remove all the typescript. So if you don't want to do typescript, you can do that too. But I just think that typescript, it makes even the learning experiences better because you get auto-complete and it just, that's an enormous benefit as you're learning. So yeah, typescript all the way. All right. Great. Another question from Luke. One minute, so be quick. What are your opinions on linting your tests? I lint my tests. There's a linter plugin for testing library that you can take a look at. I think there's actually... Yeah. So I would recommend using that for your tests. It helps you write better tests. And yeah, I absolutely lint my tests. I also write my tests in typescript. It can be kind of annoying sometimes, but you kind of figure out clever ways to, especially like with mock functions and mock modules. But yeah, I recommend typing your tests and linting your tests. Wow. I was not expecting that, to be honest. Well since that was our last moment, my memory is slipping, but do you have a speaker room available? Are people able to continue the conversation? I'll be in the special chat for folks to chat with right after this. Awesome. Well, that's it for our little fireside chat, Kent. Thanks a lot for joining. I had a lot of fun and yeah, unfortunately I can't continue in the spatial chat because I'm needed here. But yeah, hope to see you again soon. Yeah, likewise. Thank you so much. Take care.
31 min
14 May, 2021

Check out more articles and videos

We constantly think of articles and videos that might spark Git people interest / skill us up or help building a stellar career