Few seats left for our Advanced Backstage Plugin Development workshop on December 7th. Secure your seat today!

106: Microstates

Hosted byCharles Lowell, Taras Mankovski, David Keathley

July 20th, 2018.

In part I of The Frontside's microstates series, Charles Lowell, Taras Mankovski, and David Keathley talk about state management that’s easy and fun and transactionality.

This show was produced by Mandy Moore, aka @therubyrep of DevReps, LLC.

Upcoming Conference Talks:


CHARLES: Hello, everybody and welcome to The Frontside Podcast, Episode 106. My name is Charles Lowell, a developer here at Frontside and I'm going to be hosting today's episode and we're going to be talking about microstates with fellow Frontside developers, David Keathley and Taras Mankovski. Welcome you all.

DAVID: Hello.

TARAS: Hello, hello.

CHARLES: I'm really, really excited that we finally get to talk about this but before we jump into this, just wanted to let everybody know that last week, we published our roadmap for Big Test, which is a JavaScript acceptance testing framework that we've been building right here, in-house at Frontside, which can work with both Ember applications and React applications, Vue applications. It's in alpha phase and we're looking for feedback but we're really excited about it, so we're going to leave a link to that in the show notes and please go check it out.

All right. Finally, the moment has come that we actually get to talk about this publicly. We've been publishing things about microstates for a while now but we feel that it's ready to share with the world and that's really, really exciting. We should probably like wind the clock back a few years because that's about how long we've been working on this and talk about, kind of the why and the what of what microstates are. It's kind of a weird word. What are we doing here?

TARAS: Yes. That's interesting question because we were working for so long and after all this time, what is this specifically we're working on. I can speak personally from my personal motivations because we have conversations over the last two years to why we were doing this and I think for me personally, it's always been that I've been mentoring building complex applications for almost five years now and one of the things that I find consistently is that there are patterns for how to build complex stateful UIs that the required solutions, that are fairly reliably consistent.

I can teach people certain patterns and then they can use patterns to build complex applications and those patterns scale really well, the challenge is that, there's not an easy way to express them and it's different for every framework. The way that I would teach somebody how to do it, for example in Ember versus how they would do it in React, even though the pattern itself is the same but the implementation of the pattern is different but it's different in such a way that it's very difficult to see where the consistency is, what is the same about these two patterns.

It shows me that there is room for improvement. In the same way that if there is an opportunity that in the future, components will get unified under one umbrella or what component spec. The fact that we do states differently across every implementation, across every opinion, it suggests that what we might be missing is something that would unify across frameworks, how we actually do statement management.

CHARLES: Yeah, that makes sense because as much as people try to buck the trend of MVC, it can't keep coming back in the rear view. Hopefully, not like a horror villain but like a caring friend, like a reliable pattern. I remember when I was in writing Java applications back in the day, the most important thing when you were writing a swing application was making sure that you had your models right. If you were modeling either a form or a dialogue or a set of pages, the most important thing was to have those models.

Back in the day, we modeled those things with event listeners, very similar to where the way like a Backbone application used to work or if you're familiar with Ember Object, the way that it worked basically, adding observers to some model so that when that model change, you were able to react to that. Your representation was able to be 100% dependent on this model. That was like a Java Swing Application, which was then inherited from this pattern from Smalltalk but we keep on seeing this again and again and again. There is this piece, this state, that if you're going to be representing something, then you need to be able to get at the meat of it. That's ultimately what the model is in the MVC pattern.

I remember when React first came out, everybody was like, "Oh, MVC is dead," but now in terms of state management and all of the state management solutions that we see, what that really is, is the model trying to reassert itself. Because it's such an important piece of your application, it's going to rear its head and you cannot escape it. I make it sound like a problem. I guess, it is a problem but there's a proliferation of solutions out there that are attacking the problem in order to represent something you have to know what that something is.

DAVID: I think when you only work with the one framework, your perspective might be very much shaded by your experience. If you're really good at building React applications and you have subscribed to POJO is king and you don't need anything beyond that, then you're going to get good at handling POJOs and you might not realize that there is actually limitations to what you're doing. I think the same applies to people who use Ember and they're using Ember Object and they have computer properties and now, in Angular, there is observables and subscribing to streams.

What was really interesting with the work that we've been doing with microstates is that and one of things that have notice with other state management solutions is that a lot of them emerge as, "Oh, I had this insight. I spend a couple of days working on it. Here's my proposal, in a way, for managing state." I think that's great because insight and understanding is really important but sometimes, taking the time to design something can be really helpful. I think that one of the things that are really interesting with microstates is that because we've been iterating on it for so long, they've been testing it for so long, we've been able to do something in microstates that's really difficult to do when you have a production application that you're like, "I want something different. Something is not enough but I don't have the time to really make it better," or, "I have an idea. I want to release this thing but I don't have any other collaborators that I can talk to and really flush out these ideas," so what you get is a solution but it doesn't always touch on a lot of the angles that you want to touch on.

It's really interesting when you think about --

CHARLES: I'd say, we should talk about those angles, like what are some of those angles?

DAVID: Yes. You could think about this from a few different perspectives. I would say, from Ember perspective because that's been my personal starting point. For anyone who was listening. Ember and Vue, for all intents and purposes, in this scenario are very similar because they have the same primitive, which is the computer property. The computer property in Ember, the computer property in Vue is almost identical from the way that you actually consume.

In Ember, you would use computer properties and in Vue, it's just computer properties to create derived state from data that's passed into the component and then, you would use is derived state to basically, decorate information that comes with the props, so you could present it in your component. This pattern is very nice. This pattern is being used to build LinkedIn, to build applications at Apple. There are huge implications being built by Ember community using this patterns and I'm sure something similar has happening now with Vue but this pattern doesn't really exist in React in the same way.

You might opt in to start using this pattern if you start using [inaudible] but it's not quite the same. Writing computer properties in Ember and in Vue is essentially free because it's so effortless, where in React, if you're using [inaudible] to create cache computer properties and memorize computer properties, you are really opting into doing a particular way of computing this properties and writing selectors that's not trivial. I think that's one thing that is available in Vue and Ember and it's a really effective pattern and you can do it in React but it's actually not possible in Angular, unless you're using, I'm guessing something like ngrx, where you can do similar things to what you would do if with redux in React.

Creating derived state that you can consume in your component is one of the things that is possible but it's not possible consistently across all frameworks. That's just one of the things.

CHARLES: I agree. I think that was one of the things that drew me to Ember at first coming from Backbone as I did and honestly, from the models in Java was that, in order to compute anything, you had to install a listener and then eagerly make that computation and store it somewhere, where as those frameworks, it made you feel effortless where you can just decorate some state and derive it and the information is there, the computation is there when you need to reach for it but you don't have to do expend any extra effort aside from say, "This state is derived off this other state."

I think another case that I came across was immutability. Immutability is a means but the end is to have consistency inside your application. I first really started running up against this when I was working with forms and since then, I've realized that actually, a lot of the pain that I was feeling was because things were not being immutable but this is where the fact that things weren't immutable ended up causing me a lot of pain and headaches and I was having the code into being a lot more complex.

Essentially, when I wanted to make something transactional if you're editing a form or something like that, then you need to essentially store off a copy of your current state. If you're editing some object, I want to say, open up a dialogue and I'm going to edit it and then I'm going to commit the changes back to that dialogue. What I'm modeling there is a transaction so I kind of need to shave off a copy or make a Xerox copy of the object and then make the changes inside to that object and then somehow, try and merge those changes upstream and then, get them back into the main lines. Really, like a very Git-like operation.

It wasn't just at the object level. When you're doing things like dirty checking, you need to make what is the original value of a field versus what does the input currently have. I might be doing any number of transformations in between the actual physical representation of the field, maybe it's a date but the user interacts with it in as a string, so there has to be this kind of parsing serialize thing sitting in between that value and you need to basically, keep a copy of that field as a mini transaction within your macro transaction because you want to say like, "Is it dirty? is it really the same object?" because if you can just do an object comparison, you have to get into all that hairy like equality checking and it gets really complex really fast but it turns out that a very clean solution to this is if you just never actually make the changes in place and whenever you are making changes, you're generating new information without destroying the old information because when a form is the case where we really come up against this, where we're modeling the same object over time, so a form explicitly models the change of an object.

Change is part of what it is and so, the definition of change is being able to compare something in its prior state to be able to compare it to its current state. If you're making that change by destroying the prior state, then you're going to run into a lot of trouble. It just turns out that when you're working with forms and it turns out there are a lot of use cases like this but this was the one where I first really just couldn't even... Without immutability, you want to model your change as always rolling forward and not ever destroying prior states but being able to pick and choose and can always be able to look back to where you were and compare where you are now to where you were.

That's where immutability comes when you're modeling change. It's actually much easier to model change when you have explicit states that represents what was and what is. But when you start doing that, then things get a little bit messy. You have to do the compromise. There's a tradeoff, like when I say, "Set this property on this object," that is easy and that's something that I think that we can all understand. We're not idiots. That's why immutable models are very conceptually easy to grasp, to wrap your head around.

If I'm modeling myself and I'm saying, "Set hand position two feet up in the air and now I'm raising my hand." That's easy to understand and it turns out that when you're changing a data structure immutably, then you have to, for example model a simple set operation as duplicate and swap. Or if say like you're swapping out one property in an array, you actually model that as a map, where you swap out the one element at that index, rather than just saying, "Just set the thing at that index," or if I'm changing a property of an object, I want to copy over all the fields except for that one which I'm going to change. If you start to do that then, you realize the benefit of always being able to look backwards but you've now introduced overhead and complexity in your code, so you've made a tradeoff. I think that's a lot of people look at saying, "I want this to be immutable," then they actually have to come to the grips with the complexity that's going to introduce.

There are libraries like Immutable.js that do make that a lot better but even they have the problems. I can talk about those but one of the cases is like being able to reason about a series of states for your object, rather than just only having one copy of it ever and being stuck with having to deal with it as it is.

DAVID: Yeah and I think that really important too is being able to mentally track where you've been because in my experience on other large Ember applications, I've run into so many different bugs that I could really track down to people or rather, places in the code where things are just reaching in and mutating your model, where it wouldn't make sense, so as you're writing your code, you're in a completely different state than you might have expected.

CHARLES: Yes and it's easy, right? But essentially, when you have some model, you've basically got a global object. There's a lot of recomputations that needs to happen when those things change. As a result, if you look at the actual code that supports computer properties in Ember and I'm sure other frameworks as well, some of the most hairy and complex. If you look at the chain watchers and the chain nodes and all the stuff that's required to support things like computer properties, it's amazing that it works as well as it does. I've tried to actually open that up and understand that code on a number of occasions and every single time, I had to walk away in defeat.

TARAS: This problem exists in React applications and Angular applications because if you think of the challenges of working with set state, I think one of the problems with working with set state in React is that your complexity increases very quickly. It starts off kind of simple when you're using immutability with set state in React. They're kind of very light and then, as you start to add features, your complexity starts to grow very quickly in a way that you really start to run into limitations. You start to confront your personal limitations of your abilities to work with immutability, so you're limited to doing immutable structures that only a few levels deep because your ability to use destructuring to create new immutable object is limited by expressiveness of JavaScript. It's gotten a lot better with the destructuring but if you do it a few levels deep, I think everybody's familiar with what happens. It starts to get really hairy. You kind of loose track of where you are.

CHARLES: Yeah. The signal-to-noise ratio increases because after probably, two or three levels, the majority of your code has to do with destructuring and restructuring and very little of the actual change that you want to make.

TARAS: And this is kind of interesting because people talk about declarativeness versus imperativeness but at certain point when you have a complex immutable state change, if you're doing with destructuring your code, even though it looks declarative but there is so much processing that you're doing, it's actually kind of losing the benefit of its declarativeness. It's actually starts to look more like imperative code than it does what you would expect a declarative system to look like.

I think this touches on the other aspect. It kind of compromise that when you work immutably, when it compromises, you make a serialization. Your ability to represent your state as a POJO becomes restricted by the fact that you have this complex system that's wired together and you have systems like zones in Angular and in Ember Object that are able to keep track of changes in these objects but you don't have a way to restore those objects. You don't have a way to do more sophisticated things that you might want to do, especially in situations where if a service feeding you, what do UIs going to look like. In that situation, it's really helpful to be able to say, "Here's a POJO that I got from the server. I'm going to use this POJO to build this component tree that the user is going to interact with and then, as the user interacts with it, I'm going to then, capture that state, serialize it and put it back in the server."

When you're using something like Ember Object or if you're doing this kind of stuff in Angular or even if you're doing this stuff with React but without using something like redux, you essentially end up doing so much wiring to accomplish that. By the time you finished writing your application, you've written a ton of code just to handle this particular use case and if you have to do this again in another application, you just rewritten that kind of code in a new application as well.

CHARLES: It reminds me of the concept of a Smalltalk image, like there's no way to really get at the state of a Smalltalk thing. It's almost like you're dealing in docker containers and not actually being able to write that state down into JSON or something like that.

I'm trying to casting about for an appropriate analogy. Maybe that's not a good one but what's actually happening cannot be made orthogonal. It can only exist in that one run time that you're currently running. If something wrong manifests itself, reproducing it can be extraordinarily difficult, right?

TARAS: Yeah.

CHARLES: Imagine if there's some render cycle that's making a bunch of mutations and there's this process that you stood up and it runs in completion, some signal comes in and those effects are like ripple through the system, there's no way at any point to have any other representation of that system than the running system itself.

TARAS: There's another element to this, which I find really interesting. When you're thinking about how to architect complex UIs, it's actually helpful to get really clear about what kind of changes are happening. A lot of times when I want to see beginners are writing, especially if you task someone who is a fairly junior at building a single page applications, a lot of times what will happen is because they don't have a clear mental model of what is going on in regards to state. They end up setting a lot of properties. Every operation, every time you have an event handler because they don't have a clear model of what's going on, they end up setting like five or six or seven properties.

That kind of signals that they don't have a clear picture but what that also does is a lot of times, they usually comes together with cascading state changes. Usually they're not representing a single operation as a single state change because there might actually be a bunch of things that are happening because what they're doing is they're massaging the system into submission. Not like they're not in control of the state transitions, so they use, essentially time and their dedication to kind of sort it out and make it work and eventually, would that ends up looking like that if it works for most of the cases that they are able to test for or that they able to manually see. But then they AR, not accounting for problems that they're not able to understand right now, they become discovered by users when users start to interact with the system and with the components or with that application and the application is there to get into some funky states.

The tools that we have, they don't prevent that from happening. They just --

CHARLES: Right. They don't force you. I think what you're saying is that ideally, you want to model your UI as a set of transactions on your state, that you want transactionality to your state so that I basically am saying, "I'm not going around and setting seven properties in reaction to this one event." I'm saying, "This event triggers this transaction and that transaction clearly bundles up every single operation that needs to happen and the tools don't enforce that." Is that a fair --?

TARAS: Yeah and then, the problem is we're working with component trees. You start off with having a set of requirements and over time, the requirements change. As the business unit understands your application better, they give you more direction of how accounting should work and then you find out that there's more interconnect at stake but then what's happening now is that the cost of refactoring those state that spread throughout the components, whether that be with set state, whether that be with the actions in Ember or even in Angular.

What you end up doing is you start to change the system but change is not trivial because the actual process of changing where that state lives is not linear. It depends on the complexity of the code that you wrote and it just gets really hairy very quickly. That's where companies end up losing a lot of time. A developer could start off with a requirement, you build something and then a new requirement comes in and instead of it being a simple change, it turns into a week or two weeks refactor because you now understand the state ownership should be different. The state that you have should be in a different place. You have to manually make that change. You have no obstruction to help you express that in an easy way.

CHARLES: Another thing, because we are doing a kind of a roundup of all the things that you need to account for when you actually embark on managing your state. Another is actually constraining the amount of computation that happens. If your system is based on listeners and large chain reactions of things where it's like, "This property changes so I need to notify these other 10 dependent properties that change." You can do a lot of unnecessary computation, especially if nobody is going to render.

That's kind of the thing that you have to do if you're going to be immutable. You have to eagerly walk those change to see which objects are affected so that you can then invalidate those caches. A system like Ember Object, I don't know exactly how Vue works, it mitigates this somewhat by the fact that the computer properties are lazy but you still have to walk all of those chains. That can actually get out of hand. They're eager, not lazy.

Then the other concern that you have, where you normally have to make a trade off around is around composability. One of the things that's really nice about immutable systems is they're very composable. If I've got some object that does one thing, I can then just set that object onto another object just by mutating one of its properties and I've effectively composed them.

I can then install listeners onto that thing or I can compute properties off of that property and they can post pretty well. That's something that you get but then of course, you're losing all of the benefits of immutability, so things like Immutable.js don't really compose very well or redux doesn’t really compose very well. The concept of taking a redux store and embedding it into another redux store, you just don't see that. I would never distribute and I think ultimately, the litmus test there is would I be able to share it on something like npm. Nobody shares an npm package that's just a redux store that you can dispatch actions to and observe and use it with your other redux stores.

When it comes to a system like Immutable.js, that does make transitions a little bit easier over lists and arrays and maps but you still run into the exact same set of problems that you have when you have lists of maps of records and you don't really get any help there, so you have to make this tradeoff between immutability and composability, whereas a system like MobX or Ember object actually quite composable.

Before we start talking about microstates, I want to say that you just throw those in there because there is just a lot of concerns out there, a lot of edge cases that actually build up but through the course of a real application, you will encounter them all. You might be making tradeoffs at the beginning that you're not realizing that you're going to need or are going to get you into trouble later on.

DAVID: This actually happens in the Angular community as well because there's something really great that's happening with observables in the Angular community. I think everyone's embracing them wholeheartedly and I think that's really been pretty great to see but observable streams of composable, but objects that have on them observable stream providers of some kind, like if you have something that you can subscribe to and that is part of a property in a class, composing multiple classes together and consuming properties from those classes, there is no mechanism for composing that. That kind of composition has to be done manually.

Again, you're kind of manually wiring together a bunch of objects and the big challenge is that you are manually subscribing to all those streams and unlike what you have with components. Components have lifecycle hooks. When your components is being torn down, you know you can perform some operations. If you need to remove an event listener, you have a hook where you can do that when you have a class instances like JavaScript class instances that have on them properties that have observables that you subscribe to. There are no tear down hooks for class instances, so there are no obstructions from managing unsubscribing from those streams.

You essentially end up having the foundation that you can use to build complex reactive systems and you can subscribe in there really fast but wiring those things together at a bigger scale is simply not there. It's something that you have to create and enforce yourself.

CHARLES: Right and I think that's probably a perfect segue into talking about microstates, which is the project that we've been alluding to for the past 30 minutes, that is I think in attempts to solve these problems and make sure that you don't actually have to compromise on those things, so you can reason about things locally but have those things be composed into a greater state. But also have them be immutable so that you can look at past states and reason over a data structure over time as opposed to just in one instance.

Also, have an intuitive interface that when you're making these changes, doesn't look like half of your code is unpacking some data structure, flipping some bit in it and then repackaging it back up again. That's the context. Should we start talking about what microstates is and how it addresses those?

TARAS: That's a good next step and when I start working in the ReadMe, I end up actually, I think I wrote about 40 pages. One thing that's interesting about microstates is that and this was part of the design of microstates from the work that we've done is that we intentionally wanted to make the number of ideas that you haven't microstates very little, so when you use microstates, the number of concepts that you need to remember in your mind is very few.

It is a conceptually a different way of thinking about organizing your state in the same way that shifting from managing DOM elements directly to having an obstruction like component that declaratively applies changes to your DOM tree. In a same way, microstates is kind of an abstraction that allows you to declaratively describe how your state will change and it will manage the transitions for you and allows you to give the state transitions names and it allows you to give your states names as well, so you can actually name things. You don't get a POJO that has a shape but has no name. You actually get to give things names.

CHARLES: So, why don't we start? I have a list in my mind. I should probably write it down of the things that we just talked about but I think the things that we talked about are ease of representation, like conceptually easy, transactional, basically serializable and immutable, lazy and composable. Those are like five or six things. But I think there are kind of aligning principle around which we gathers that the state management should feel easy. It should feel fun.

One of the things that is awesome about working with components, whether you're using web components or React components or Ember components is when you get it right, you're just snapping these things to feel together and it feels great. It's like I'm just passing properties and render blocks down the tree and the framework is just doing all of the grunt work for me and I'm just operating at a very high level. That's what organizing principle with microstates as it needs to feel easy. Maybe we should start there and just say, how does that easy and fun line up with each one of those kind of unique problems around which we typically have to make tradeoffs? We could start with the interface of making a change.

TARAS: I'll go back to the starting point. I remember what got me first interested in microstates is Charles, when you said that, when you have a number, there are certain operations that a number can do. We really don't need to be writing an increment operation for every... Like if you a have a number, you can increment it and decrement it.

CHARLES: Honestly, every time I see state management tutorial and they tell you how to increment and decrement a number with it and you write the increment code and you track the thing and you store it back into the store, at this point I'm still annoyed with those tutorials because I'm like, "It's a number. We know we can increment it. Just show me where to plug in the code. I should not have to be writing an increment method."

TARAS: Yes. And that's the kind of starting point. There are certain operations that you can perform with the primitive types. If you need to add a number or if you increment a number, we already know how to increment the number. It's part of microstates. But that in itself is nice but that's not that important. I think what's really important is that when you need to put a number into another data structure, let's say you have a nap and you're like, "I need to..." I don't know --

CHARLES: Let's say, like a click tracker that has a number of clicks.

TARAS: Right. By itself, you can increment the click tracker but if you need to put a quick tracker into another app, essentially you can compose it in and you don't need to figure out how to wire the actual mechanism of how to make sure that you can update the property, like it's part of another class, for example, you don't need the way you would increment the number. When it's a part of another class versus how you would do it when you're working with it by itself is an approximately the same. The amount of work that you need to do to actually perform that operation is the same. Your complexity doesn't increase as you compose one data structure into another.

CHARLES: Right. You can just say, "This is an app. It's got a click tracker and this property is a click tracker and I have to do nothing else. I can register clicks on that thing. It doesn't increase the complexity of application at all."

TARAS: There's no wiring. Now, you added some new state, that state is very explicit and it's really clear that it is not impacting other parts of the state. You can operate with this thing. If you change it, it's going to work properly with all the other things that are in the type that you are adding this counter to. Those things are just going to fit well together and it's not going to break if you need to transition one more thing. All of the other transitions will work the same way. I think that kind of consistency is really meaningful, over time especially when you start to increase the amount of state that you manage in your application.

CHARLES: Just the ability to work with types and just have kind of those implicit operations and have those things compose, kind of indefinitely. Moving down, we talked about easy and the other thing I would put on that is that the way in which you express those transitions, for example if microstates comes bundled with numbers and Booleans and strings and arrays and objects and kind of the stable of types that you would expect in any JavaScript application but those types are expressed using this way for expressing types, essentially.

When you actually do make a transition, it feels very object-oriented, I would say, even though it's not. It feels like you're mutating but you're actually doing a transition. Does that make sense?

TARAS: I think for anyone who is familiar with what it's like to write queries for GraphQL, if you're not familiar with it, it's fine. You can get a sense of that from microstates but if you're familiar with the ease of just writing a query and if your backend knows how to retrieve the data, then your queries will just give you the data that you want. That feeling is really powerful and just being able to write the query and just gives you what you want. Microstates is kind of like that. Actually, the inspiration came from experiences with GraphQL, which is that sense of ease is what we wanted to have in microstates and so you get that seems sense of like, "I can just do what I want and it just going to work and with this other thing and it just going to work," and you're just like flying through, like adding states to your application and it's just working for you and working for you and working for you and you don't have to do monkey work like gluing things together. It'll change how you are working before because you have a way of opt working with these things at a higher level.

CHARLES: Right. Let’s talk about transactionally or should we talk about immutability? How does this make immutability easy and fun?

TARAS: I think one thing is that you don't have to write reducers and you don't have to do destructuring by hand. I think you have a way of expressing. Thinking about this, if you have a component tree and let's say you have redux and then a bunch of components like your parent component, your root component has some state using sets date and then components further down the tree also have state. You could actually express that as a microstate.

What you would do is, essentially the parent component state would be the root and then the children's component states would be composed into it. The nice things about doing that is that at the root level, you have access to transition state of the children declaratively. You know where the states for those children is on the route type and you can write transitions that are going to declaratively perform multiple operations on the children state and I suppose it'll restructure to what happens with components but if you don't use this, you might have multiple sets date operations.

The process of wiring data from the root down to the children is kind of complicated, where here, you have a way to represent that and perform a lot of transitions in the way that is going to be just easy at whatever level you need to operate at.

CHARLES: Right. I think, for people familiar with redux, in redux you act globally and then you react locally, if that makes sense, so you dispatch an action to the entire store and every single reducer can see that action. There's ways to manage that but effectively, you have this one atom and then you have the reducers that kind of act on local state, whereas with microstates, you're basically acting locally. You're reacting locally but the effect is global.

TARAS: You're participating globally.

CHARLES: Yeah, participating globally but you never have to consider the context that is above your own, so you never have to be mindful or cognizant of the context in which you're enclosed because from your perspective, it just doesn't exist.

DAVID: Every microstate has a set transition which is the basic transition that you can invoke, essentially in any type, so what's interesting is that it's amazing how powerful --

CHARLES: So, we should break it down really simple. Basically, when you create a microstate, with a type, you say like, "I want to create a number with the value five," and then I can just say, "That returns a microstate," and I can say, "microstate.set 10," and that will return a new microstate who's also a number but the value is 10 and that's available on all microstates.

DAVID: Yes. If you have a tree of components and your state is presented by a microstate at the root level, then what you can do is you can invoke the transitions on any part of the microstate and it will just know how to properly create the next microstate for you. The example that Charles you gave of one number so that number can be inside of a class that represents state for a particular component and then that can be a part of another class or represents a state for another component but then, when you invoke a transition on one of the leaf nodes, an equal sets state on one of the leaf nodes or equal set on one leaf nodes, it will respond locally but it will actually reflect the changes globally. At the root level you're going to get a new object that causes the components to update.

CHARLES: You know what? I have another concern that actually just popped into my head, which is something that I've certainly struggled with in every single application of notable complexity is stability of value. We should put that on the list. We're almost out of time to talk about this. We spend too much time... Well, not too much time, of the issues of state management, which I think you can't spend enough time talking about but I did want to pile one more on there is when you're making that transition, where you're acting locally but you're participating globally.

For things that are unaffected by your action, remain unchanged. That is a super power. It's actually very hard to do with a lot of state management, especially when you're cloning a bunch of stuff, being very judicious about what you don't want to clone. Where this really comes into play is if I want to re-render something. A lot of times you have to jump through a bunch of hoops to tell, did my model really change or did only referentially change? With microstates, when you make that local change, if you're embedded in a very large graph of objects, obviously all of the objects above you are going to be changed but what about things that are off to the side of your siblings. They're outside of that scope of that change. They shouldn't be cloned. They shouldn't be copied over. They should remain the same.

If you're doing a re-rendering based on the changes that are happening, that's going to be a key feature because you're not going to have to write, basically any hooks to say, should I have to re-render my component. You can always rely on triple equals.

DAVID: This quality is going to describe the structural sharing and some of the other tools that are available but I think that's how it's accomplished. I think one thing that's a little bit different with microstates is that when it comes to structural sharing, it's not that difficult to do if you just do structural sharing and value. Meaning that you can do structural sharing on a value using something like lenses and not a lot of people are familiar with lenses in JavaScript but it's actually only three functions that you can use and it can give structural sharing on complex POJOs. It's pretty easy to use relative to how little people know about it but what that doesn't do is it doesn't allow you to graph of objects that have their own operation that you can invoke and that will perform structural sharing. That part, I know that is not available in any solution, I think at the level of completeness and luxury that microstate provides when you write things.

CHARLES: Right because every piece of the tree kind of comes bundled with its own things that you can do with it.

DAVID: I don't think you should jam everything into this podcast because there's a lot to talk about it. I think one of the things and we've talked about this a lot, which what we want to do is create an implementation for an idea of what it would look like. What would it be like if we had a composable state primitive that we could use to describe state and share state solutions in the same way that we share components like react-virtualized or whatever your particular frameworks or popular component may be. What it would look like if we had solutions to state problems that we could share and we could --

CHARLES: I think the litmus test of an awesome solution is like you look at this current crop of MVC frameworks and what's so awesome about it is they're sharing. That can happen, right? If I'm writing a React component, I can publish it on npm and other people can use it. If I have an Ember add-on, I can publish it and people can use it. They can consume those components. That's awesome and I think it's the hallmark of a great system.

What would it look like if I wrote just the state piece of a file upload and I could publish it npm and then anybody, in any framework, could actually use it with their framework without paying any penalty. What would it look like if there was some transactional data store that could be built and shared and the hooks into any framework were minimal. The possibilities are really exciting around that idea, whether it is realizes microstates or not. But clearly, we feel this is something you should be able to do.

DAVID: I'll add one more use kind of use case that I personally find really motivating is that there's a lot of companies that are investing into building single page applications and a lot of times, what you see happening is they're building a very similar application to what they had before because their business hasn't changed. The technology has moved on so solutions have improved. The demands for better user experience have increased but the actual business and how the things that people have to do on day to day within their company hasn't changed.

What we're seeing right now is we're seeing the same, like whatever was written before in jQuery or an AngularJS is now being rewritten in React or Angular or whatever you might Choose. But whatever you like to see is it has a situation where the domain specific logic of your business is represented as a data structure that knows how to, potentially, in the future talk to the server and retrieve data from the API because that's likely not going to change.

But you can use that like that's been tested and published as an npm module within your enterprise and then you can then consumed that in any framework and it's actually easier to do it this way, than to implement it in a framework-specific version of their state management. That's the part that I find the most exciting.

I think one of the things, just to connect to the goal is that, we would like to keep this conversation going. If you're interested and I think this is a kind of a call to our audience that if you're interested in this, we would love to have in our podcast to talk about these things because I think there's a lot of things that microstates really is a beginning of a conversation. It's not meant to be a statement. It's meant to be a proposal that we can just talk about this.

CHARLES: I agree and that's one of the reasons we're keeping it very small at this point. The core library of microstates is not setting out to accomplish too much. In the core library, there aren't even any side effects. It's actually impossible to have side effects. It means that it cannot be used for anything except for the model but that's very liberating and it let us focus on what would a system like this actually look like.

DAVID: It's really exciting because this has been the biggest metric but we have microstates in the JavaScript weekly and that was great and then it got circled around when people know and it's like 800 stars now, which is not a really big deal. It's funny because somebody commented like, "How can you put something in production that only had 100 stars?"

CHARLES: I think it's just important to realize that this really is the beginning of a conversation. There's a really exciting set of things to come. We haven't even talked about how we're going to model side effects, although we're going to use microstates to do it. We haven't even talked about what the various framework integrations will look like and what are the best practices for using this to organize state in your application. We've had some lively discussions internally about what that looks like. There's still a lot of questions but it's going to be a really, really exciting and edifying experience to get to answer them.

DAVID: Yeah, it's pretty exciting. I'm excited too. There's been a lot of interest from people in microstates, so it's going to really great. I'm looking forward to meeting people and having conversations about how we can use microstates because I'd love to have someone create a really great solution that I could just take off the shelf and just use and not have to implement them myself.

CHARLES: All right. Well, I think we could talk about microstates for at least the next three hours but we have to give everybody an opportunity to, at least like go to the bathroom or something. Microstates will return but if you're interested in learning more about microstates and you happen to be in one of the many places on which we’re going to be presenting on microstates in the future, who knows? Maybe you can come in and join the conversation in person.

Taras is going to be speaking at Toronto.js on July 30th. He's also going to be presenting at Manhattan.js on August 8th and then, yours truly will be presenting on microstates at React.js Austin on the 6th of August. Come out and see us. We'll drop those in the show notes and it's guaranteed to be a good time and we'll have that conversation.

Until then, we are the Frontside. We lead with the why, the how and then the what, if you're interested in working with us and that helps us that we guarantee the lowest total cost of ownership for your application. We're always looking for feedback. If you have news items that you'd like to see at the head of the show or just any feedback or questions, we would be happy to answer them.

Thanks today to Mandy Moore for producing our show and next time, we'll be talking with Kristian Freeman about what it's like to run an online conference with Twitch, so I'll be looking forward to that. Bye David. Bye Taras.

DAVID: Yeah, thanks for having us.


CHARLES: Yup, and bye everybody. See you next time.

Next Time:

Running An Online-Only, Free Conference on Twitch with Kristian Freeman

Listen to our podcast:

Listen on Apple Podcasts