1 I'm very happy to be here this morning. I submitted this talk, seems like a long time ago now, when the RubyConf call for papers was open. After I submitted it, I thought, “Yeah...they’re not going take that.” Mostly because the abstract I wrote for the conference proposal was essentially a rant. But what I had forgotten, of course, was....
Rubyists do. With the possible exception of the major political parties this close to the election. But don’t worry, after Tuesday we’ll be back on top.
3 I was still surprised, though, a few weeks after they accepted the talk, when Evan Phoenix asked me if he could make this talk a plenary session. I did have to look up what "plenary session" meant. But I was super excited when he asked, because I love RubyConf. I love you guys!!! This is my fourth RubyConf, and every time I come I have a marvelous time, I meet awesome people, I learn all the things, and it’s great. So I am thrilled that I have tricked Evan into giving me this hour to rant in front of you. And I did trick him - because, despite the abstract, this talk contains absolutely no ranting whatsoever. Sorry Evan.
is the set of factors that affect whether or not we write good code. No one writes good code all the time. Ah, present company excepted, of course. And the factors that go into whether or not the code we write is good at any particular time - there’s a lot of them. And it’s complicated. Now - I like to write good code. I could say that it makes me happy to deliver business value, or it makes me happy to create a product, and those things are somewhat true. But really - I like to write good code just because good code makes me happy.
clearly, this middle developer in this picture has just gone into code she wrote two months ago and finished a very satisfying refactoring. So for me - code makes me happy when I can understand it two months after I wrote it. Code makes me happy when it’s easy to read, easy to use, and easy to modify. But I don’t always feel like that. I want to feel like that more. I want to write good code more of the time. And that’s what this talk is about. Maybe I should have retitled it,
And maybe I should have, actually, because that sounds even more rant-like. Anyway - that’s what we’re going to talk about. Now I started research for this talk by looking into how other people define “good code.” Personally, I like my defintion, but it is subjective. It’s qualitative. So I wanted to see if there was a quantitative way to evaluate the quality of code.
like forty years to get a more quantified definition of good code. And it turns out that good code is computable. In other words, it is possible for a computer to take two pieces of code, and grade them against each other, and say which one most developers will think is BETTER. It’s a very complex computation. The one thing that is useful for us out of it is that good code is defined by its pattern of dependencies. So. Let’s look briefly at what a dependency is and how we incur them.
Here some code that many of us have written. This is a Rails controller with a show action, and on line 3 we incur two dependencies. First, we depend on the Stuff model existing, and second, we depend on the Stuff model having a class method called “find.” Dependencies are important because the dependencies in a project dictate whether or not code is easy to change. Let me explain a little more what I mean by that.
conceptual diagram of a Thing that lot of other modules depend on, so A, B, C, X, Y, and Z all depend on Thing. This code is said to be tightly coupled to Thing, because if it changes, then potentially, all the modules that depend on it also have to change. And then all the modules that depend on those have to change, and that’s how we end up with a codebase where changing code breaks something far away that seems unrelated. Now I should point out that not all dependencies are equal.
Gard aquaduct in France. It was built by the Romans in the 1st century AD, which is almost 2000 years ago. There are many interesting features on this bridge, but one of them is that it was mostly built without mortar. The stones in those arches are not cemented together. Their weight, and the physics of arch structures, keep them in place. So - this bridge has incurred a dependency - on gravity. So if the way gravity works changes, this bridge will need some refactoring. But given how often gravity changes, the Romans probably made the right decision when they tightly coupled their bridge to the force of gravity.
coming back to our software example, depending on things that don’t change very often, such as pieces of your framework (hopefully), is safer then depending on things that change more often.
So the key to keep in mind here is that good code is defined by its pattern of dependencies. And that seems to work equally well whether it’s a computer or a person deciding whether the code is good. Right now, since we have a computable definition for good code - though I personally still like mine better - let’s back up and take a look at the factors that affect whether or not a particular person will write good code.
for sure, is their knowledge of object-oriented design principles. There have been several awesome talks about object design, actually, during RubyConf so far. Knowing about OO design means both knowing the actual principles, such as SOLID, the Law of Demeter, and so forth, as well as knowing when it is appropriate to use the principles. And certainly people do write good code without knowing all of the principles. But in general if you want to write good code in an object- oriented language like Ruby, these things are quite useful, and worth studying. At their core, all of these object-oriented design principles are advice on how to manage your dependencies so that you end up with good code. Let’s look at the Law of Demeter for a moment so you can see what I mean.
Law of Demeter says, in its very brief form, that if you have an object, it’s fine to call methods on it directly, but if you start chaining methods you should think about whether there’s a way to get the information you need more directly. And the reason is that the second line incurs a single dependency, but the third line incurs a whole chain of them, and if any of those things changes, including the intermediate ones that we’re not interested in, this code might break. So the Law of Demeter helps you avoid the far-reaching dependencies that characterize bad code. And if you analyze other OO design principles you’ll find that they’re pretty much all little patterns that help you incur the right kind of dependencies in your code. Now if all this talk of “design principles” is a little too abstract, or if you just want to brush up,
practices 3. Book cover copyright Addison Wesley 16 There’s a fantastic book that came out recently - Sandi Metz’s book Practical Object-Oriented Design in Ruby. This helped me see object design differently, and is very, very concrete. Highly recommend. So - knowledge of object-oriented deisgn is ONE factor that affects whether or not we write good code. Another big factor is the development practices that we follow. Over the last 15 or 20 years, we as an industry have decided that we write better code when we use agile development practices. Now that’s small-a agile - not any particular flavor of it, but in general - focusing on the user of the system, and keeping feedback loops as tight as possible. Test-driven development falls into this category, as well as pair programming, fine-grained user stories, all that stuff. You need to pick the ones that work for you. Now I suppose those two factors are fairly obvious, and easy to identify. They are the ones we focus on when we’re trying to improve the quality of our code. Because there’s always another screencast about object design, and there’s always another blog post about agile. But there’s a third factor that affects the quality of our code, that we typically we don’t think much about in this context, but it actually has the biggest impact overall on the quality of our code. And that factor is, of course, other people.
build anything interesting, you do, usually, have to work on a team. And in a team of this size - seven people, probably about average - just by the numbers, a single person will personally write less than half of the code in the project. Probably significantly less than half. Most of the code you encounter every day will have been written by someone else. So in some sense, if you want to work with good code and write good code, you and your team sink or swim together. Now the team itself is more than just the individuals that compose it. It’s also the communication patterns between those people. Now that’s an interesting thing. Because while you can’t affect the other individuals very much, one person can have a huge impact on how communication works within the group. And it’s been my gut instinct for a while now that it’s the communication patterns in the group, rather than the individuals within it, that determine whether or not they produce good code together. So I set out to see if science could help me confirm or deny. One of the difficulties in studying this type of thing is that the communication patterns within a team are often not as obvious as the arrows on this diagram.
of communication, such as manager-report, are easy to see, and to visualize. But informal communication, which science tells us is more powerful within the workplace, is very hard to see. I like this quote that Jim put up yesterday, so I’m going to put it up again.
are hidden because of their simplicity and familiarity. Wittgenstein 19 Formal communication is when you’re just talking about work and usually occurs at work. Informal communication includes things like chatting in the kitchen or on campfire, going out for lunch, and having drinks. Emails and conversations can include both types and may switch back and forth rapidly. Informal lines of commuication are ad-hoc and fluid. They’re constantly changing, and they are largely invisible to us. But if you can’t see it, you can’t measure it, you can’t diagram it,... how do you know if it’s good or bad? And how can you improve it? To answer that question, we have to go BACK IN TIME... to 1968.
paper called "How Do Committees Invent?" which is now - I”ll save you the arithmatic - 44 years old, and definitely still worth reading. In the paper, he outlined Conway's Law, which says:
design whose structure is a copy of the organization's communication structure. Text of paper: http://bit.ly/RrHWvp 22 It’s a bit of a mouthful, so I kind of like the way Eric Raymond restated it -
get a four-pass compiler. 23 Now over the years many people have interpreted this as kind of funny, and not quite serious, perhaps in the same vein as Atwood's Law - anyone know what Atwood’s Law is? "Any application that can be written in JavaScript, will eventually be written in JavaScript." And so many people have looked at Conway's Law and thought about it as the computer science equivalent of "people look like their dogs."
Specific Common functions Resulting software 25 Here's what their team structure looked like. They had a set of coordinators, and then they had a group from the army, and they had a group from the navy. The software that they produced looked like this. They had a group of common functions, and then they had an army-specific module, and they had a navy-specific module. They built software whose structure was identical to the communication patterns of the teams that produced it. When I read that, I thought, “huh. that’s interesting.” But I wasn’t completely convinced that it was still relevant, A, because people came up with some crazy stuff in the 60s, B, it’s over 40 years old and computing was a different thng back then, and C, I wasn’t entirely sure that something that applied to the military would necessarily apply to the kinds of smaller, less-formal organizations that we find ourselves in today. But then I thought back over the jobs I’ve had....and I realized that several of them provided me with perfect object lessons in Conway’s Law. I used work in a big company once...
looked kinda similar. The project I was on was new & innovative, and the team was actually amazing. We had a few people who had been around since the starting days of the company and had built some of its larger products, we had people who wrote books about programming. I mean - we even had the guy who wrote Freecell.
copy of this book. Now, at the time, Code Complete was pretty cutting-edge. It’s still pretty good, and I recommend it if you haven’t read it. Don't let this, ah, put you off too much. So it started well but as we got going we found ourselves really struggling.
piece of the project to work on, and once the pieces were “finished,” we started the “integration” phase and try to get our pieces to talk to each other. If you’d like to know how that integration phase went, you may refer again to the above diagram. Months of unbudgeted work. And trying to get these interfaces to line up required hacky changes to the internals of these components, which we did quickly because we weren’t supposed to be spending that time at all... And the whole thing felt terrible. I knew I wasn't doing good work. But I didn’t understand how we had gotten to that place. And for years after that experience, after I had left the company even, I wondered: how did our code get so bad? We were smart, well-intentioned, knowledgable developers, but looking at our output - you really wouldn't have known. How did we manage to build modules that completely failed to talk to each other? This was the first time I saw Conway’s Law in action, because we - the developers - didn’t talk to each other. The company believed...
individual office, with a door that closed. And they also weren’t big on meetings. Now this sounded great to me, before I joined - because it sounded like I'd have lots of time to focus. And I did - I spent a lot of time in that flow state, being super productive, and completely focused. Sadly, I was usually focused on the wrong things. Now, I thought they were the right things. Certainly last time I’d talked to anyone, they were. This culture feels like a bizarro other world - because it’s so different from what I do now - but I could go entire days without talking to any of my coworkers, except maybe awkwardly passing them in the hallway.
week, and those were the only times we really talked to each other face-to-face. We wrote a lot of email, but of course it was easy to misunderstand. So in the end,
modules that couldn’t talk to each other. We built a copy of our dysfunctional communication structure. Another project I was on, more recently but before I joined Pivotal, was another priceless example of Conway’s Law in action. This was a small company...
PHP Java R Python C Hadoop 33 ...whose hiring philosophy for developers was to get really smart people from top schools, and then let them do whatever they wanted. There were eight developers. Their application had a service-oriented architecture consisting of eight services… …written in eight different languages.
really, I wanted to show you that Conway’s Law, despite being 44 years old, is in effect today in organizations of all sizes. Even, probably, the organizations that you’re a part of. And it’s really obvious at the architectural level. Conway’s Law says that organizations draw technical boundaries where the team boundaries already exist. But what we’re interested in, is what happens inside one of those teams? Well as it turns out, there was a paper in 2007 from Carnegie Mellon, where researchers studied exactly that. They took Conway’s Law as a given, and looked at what happens inside one of these architectural boundaries.
code base • Good communication 35 Specifically, they were trying to figure out what characteristics of a team led that team to produce good code. They found a bunch of characteristics that mattered, as you might expect, including the technical qualifications of the individuals - makes sense, a team where the individuals have some experience with the technology will do better. Another significant factor - their previous experience with the code in question. That also makes sense, right? If you’ve seen the code before, you’re familiar with its patterns and can more reliably put things in the right place. But the strongest predictor of good code out of a team was good communication between the team members.
quality of the code you produce. 1 36 So here’s the first thing I want you to remember. Now this is fascinating. A, because it’s pretty awesome to have science back up my gut instinct, and B, because if good communication leads to good code and bad communication leads to bad code, then...that sounds a lot like the code reflecting the communication structure. Conway’s Law is fractal.
we see technical boundaries forming along the team boundaries. But it’s also reflected, in a more obscured way, inside a single codebase. It is definitely less obvious. At the macro level, you can look at the organization, and match it up with the software they produce. At the micro level, the structure it’s reflecting is less visible, so you can’t start there and look for mirrored structures in the code. But - you can go the other direction. You can start with the code, and let it tell you things about your communicaiton. Let’s look at a really simple example.
where the view helpers were cartoonishly enormous. Dozens of files, hundreds of methods in some of them. They were all included in all the controllers and all the views, and it was horrendous. Now I was going to just dive in and start factoring out the stuff that didn’t belong. But before I got started, I got curious. Why would anyone do this? So I poked around a little bit, and I discovered that they had an absentee architect. They called him the seagull architect, because he’d fly in, crap on everything, and fly off. One of the pearls of wisdom he had dropped was “thou shalt not put things in the lib directory.” Now...I appreciate the sentiment, actually, because for a long time I thought of Rails’ lib directory as every app’s junk drawer.
his word - lib was nice and tidy. But they’d just moved the junk drawer into the view helpers. I learned two things: First: that I needed a way to make the seagull architect less disruptive. Second: we had a rule of thumb that was generally good, but needed more context and explanation as the developers who were using it gained more experience. I let the bad code point me towards the communication issues that had caused it. Let’s look at another, more concrete, example.
end end 40 Here's a piece of code from a completely different project, and I would characterize it as slightly smelly. We have a Rails model that is including the URL helpers so that it can put its URL into its as_json method. Now this is *not* how you actually include the URL helpers, but it doesn’t matter because you shouldn't be doing it anyway. Now when I came across this code, at first glance, it was a simple object design problem. This code violates the single responsibility principle, which means, it does more than one thing. Models are for persistence-related code - that's their main purpose. But this stuff has nothing to do with persistence. So, what that means is that there's probably another object trying to get out here, maybe a presenter of some sort, like this.
as_json {url: stuff_url(id)} end end 41 That's a pretty straightforward refactoring, uncontroversial these days. The objects would be better factored, and in general we’d be improving the quality of the code. But before we actually do the refactoring, it's worth asking - is there anything else we can learn from here? Let’s go back and look at the code.
end end 42 Well, this code came from one of my former teams, and I learned two things actually from looking at it. The first thing was that the presenter pattern hadn't made it far enough out of my head. In this codebase I already had introduced presenters for other models, but it was a big enough team that obviously not everyone had seen them. I hadn’t communicated that decision effectively. The second thing I learned was that some of the developers on my project were ready to hear the more nuanced version of “Skinny Controllers, Fat Models.” Let me explain a little bit what I mean by that. The developers who wrote this code had been programmers for a long time, but had only been doing Ruby and Rails for a few months. One of the things we talked about was “don’t put model-related logic in the controllers." Right? Skinny controllers, fat models. And they had done that, which is awesome. Because they totally could have put this code in the controller...
stuff_url(@stuff) }) end end 43 like this. But they had listened, so they moved it to the model. So the second thing I learned was that instead of giving them the short version of this rule of thumb...
code related to their main purpose: persistance. Move other stuff out to non- ActiveRecord classes. 44 I should be telling them the long version. ... Yeah. I guess that last one doesn’t roll off the tongue quite as easily as the first one.
more nuance 2.1. Developers actually listen to me o.0 45 So looking at this "software design problem" and really digging under it showed me two specific problems in my team's communication structure that I didn't know we had - the hidden silo, and the new threshold. And as a corollary to that, I also discovered that other developers actually listen to me! That might be the most amazing thing on here. We used Conway’s Law in reverse, to let this code smell shed light on our otherwise-mostly-invisible communication structure. And what we got back was really valuable. If I tell everyone about presenters, and I revisit where you put model logic, then I address the root of those code problems. Whereas, if I had just refactored the code, I would have just fixed the symptoms. Over time, fixing the code AND fixing the communication difficulties that the code helps you identify pays way more dividends than just fixing the code. Or in short -
something. 2 46 Every piece of bad code is trying to tell you somethng. When you figure out what it is - fix that. Because fixing those communication problems increases the quality of the code that all of you write.
last piece of this puzzle together, and talk about how you fix communication problems. I’ve been at Pivotal Labs for almost three years. If you aren’t familiar with our process, we do all pair programming.
code into an act of communication. It means I sit with someone, all day, and we talk about the code we’re writing. It’s actually really fun. So for almost three years, minus vacation and holidays, I’ve been pairing eight hours a day, five days a week. And in that time I have become the carpenter who has any tool you want, as long as it’s a hammer.
And a lot of these communication problems sure look like nails. (deep breath) Now I did SAY that I wasn’t going to rant. But why is it... why is it that some developers are so rabidly opposed to pair programming? I mean, leaving aside whether or not it’s effective, since there are have been loads of studies confirming that. I mean, why is it that some developers just don’t want to do it? And what I think is that they don’t want to do it because they don’t know how to talk to people and they think it will be awkward. And I think that because I thought that before I started doing it. But, I have some news for you: you are a human being. And learning how to talk to other human beings is a skill that you should learn. Easily talking to people is not a talent that you either have or do not have. It is not. It is a skill that you can learn, and get better at. And I think you’ll find that spending your day talking to another developer is significantly easier than spending that time talking to a normal person. But - the practice still counts! About 3 months after I started at Pivotal, I went to an alumni event for my university where there were normal people. And I was shocked to discover that I could make small talk with them, which I’d never really been able to do before. And not only that - I was pretty good at it. I asked them questions about themselves, kept the conversation going...I went home that night and I sat down and I said, what just happened? Am I becoming an extrovert? Now I won’t tell you it’s not difficult. It is exhausting at first to spend all your time talking. The first few months I was at Pivotal, I was so exhausted that I really don’t have clear memories of that time. The only other time in my life I’ve been like that is after I had a baby. So, it is hard. But it is worth doing. No matter how awkward you are now, you can learn to talk to people. Because that skill, that ability to connect with people, is what makes us human. And that’s all I have to say about that. (breath) Ok, We’re back on script. So!
I’m going to tell you is that if you are really serious about increasing communication among your team, pair programming - even ad hoc, even part time, even temporarily - is a very efficient way to get you all to talk to each other more. There are, of course, other ways to get people talking so let’s go through them. The first one is create opportunities for informal communication.
Science tells us that informal and accidental communication are critical parts of your team’s structure. People need to have opportunities to have conversations with people that aren’t totally about code. There are lots of ways to create this space. One of the most effective is to sit together in the same place. You overhear each other talking about what you're working on, and can contribute if there's something you know about. This works for distributed teams too, via campfire room, irc channel, google hangouts….any number of options. Another way is to create a space in your office for people to relax. Sometimes it's as easy as bringing in a couch to the office, or a buying a fancy coffee machine that people have to show each other how to use. Look for ways to bring up the ambient communication in your work space. Second:
you need to talk to someone. Face-to-face is still way higher-bandwidth than any other form of communication we have. You get lots of meta-information, along with the actual information - meaning the body language, the facial expressions - which in this picture I love, because they’re so goofy - the voice intonation, and so on. If face-to-face isn't an option, then in descending order, of usefulness, we have facetime/skype/google hangouts, phone, IM, tracker comments, email, ...carrier pigeon. I don’t really recommend those last three if you can avoid them. And finally,
communication • Use high-bandwidth communication • Teach/learn/practice 54 Now one thing I will say about these is that you will have the most impact if you’re a team lead, but they’re still effective if you’re not a lead or a manager. I don't know how many of you ever read Dear Abby - as a kid, we had these things called newspapers that showed up at our house every day.
read in the newspaper every day, long after I decided that funnies were a waste of time. Dear Abby is an advice column, and people wrote in, usually asking advice about problems in their marriage, and Abby's advice to those folks was remarkably consistent: go to couples' counseling. If your partner won't go with you, go by yourself. Now that sounds weird, to go couples' counseling alone. But Abby knew that learning communication techniques and taking them back home often improved the relationship, even though only one person was going to counseling. And these communication techniques are very similar - you can do them yourself, and both you and the rest of the team benefit.
want you to remember, is hack your team! You have more power than you realize. A single person has remarkable power over the structure of communication within a team.
illuminate problems • Fix them! http://www.flickr.com/photos/uggboy/4066337480 57 So go forth! Do these things! If you do, I think you’ll find you’ll spend more of your time writing good code. And I hope that makes you as happy as it makes me. Thank you very much.
sarahmei
yodakeisuke
mkdev10
nwiizo
masakiokuda
databricksjapan
tenten0727
kameitomohiro
dictoss
miholovesq
takabow
yuitosato
naitosatoshi
skipperchong
bryan
chriscoyier
yeseniaperezcruz
addyosmani
tammyeverts
morganepeng
eitanlees
sonatard
bluesmoon
chrisshort
![class StuffController def show @stuff = Stuff.find(params[:id] end end 8](https://files.speakerdeck.com/presentations/50960d822fd11800020115e8/slide_7.jpg){kind=link}
![Thank you! Sarah Mei Pivotal Labs @sarahmei [email protected] http://www.flickr.com/photos/megaanliddle/4568674938 58](https://files.speakerdeck.com/presentations/50960d822fd11800020115e8/slide_57.jpg){kind=link}