This course is archived!


When Testing Makes No Sense

Share this awesome video!



SymfonyCon 2018 Presentation by Miro Svrtan

If you look at the stage of any conference in the PHP world, people are preaching testing, testing, testing ... If you on the other hand look at the community, the percentage of people writing tests is really low.

As a person who went from 'How can I ask for more time/money/resources for testing?' through 'ask for forgiveness instead of permission', to person who writes tests a lot, I still believe testing doesn't make sense. No, it doesn't make sense for all and everyone, often enough it makes no sense for me too.

This talk will explore that fuzzy line when you have to shift your mind from one side to the other: in both directions.

Everybody's got their caffeine? I know it's the last session of the first day, so, everybody needs it, right? No? Somebody can live without coffee. I need some help with how to do that.

The Back Story

So, first off, I see a lot of young faces in the crowd, so, do you know what this thing is? That's how old when I started developing professionally. So, almost 20 years now. So, I started off doing small websites because at that time people just wanted to have some internet presence. And far away from any kind of rocket science, even by those days, and for most of you today, it would be like, yeah, really what's the problem there? But that was the time that we didn't have cool things like frameworks, templating engines, database abstraction layers or ORMs. We had to handle all of that ourselves, but this was actually the fun part of the job. The problems started when you had to make it work for things like Netscape or Internet explorer 5.5. And if any of you are doing front-end these days and think you're screwed, trust me.

And, I got so uninterested in development that I actually went back to university doing some freelance on the side, earning some money. But about 10 years ago I found a team, I found a company, I found a really great job. It was a small product. It started off, let's say, like a dog house - really, really small. As we kept working on it, it kept growing in both features, customers, money. So, I was good with, let's say, carpentry.

So, in few months we had a shed, which meant more and more people could fit in. They wanted more and more, more money was coming in, more customers, more content, more everything. So it's grew again. You understand that you can't use wood to build skyscrapers, right? You have to change your base. So yes, I had to learn how to use concrete to get a big, better stuff done.

But what actually, my code looked like, something like this. You know, and to be honest, probably, in this case, this was done by somebody and then by somebody else. And for most of it was us, us. And, okay, it was ugly, but it worked. But what our customers wanted was bigger and bigger and nicer and not now concrete. It's aluminum, it's glass, it's shiny.

But as we kept growing, we kept failing. Because, hey, I didn't have the experience needed for that project. I used to work in small websites, small apps, not 100 million user visits a month. And what customers actually paid and what I thought was, oh, this is what we're doing. But in honesty, most of the developers don't understand that businesses want a nice feature, another nice feature, they really really want something to just glue them together. By design, because businesses don't know always what they want.

So, this project has took about five years of my life. And the next project that I came on, yeah, it's going to grow like that, right? It's going to be that small product is going to be the biggest thing ever. Unfortunately, what I was suffering from was Stockholm Syndrome. Because in reality, not every project or product becomes big.

So, let's say that you want to buy a dog house, like a real dog house. You have a dog and you want to buy a dog house. How would you feel if I came to you and told you that the price for this nice small dog house includes something like this? Because it's going to be a skyscraper so I want to do it before and avoid all of the problems. Right? Found yourself any time in that situation? So, Donald Knuth said: Premature optimization is the root of all evil. And, that's why I want to talk to you about when testing makes no sense.

My name is Miro Svrtan. You can find me on Twitter using this handle. So why am I talking about buildings and testing? What's the connection? So, who here is testing? It's an easy question, right? Okay. Who here isn't testing? Again, so, this was a partially loaded question. Sorry for that. But actually all of us test. You write some code, you open your browser, you check that it works - it's also a testing. So, I hope nobody just writes the code and pushes it to production. To be honest, I did it few times. But not as a normal way of working.

So, let's say that we have a small website. No Mars rover thing, just a small website. And we have a bug. How much time do you think after somebody reports it will take us to fix it? Five minutes? An hour? Half a day? How much time does it take you to deploy? Probably 5 to 10, maybe 15 minutes. Right? So, why would we test and invest in testing when we can fix something quickly and deploy it everywhere?

But the problem was that about 7-8 years ago, a lot of web developers became phone developers. Been there, done that. A few of my colleagues went that way, but they didn't have the mindset of testing. Which is pretty nice for Google store because it takes about 3 days after you fix it for people to get it. But what if it's an iPhone? iOS app? 6-7 years ago it would take from 2 weeks to 3 weeks, sometimes even a month to fix something.

What if you're doing desktop applications? You have a bug and now all of your customers have to be somehow contacted to download the new version because, okay, maybe Mac users do have App store, but Windows users not so much. How hard does it then to fix a bug? It might not be a critical bug, but it might be something that bothers your customers, your users.

Hey, we have hardware. But that's at least easy, right? Because, this is a router, so it's connected to internet, right? So it's easy to fix it. If we have a bug, we can just ship it to all the routers. What if the bug is in the piece of code getting the update? Are you going to send all of your staff to the whole continent, to the whole world to fix people's routers?

When should we write tests?

What if it's... what if it's a washing machine? Hopefully, it's not connected to internet, but you find out that you shipped some of it and you can't wash cotton on 60 degrees. So, you now have to call all of your mechanics, all of your service departments, to contact the customers to flash the washing machine. So, when should we write tests? When it's a high cost to fix? Because, maybe, sometimes it's cheaper to test than to try to fix it once it happens. Maybe when the cost is larger.

Let's go back to the website. So, should we ever test websites? We can fix it easily, right? Yep. Right? Anybody? Nobody? Everything test? But banks do use websites as well. I mean I hope you don't experience a bug where you send the money instead of receiving it. The other way around, probably, you'd like it. Right?

So, let's maybe look at it as when it brings value. Because value is not just the money, it's how much your consumers love the product, how much your clients are happy with your work. Maybe for you, if you're a development company, maybe because of, maybe, it would bring value in the sense of people not leaving the company often because they would be more happy? There's a lot of cost when a disgruntled employee decides to leave because then you have to get somebody in. It might take about 6 months to get the new person in. So, let's define the value as benefit minus the cost.

So, what I'm trying to say that if, let's say, you have a tomato farm that produces a lot of tomatoes, you probably shouldn't use the same tactic - you might use on your own tomato garden. You maybe like your tomatoes a lot, but would you, um, would you create a whole lab? Would you test it every day? Probably for tomato farm you would because if one goes down - you might lose everything. Everything, if you get the disease. In case of your tomatoes in the garden, you can just go to a farmer's market or a supermarket and get some others. Okay?

So, value is benefit minus the cost. Of course, I do understand that some people might invest a lot of money into their tomato gardens because it's a hobby and we tend to spend a lot of money on our hobbies, but let's look at it as a business. The benefit minus the cost. So there is some, before I go further, there are lots of different things about testing. So what I'm going to try to focus on is writing test, automating the tests, the ones that help you ensure that the app does what was expected? Okay?

So, who here writes tests in that way? Okay. Who tests everything? Okay. 6 to 7 hands. So, actually, can you raise your hands up? Who tests everything? How much security testing do we do? Performance testing? Visual regression testing? That all is testing, right?

So, why I decided to do this talk was people came to a stage like this and said we test everything. And a lot of people understood that, yeah, to be a good developer you have to test everything. And they started testing, but they started testing the wrong things. I've done it that way. And, 6 months ago I did this talk in Amsterdam and I spoke to another speaker afterwards. He wasn't at my talk, but he was interested in what I was talking about and he told me: Hey, yeah, well no, you shouldn't be talking about when not to test because we test everything. I said: Okay, so performance? Oh, no. Security? Some. Visual regression? Nah, we actually TDD everything. Wait, you're using Symfony and you TDD everything? Yes. I was really interested because I found TDD to be very wrong for Symfony, the Symfony part, like using the Forms, using the Doctrine and stuff like that. And I was wondering, okay, how do you do that? Oh, we don't. So, wait, you just told me you test everything and you do it TDD way, but you don't do the UI, the infrastructure and stuff like that? Nope, that's the UI. So if I wasn't really interested in that part, like how do you do that? I would think, yeah, I'm the stupid one because I don't test everything. Please, don't be.

Fallacy #1: It makes my project more expensive

There are a lot of fallacies with testing. 6 or 7 years ago, when I wanted to start writing tests, one of the biggest problems I had, I was a team lead at the time, was... So similar feature was about 100 hours. Now I need to set an estimate of about 150 hours because, well, writing the code, writing the tests, right? So it's gonna take more money from the clients. They're not going to be happy. What I didn't think of - do you charge for manual testing? So, do we say: Hey, it's gonna take me 7 hours to write the code and X amount of hours to run it in the browser as I click save on the code every time. We think of it as the same thing, it's part of development.

And, if you have to explain to anybody, try to explain to them the bugs cost too. So most of the companies should have, should solve the bugs for free. Right? I mean I know of companies whose whole business plan is to charge for bugs, for bug fixing. But you might notice that that's a pretty bad business model because somehow they want to produce more bugs because that means more work for them. Right?

Anybody here has a car? A new car maybe? Or somebody's buying a new car? Recently? So, let's say you want to buy this car. You go to, I think it's an Opel dealership, and you say: Hey, I want this one. How much does it cost? Oh, 20,000 euros, let's say, I have no idea, please, don't get me on the numbers. Okay. So, it's tested, right? Yeah. So can I get the version that's not? It's going to be cheaper, right? Because why would I pay for testing?

So products don't come out without those things. I mean, when you buy food, you expect it to be tested, right? You don't want to be poisoned. Try to ask for forgiveness, not permission. Just try to do it. Try to do it in a small way. Start building up your tests. Every new feature, add more tests. You don't have to do everything the first time.

Fallacy #2: It takes a lot of time

Another fallacy is that it takes a lot of time. That might be true for a first timer. Because there is a cost of learning. I mean, I guess, most of you that had a car that are driving had to go to school for driving, right? It took like 20 to 40 hours. Right? So, you don't take that into the cost of every time you go to the shop. Right? It's learning. Yes when your first project, it's going to cost you, your company, your product team, whatever. But it's learning. You wish don't learn.

Fallacy #3: It takes extra time

It takes extra time writing the code, writing the test. Right? So, if I rock, if I write 10 lines of code, I need to write some lines of tests as well. So, that takes time, right? Manual testing takes time as well because you have to Alt+Tab or Ctrl+Tab something, check your browser. So, if you write your tests as you're writing your code, you don't have to do that manual check. You can just run the test and it will tell you that, what you expected, works.

Fallacy #4: Ship PoC to production

So, for some years I thought that if I implement something that's ready for production, let's say, you want to implement a new social login or a new payment gateway and you get one PHP file where you see that it works, then that's it, right?

Phase 1: Exploration

But actual feature lifecycle is the part where you explore. You look how that system works. It could be a library, it could be a new database, it can be a third party integration, it could be whatever. That is the part where you don't test. You should explore, you should enjoy, you should understand what you're doing.

Phase 2: Modelling/architecture

Then you start modeling. Again, we're not testing, you're using tests but you're not testing - you're modeling. And this is a very confusing thing, when I try to explain it to people. If I'm writing tests, right? How is that not testing? So, if you are in a car, it doesn't mean that you are the driver. Okay? Uber and Taxi doesn't work that way. Okay? So, yes you could be using tests just like you use a car, but that doesn't mean you're driving it.

Phase 3: Development

The third part is development and that's where testing needs to be done. A lot of it. Because you're not sure about all of the edge cases that could happen. Here tests have a great value.

Phase 4: Production

And usually, when I go to production, I delete some of the tests. Because when I was writing them in the previous phase, I wasn't sure which of the tests will have value later and I don't want my tests to run for hours just because during the development phase I wasn't sure if that can break or not. I'm not saying delete all of the tests, but be... feel free to delete some of those that you do not need.

Fallacy #5: 100% code coverage

100% code coverage is something that I had a lot of problems with. And I hear from the others. If I don't check the code coverage, how would I know if something is tested or not? So you want to go and have 100 percent because in that case you know your app is tested. So, what usually people start is, well, let's get the code coverage as high as we can, as cheap as we can, which means testing getters and setters. It can be fun for like one or two entities, five-ish is Ballmer's peak. Oh, need of Ballmer's peak. About 10 - I'm happy if everybody doesn't quit their company.

So, should we test something like this? I really don't see the value. When I moved to using commands and events, command and handler events, command handler patterns, I found that I spent a lot of time on testing those setters and getters, once again. What now I usually do is if I do use code coverage, I just put all of the commands and event namespaces into ignore. And I just don't care. I want to achieve the higher percentage, but I don't want to know about those DTO classes.

Fallacy #6: We have "tests"

We have tests. I've heard some people complaining a few years ago about they wanted to participate in open source project. And they implemented the changes because implementing those changes was rather easy, but making all of the tests work was the hard part. You don't want you and your fellow developers to be in the position that, to change the code, is so hard because of the tests. Because people in that case will not test. I mean if you have a car on the bottom of a lake, what's the value of it?

Fallacy #7: Writing tests later is OK

I mean, and that most often comes from: Let's write tests later, which is most often coming from management. And as a disgruntled developer, I might come, I might build the whole feature. I know everything works and before merging it, my manager comes and says: where are the tests? We're not merging that without tests? No. Everything works. I don't care. Everything works. No. You have to write those tests.

What do you think? What kind of tests am I going to write? I'm just going to look. I'm going to split my screen. I'm just going to look at the code on the left side and replicate it on the right side. Because that's going to get shit done. So, yea, I'm just gonna add a simple case for this, maybe.

Fallacy #8: TDD has something to do with testing

Who here knows what TDD is? Anybody practicing it? Okay. If I say, if I say test driven development - is that TDD? No, sorry, another loaded question. Test driven design. Not development, design. You're supposed to use TDD to build your architecture, to model it. You can write your tests first and then your code - that's not TDD. TDD is that small red, green, refactor, red, green, refactor, red, green, refactor.

Anybody's remembering this tweet? Anybody's remembering the shit storm coming after it? Where everybody came with: Haha, you see, testing makes no sense. You shouldn't test anytime. Why were you forcing me to test? People don't understand the TDD and testing are completely different things.

Uh, do you all know who DHH is? So, he's an author of Ruby on Rails framework. He works for a small company called Basecamp, used to be called 37 signals and they have a few small products, really, really small products, really cool products, but really, really small products, that they have been building for about 10 or 15 years. What kind of modeling would they need when they know what they're doing? There's no modeling in it. They just need to test that what they want works. It's not really modeling. And, just to try to explain to you, that's one of the rare companies that would come to their clients, when they would ask for features, they told them: No. We have our product, we love it, as simple as it is. If you want more features - use Jira. Use this, use that. Thank you. Here's the door. So, don't forget the environment that this comes from. They don't build that many features.

Fallacy #9: TDD all the way

If you ever looked into TDD, one of the things that even the people who invented it said, you have to do everything the TDD way. Only in the last few years they started saying: Well, we were wrong. Because trying to build everything TDD way takes a lot of time. And if you're yet another payment gateway, yet another CMS, yet another E-commerce and you know how everything has to look, what part of modeling do you have to do? It's yet another house. Maybe has another color, but that's about it.

I'm going to guess that some of you changed the company a few times. Have you ever come... have you ever changed it to company that has good tests? Whoa. Really? Wow. Three hands. Wow. So, my previous employer had such a great coverage that I started shipping the next day. I came on Monday and started working on Tuesday. Because it was easy to onboard me because I could have gone through the tests to understand what and why are things happening and even if I screw it up, because it's my second day, people, the CI says: Hey, this doesn't work anymore. It's much easier to add new features because you know that you didn't break older ones. You know that the older ones still work.

So, what do we had? This, the previous exam? We know this suck. We have to add a feature now. The feature is that if you're not from, um, from that European Union country, I'm using Portugal since we're here, you shouldn't charge a VAT if it's a company. So, maybe this is the point where you should test. Switching from rather simple if, to another.

What if you have to refactor something? So you have all of your tests. They run, you just change the code and all of your tests say: Hey, it still works. Still works. It's okay, everything's good.

So let's make this bit more readable to avoid nested ifs. So this is what happened. This is what I have. Can anybody spot the bug? Where? Here? Nope. You always have to charge VAT for that country if you're selling something in your country. Any others?

So, I just wasted two minutes of your life trying to understand if this was okay or not. So if we had tests for this, you don't have to look at this. You know it works. It's easier to rewrite things. And running tests means that we can do some fun while they're running. I mean at least for us in PHP world because we don't have compile times, which I'm something that I'm actually missing. But that's a long story.

Cleaner code. Trust me, it's really hard to test shitty code. If you write your tests as you're writing your code, you'll will write cleaner code. Because for a 15 line code, 15 lines of code, you might have 300 lines of test. And then you add an if, which means 600.

But maybe if you're building a small app, something that you have built often and that you understand. I don't think that testing brings any value. Let's say you are bringing, you're building a one-off app. Again, you can easily manually test it. That's okay. You know it works, but what happens a lot is, that simple app becomes a bigger one and a bigger one and a bigger one. When that one-off app for one concert in Lisbon becomes an app for all the concerts of that group in the world or all the concerts in the world for every group, you can't use the same thinking. When you're exploring something, if you're trying to learn a new language, a new library, a new framework, a new something - don't waste time testing. Enjoy it, learn how to use it.

Last but not least: job security. If you're a key person in your company and you have no tests, you cannot be fired. You can ask for whatever you want - you'll get it. I want a new car, I want the best Mac, I want 5 month vacation - you'll get it. What people keep forgetting is that at some point they might look for another job because they're not happy where they are or they have to move to another city or they get the shitty manager because their previous one decided to go somewhere else. You will not find a senior position that says that doesn't have listed testing experience, testing knowledge. I didn't trust people when they started telling me that 6-7 years ago, but for last 3 years, no senior position ad that I saw, came without testing experience. So, it's gonna be really, really hard to move to find a new job and it's just getting worse and worse and worse. So if you're like, my grandma had a few chickens. No fancy lab to test them. Please, don't think that you can use the same principles to have a chicken farm. It's different mindset. Thank you.