A fascinating conversation on The Elm Programming Language with Philip Poots.
Philip Poots @pootsbook | GitHub
CHARLES: Hello, everybody. Welcome to The Frontside Podcast Episode 56. I am Charles Lowell, a developer here at The Frontside. With me is Jeffrey Cherewaty, also a developer here at The Frontside.
CHARLES: We're going to be talking today with Philip Poots, who is a fantastic individual, who I have known over the Twitters, over the e-mails, interacted with at conferences, seen him speak on at least one occasion and today we're actually going to be talking about the thing that I saw him speak at Wicked Good Ember last June. It was actually one of my favorite talks from that conference. It was on Elm for Ember developers. Thank you very much for being on the show, Philip. Why don't you tell us a little bit about what Elm is and how you came to find out about it and really kind of dive deeply into it?
PHILIP: Yeah, sure. First of all, pleasure to be on the show. The Frontside is one of my favorite podcast, if not my favorite, given a cross-section of the Dadcast, the love of programming and balancing that with the business of programming. That’s right, I'm an independent developer. I started off with Rails then got into the Ember quite early on. Last year, I think around January, that's when I started really investigating Elm in detail.
It's actually a funny story how I came about because I was at Ember Amsterdam and it was a night where we had three members of the core team: we had Erik Bryn, he gave a talk, Alex Matchneer, gave a talk and Igor also came over because he's based in Europe. Alex always loves to investigate new things and one of the things he was getting into was Observables. I'd never heard about Observables at all so after the talk, I kind of pulled him aside and I asked him some very stupid questions. He was gracious enough to bear with me and to dive a little deeper into this stuff.
Alex is kind of a quiet member of the core team, unless he's got his drum sticks but he's the guy that rewrote the [inaudible]. That was no mean feat because I got into Ember just before that moment and the way he managed to make that incredibly easy was fantastic so I kind of had an extra ear open to what he had to say. I went on this Observables talk. You know, you start off with React that was the framework that was using Observables the most. That brought me to Cycle.js --
CHARLES: Cycle.js? I haven't heard of that. Is Cycle.js a framework built on top of Observables?
PHILIP: It is. There’s a guy called André Staltz or André Medeiros but he uses Staltz as his name. It’s largely based off the same principles as React. Cycle, basically one of his inspirations or at least one of the things which cycle is most like was the Elm architecture. He calls it Model-View-Intent. We have Model-View-Controller and Elm was model update view but essentially, the same principles.
You know, I'm the kind of guy that likes to get stuck in, to go deep and where I started with Observables then I ended up at Elm. I started playing around with that, I started looking into it and I loved what I saw. The thing above all that really attracted me to it was the pure simplicity of what was going on. It was almost like they boiled down the UI paradigm to its essence and removed all the extra cruft and you just saw what you meant what you wanted and it gave you these, what I thought at the time, composable way to put things together.
CHARLES: Can we unpack that just a little bit? I really love that idea of it boiled down to the essence of UI. I assume there are certain coordinating mechanisms that Elm employs. It’s interesting to hear you say that Cycle.js has used Elm as an inspiration. I also understand that Redux is inspired also by the Elm architecture. I'm very curious, what are those kind of essential mechanics that drew your attention?
PHILIP: I think you can look at it from two points of view. The first is, which I didn't actually learn until later but the first is essentially boiling down functional programming. You’re decoupled, you're using functions and not only functional programming and there's a lot of arguments over this term but functional reactive programming. The idea of functional programming is stateless. Therefore, time is kind of the beast that you have to deal with.
CHARLES: There are a lot of blog posts trying to explain what exactly is the view, is the template the view? Is the controller the view? What's the difference between a view and a controller?
PHILIP: The way I think about frameworks is they give you buckets to put your code. The buckets are kind of all connected together. I'm thinking at a really simplistic level. I need to write this feature, I need this bit of code, I need this bit of logic, where do I stick it?
CHARLES: Yeah, I love that. I feel like it is very much as 'data down, actions up' kind of boiled down to the essence. It’s almost better at that paradigm than Ember is in itself, with having your view as just a function. Your state transition or your update is just a function. Then your model is just unadorned data. That’s all it is.
PHILIP: The type system as well in Elm, also made it really straightforward. My model is a record. That’s it. My view is HTML in the beginning but then it moves to HTML, which contains messages. Essentially, I've got HTML and in the HTML, there are events or actions and those actions will send messages and really straight forward. Then the update function, I take in the current model, I take in the message and I decide via pattern matching what I want to do. There are a few extra bits and pieces around there but that's the essence of it.
CHARLES: Now, when you say messages, I'm thinking this is a way of declaring what actions you will take when certain HTML elements, like events happen, like you declaratively mapping an event to the dispatch of an action.
PHILIP: That's correct, with any extra data so a message is itself. Like the actions hash in Ember essentially and along with the parameters for that action, which would passing through the HTML.
CHARLES: Some might argue that if something is simple to get something working, I can have a pure function that's a view, I can have a simple data structure which is my model. I can have a pure function which is my update or my state transition, how I change my data or affect changes to the model. Some might say, "That's very simple," but simple is great for simple cases. But then there's the question of scalability. As my application becomes more complex and has the interactions become more complex, does that simple paradigm actually scale? What has been your experience there?
PHILIP: Straightforward answer -- it's a learning curve to scale. Why? Because it's so new and also because the things that you would have reached for in the past aren't available to you. When I think of Ember scaling, the scale is built into the framework. You need another component, you just add another component. You need another model, you add another model. There’s a clear story or there's a clear way that you deliver a new feature.
I think that's a fantastic aspect of Ember that you also say no. I can remember being at Wicked Good Ember and just realizing how many of the people that were speaking and how many of the people that I was talking to work for bigger companies -- Heroku Dashboard, you had LinkedIn and the fact that Ember scales across a large team size, it's a real testament to Ember, which is slightly different for the reason why I got into Ember. Also you know I had a few issues just on a side project with a friend, the pace of Elm's change meant that when you only have a limited time to devote to the project, then you don't want to be spending that time for going out where you should be in terms of the upgrade process, etcetera.
That’s a known path and I think that's a really clear advantage of Ember over Elm. The path in Elm is not as well-fleshed out and there's a bit of tension between Elm as a single page application and then Elm as an application that you stick on every page of your server generated app, for example. The main people who use Elm NoRedInk could employ Evan Czaplicki, the creator of Elm, there up at the minute is a Rails up. I think in Elm up for every page, I believe that certainly was the case, whether that's [inaudible], I don't know.
It’s not an area where it's like here's the path, go for it because the scalability of Elm, everyone came into Elm thinking, "I know react. I know Ember, the components system, 'data down, actions up'," and in React, you're really encouraged to make a component for the smallest thing on the page as a component. Then you have container components, you make your way out of the onion skin and hope you don't cry on the way.
PHILIP: But the thing about Elm was everyone jumped in and tried to do it this way. I certainly got the impression when I was beginning that the Elm architecture was infinitely Nestable Russian dolls. It is in a way but the difficulty is then passing into component communication between parent and child and people had been figuring it on a weird ways with signals to do it but then became obsolete.
The main encouragement is basically go so far with a single component that you can and then once you had problems, try not to create new components or new bits of UI but to extract the bits you need into modules. This is actually one of the things that really attracted me to Elm is that you're encouraged to lean on your programming skill set, rather than learning a whole new frameworks way of doing things so that the things that work in functional programming will work in Elm. But that's also a down side because all of a sudden, you have to exercise your programming chops. Let’s be honest, a lot of the stuff we're building we're like gluing things together. We’re not thinking up new architectures or ways of doing things. That’s definitely a learning curve and that's definitely a struggle or something that I find difficult.
CHARLES: Yeah, I feel like that touches a lot on the messages you get from the FP community. I know certainly in the interactions I've had with the Clojure community, they're very big on that and it's like, "Let's have very powerful primitives." They have that term 'decomplecting', like let's get at the core. It’s like understand that essence so if we can compose and we can enable composition of these low level functions and allow you to compose the data, then you don't have to worry so much about everything else. There will be a way. I think the counterpoint to that is that you end up with a lot of different ways because there are a lot of different ways that you can compose a very small set of primitives.
PHILIP: Yes, that's right. But I think one of the advantages of Elm over maybe other FP languages and this is where the similarity with Ember comes in as well. It tries very much to cement simple convention but not only by convention but actually baked into the language so people that are coming from more powerful functional programming languages often come to Elm and think, "Why can't I do that? Why do I have to write all this boilerplate code," and the reason is because then it's not going to be simple enough anymore for people to use it. Also the goal of Elm, which is this long term maintainability of large code bases, you kind of shoot yourself in the foot a bit just like you said, Charles.
CHARLES: Yeah, I think that's actually a great point. It’s actually one of the things that was most memorable, I think about your talk at Wicked Good was -- just a quick anecdote. I remember in 2007, when iPhone came out, I had an iPhone and my father in law came over from Finland and this is a guy who was a Nokia partisan. I mean, part of Finnish pride was everybody own --
JEFFREY: National phone of Finland.
CHARLES: Yeah, exactly. Everybody owned the Nokia mobile phone. He came over, He visited me and I had an iPhone and he was like, "Can I see it?" He took it, he had the swipe to unlock and he swiped it and it unlocked and all the app icons just kind of came right into the screen and he was like, "I want one."
CHARLES: That was all it took. I've never seen a sales process so utterly complete and so rapid in its realization. For me, I think that moment when I saw you talk was when you made a mistake like where you were trying to match against an improper attribute on the model inside the update function. The first thing that happened was that Elm caught it before you could even compile your program and the error was just beautiful. It put its finger right on and it's like you need to fix this right here so there's much tied up into that because I feel like it addresses a lot of the learning curve problems that we have in Ember.
PHILIP: I don't think that's Ember specific --
CHARLES: No, it's not, it's --
CHARLES: Yeah, it was fantastic where when you fail, it picks you up, dust you off and sets you right back on the right track. I think one, that's just a freaking awesome feature and I think also two, the thing that struck me when I saw that was like, "Wow, this community has a different focus than other FP communities that I've come into contact with because I have encountered that exact same error message in Haskell and it left me puzzling and wondering what to do. It's like, "No instance for type class blah-blah-blah for class blah." Then if you're an experienced Haskeller, it does point right to the problem in the same way that like you've learn the parse Perl stack traces. You know, you see a Perl stack trace and you understand it. But they could have gone that way with Elm but the other thing that it demonstrated, it has kind of a different focus there.
PHILIP: Absolutely and that really comes down to Evan Czaplicki, the creator of Elm. I was able to get over to London in October a couple of years ago or a year and a half ago now to do a workshop with him at the Code Mesh Conference. You know, just seeing him teach this stuff and saying go into this and talk about the things in a bit more detail, it was very clear. First of all that he'd had a negative experience picking up Haskell, I think it was and he just thought, it doesn't have to be this hard. The things aren't actually that hard. It's the way that we're explaining them that makes it hard. The things that are actually under the surface is really simple. He has a blanket ban on this kind of technical jargon.
In the Elm community, he prefers to get things really straightforward names. I think he said to me that one of his thesis advisors or his university professor said, "Evan, that's what you get when you put a usability specialist into a programming language creator's shoes," that he does have this focus where he understands the benefits of static-type systems but he also deeply cares about the experience of not only picking up the language and learning it but also the day you solve it and that's something that just shines through.
I think even if Elm never makes it into the pantheon of great programming languages like that in itself and the influence of that had already on other communities, this is fantastic. It’s the tide that lifts all boats in many ways and we all benefit from that.
JEFFREY: We kind of touch on this a little bit earlier. We’ve been talking about the ergonomics of being an engineer working with Elm or Ember. What about the business case? We’ve mentioned how Ember has prevent a scale fairly well in large organizations. What's Elm's path to being able to do that and where is the niche to that it fits in right now?
PHILIP: I definitely think Ember comes out of Apple, I believe with sprite core. That’s where it started and it's interesting to see that that's also where it's gone in terms of the focuses and making it easy to build these rich applications. I think also that Elm has a similar genesis in the sense that Evan, I believe he did an internship at Microsoft and one at Google and I think there's a conference talk as well from large JS or elm lock or something, millions of lines of code.
It’s definitely gunning for the CM area which is applications which are large and hairy and trying to make the maintainability a lot better by bringing the strengths of the static type system to bear and bring the simplicity that that enables. That means that the learning curve maybe is a little sharper at the beginning, in a similar way, also that Ember was and is. But then you should reach this point where the maintainability of the app outweighs the time spent in learning.
I think about it a bit like test driven development. I remember back when I was doing Rails and DHH had baked TDD into the Rails itself and there was the years of the testing discussions whether to test all the time, test everything, 100% coverage or even full circle tests are a waste of time. but it's a similar philosophy in that if the tests are doing what they should be doing, which is giving you great feedback, the time it takes to get up to speed in testing, the time it takes to set up testing, the time it takes to write the tests, they pay off further down the line and that's not music to the ears of the people who want to get something into production immediately. But it's definitely music to the ears of people who will be spending a long time on maintenance in an application.
That NoRedInk application is huge. They have millions of users. They’ve build software for teaching grammar and skills in the US and they talk all the time about the benefits of Elm. Mainly, in the sense of confidence, I have the confidence to go in and change this code. All of bits or majorities of our code bits, which are things that we've rather not like to touch. We kind of section them off.
If a feature request comes in and instead of saying, "Yeah, we can do that," you try and slowly push it out the door. Elm is supposed to give you then the confidence to be able to go to any part of your code base and to change it without the fear that you're going to break something or break everything because the type system, the compiler will tell you, "You change this type, you change the signature of this function, here's where it's broken and as soon as you fix the compiler errors, functionally it works, you'd probably have a few tests to test the actual business logic of it." Probably not so with technical stuff but it's a huge time saver.
That’s where we want to be as developers and our relationship with our tools. We don't want to be fearing our tools. We don't want to be anxious every time we open our editor. We don't want to fear the feature requests coming in. We want to be in control of our environment and we want to be able to deliver the business value. I think that's certainly the promise of Elm. That’s certainly where it wants to be.
CHARLES: I love that. It sounds like that is where they want to shoot for is these big applications and they do want to scale massively. Let me ask the question then. Where is Elm do you think right now today on that curve of scalability?
In contrast to a new framework every month, he was like, "If Elm is going to be around in 10 years, then this is the decision that I'm going to make." In that sense, where is Elm in that journey. I'd say it's still pretty early on. I'd say also, Evan is really focused on use cases so if there's not a use case for something, there's no reason to add it.
He’s definitely building it very slowly but that may sound like a down side right now but the upside is we're going to come up with something at the end, hopefully that is battle-tested, that fits the use case and a good example of that was the URL handling at the navigation. It didn't live in Elm. Do we actually need it because he's building single page applications? Can you give me your experiences? There are a couple of libraries also built. Then the things from those libraries were then taken a little bit like Ember as well, where you use the add-on ecosystem to try new things and then things may get brought into core or make it kind of the official stamp of approval. But it's a lot slower and not as committee-based. Evan is the benevolent dictator for life.
CHARLES: I really like that approach. I don't withhold my opinion from the benevolent dictator versus kind of the oligarchy that you see elsewhere. But I was thinking as you were describing it that maybe the framework really, and this is me, I'm kind of a little bit of a tree nerd, mostly in and around the trees that grow in central Texas but I love trees. Unfortunately, it sounds like oak or maybe redwood would have been like a more appropriate name because those are very slow growing, have a very hard wood, whereas in Elm, it's actually fairly short lived and has a softer wood.
JEFFREY: And you also don't want cedar because that will come back and bite you -- highly allergic and toxic to humans.
CHARLES: Right. Named Elm but slow growing hardwood that you can build a house out of. But I like that. It’s so important to get things right the first time because that's where you realize exponential gains. Unfortunately, there is no substitute for exponential gains, rather than deep thought. At least that's been my experience. You can realize gigantic short term gains with taking cut corners but in terms of long term exponential, ultimately those will taper off.
PHILIP: I think as well, what we've seen in Ember's experience is the decisions that you take to open up a private API or not to make something private. They limit you because all of a sudden, everyone starts using those APIs. A year down the line, you realize, "Oh, hang on. We don't want to design it in this way. We want to do it in that way," but given the commitment that Ember has to stability, you have to deal with that.
Sometimes, the dealing with that actually takes more energy, more time, more effort in education, in actual code, in maintaining two branches, all of that kind of stuff then actually takes away from the time that you could have had to think up better solutions and it's a danger, you know? It’s almost like as soon as you throw this into the water that it becomes firm. As soon as you throw it into the crowd, the crowd will take it and they will use it and your chance to change it or mold it will have gone.
Evan talks a lot with language designers and at least I got that impression that he opens up channels of communication with people who've done this before and he thinks very deeply. He’s not closed off to the idea of adding more powerful features to Elm but he would only do it when it's right to do it. It’s not just is this feature right but is this the right time for this feature? That as well is just the kind of mind blowing like who thinks about this stuff, who cares about this stuff?
If those intermediate steps happened, if CoffeeScript and Babel would have happened, would people be as receptive to things like Elm or things like PureScript? it's definitely there is kind of the zeitgeist of the development community really informs what features are even possible or appropriate tomorrow, even though they might have some ejectively sound qualities out to them.
JEFFREY: They want to make sure they're not too far ahead of their time.
CHARLES: Right, exactly. Folks got to be ready for it. Then the other thing that I wanted to ask was given the trajectory of where Elm is, some people are using it in production. You said it was around like a 0.9, 1.0. if we are actually quite happy with our current code bases, whether we're using Ember, whether using React, whether using some other framework, what are things that you can learn from Elm that you can bring home with you to your own town, wherever that is to use and make your own life better, as you develop code, to kind of increase that confidence, as you say?
I think that the things that you will learn in that process are tremendously beneficial, even if you don't ever end up using Elm in a professional capacity. The thing I think that grabbed me most is because of the way the type system is, you have to deal with every error and this is something when writing Ruby or when writing Ember, you always focus on the happy path. Then once you've got the happy path in place, often the business will say, "That's done. Let's move on."
Instead of just dealing with those on the fly, if statements like why don't you create an object that encodes the states that your UI can be in and then you react and the UI will know what to do and you write the code to do that. But in some ways, at least to me that's a lot better than kind of hit and hope when we got an error. That's, "If do this, if do that, let's nest some callbacks or let's chance some promises." The sheer act of taking a step back and this is something Elm forces you to do because the way it is, like thinking about the domain and thinking about how best to model the domain. To be honest, it's probably just good programming practice but how often do we not do it?
CHARLES: Absolutely. It seems so obvious, except it's only obvious when you're put into a system that you can't slide on it. What you're describing reminds me of this concept that I came across in the Haskell community. They talk about wholemeal solutions versus piecemeal solutions. Most of the time, they talk about wholemeal solutions as kind of finding the most general set of abstractions that solve a problem for all cases. But I hadn't applied it to this concept of that all the potential states kind of envisioning your application.
As you execute your code, you'll only come across one state of the time but kind of thinking of your code as all of these sort of quantum states of your application existing at once and you need to account for them all and draw that whole picture.
PHILIP: This is a really stupid analogy but I don't know if anyone's played Age of Empires or any other kind of war-based game where you have a map, you've got the fog of war and you can only see a few meters around you and you have to send out your scouts and you need to find them --
CHARLES: Yeah, yeah.
CHARLES: Yeah, I love that analogy. Like actually sent out your scouts to know the map.
PHILIP: Yep and I think as a person, and this is maybe why it appeals to me it's like I like the feeling that I haven't left a stone unturned. I like shipping something that I'm confident in and I know it won't bite me or if it does bite me, it's purely because it was an unknown-unknown. It’s like doing your due diligence.
But I feel like there's another stage where it's like what are the tradeoffs involved in taking this decision or using this tool and what are the tradeoffs involved in taking this decision or using that tool. Just the way of thinking in terms of tradeoffs, instead of absolutes I find really, really valuable and that's definitely something that I've been able to apply and what are the benefits of this decision. Not only from a code point of view but from an organizational point of view, from a team point of view, from a long term point of view, from a short term point of view, having a wider arena to make those decisions, it has been really helpful for me, valuable on both side of them.
CHARLES: I can see it. I believe you and I think we're coming to the end of our time here, I think I'm going to make the same recommendation that I made to everybody last week when we were talking with Toran about Redux and that is you should go out and you should try Elm. It’s really easy to get started and it will guide you. The guides are excellent. The air handling is excellent and yes it is quirky and weird but it supports you. There’s going to be some big functional programming boulders thrown at you but it gives you a [inaudible] to deal with them.
JEFFREY: I think you need a suit of armor so we can stick with the Age of Empires metaphors.
CHARLES: Yeah. Go out there, try it and experience the things that Philip is talking about because I feel like they are very real and they will have a good effect on you, whatever it is that you do the next day or the next week or the next month.
PHILIP: They'll challenge the way you think, they'll give you new perspective and they'll give you a good counterpoint as well. You know, elm is not for everyone but only if you try it will you understand that. I think that's one of Evan and everyone in the Elm community is try it yourself. I'm not going to say Elm is right for you, Elm is right for your business. Rewrite everything in Elm, not at all, try it out, try it on a side project, try it to just having fun hacking around. If you do want to make the jump to using it in your company or whatever, don't take a decision to rewrite everything. Take one very small part of your app and do it in Elm and see how it goes.
If you don't like it, you can take it out. If it works, you can build another small thing. This kind of gradual approach were far too quick to jump on a bandwagon: trash the old thing, do the new thing. Especially now, when we are building applications that are going to be maintained, when we are building applications that are going to be around for a few years, it's also just not feasible and sometimes it's not right. You just got to make that decision maturely rather than I want to use this or I want to use that, which can be tough. But ultimately, you want to deliver the value. Try Elm at Elm-lang.org/Try. You can try on the browser and see how you go.
The official guide is really good. There's also Elm-Tutorial.org, which is kind of a complement to the official guide. Pragmatic Studio have released recently an updated course for learning Elm. James Moore at KnowThen.com. He has a couple of Elm courses, one is free introductory and then there's a paid course, which goes into deeper topics, whether text or whether video, whatever your thing is, there's resources available to get you to the point where you'll be able to make a decent decision.
CHARLES: Well, fantastic. I'm certainly excited. I haven't dip my toes in Elm for a while and I'm actually, after this conversation pretty stoked to get it on again. I hope everybody else does. Thank you so much, Philip for coming on to the podcast. This was just a fantastic and enlightening conversation.
PHILIP: Not at all. It's been an absolute pleasure. Thank you.
CHARLES: All right. That’s it from The Frontside. Remember to get in touch with us at Frontside.io, if you're interested in UI that is engineered to make your UX dreams come true.