Florian Rival PARISS: Okay. How was everyone's lunch?

[ Applause ] Good. I had a really successful lunch. I finally

beat all the kids to the desserts. I didn't get any on Monday. So, feeling good. Can we

give a round of applause for our next speaker speaking on native web apps, Florian Rival.

[ Applause ] FLORIAN: Hi, everyone. My name is Florian,

I'm working as a software engineer at Google. And today I want to talk about something that

I really like. It's applications. And when I say applications, it's any kind of application,

but in particular the ones that allow people to create things or to do things that they

were not able to do before. So there is something that I really like. It's video games. That's

why I created GDevelop. It's an open source and cross platform game engine that anyone

can use to create any kind of games. Why anyone? It's a visual editor for games where you can

create. And there are some kinds of visual programming. I will show an example. And anyone

can use it, even those who don't know thousand create games. And then you can get an HTML

game that you can put on the web. So, here is what the software used to look like. I

say used to because it's a software that was created, in C + +, it's a desktop C + + application.

You can see on the right, there's a list of objects in the game. And in the intermediary

of the game. Is, here it's a platform. And what we want to do as an example is to have

when the player is stepping button, we want the button on the platform on the right to

go up in the air. I will show how to do it using visual programming. And so, here we

go. I go to the events tab and I'm adding a new event. An event is a condition, an action,

kind of like an if then. I'm adding a new condition to check if the player is collision

with the tree. And if that's the case, I can add two actions. The first action would be

to change the animation of the button so it's going to the state where it's pressed. It's

animation number one. And I'm adding another action, which is an action to move the platform

up in the air. So, I'm adding to the platform a false on the Y axis. And now, if I press

play, the game is compiled to demonstrate and running in the browser. And we can verify

that the button is working. It's kind of simple for people that don't know how to program

and it's only based on logic. Syntax. So, a few games have been done with it. So, a

few were working quite well. And the game with the cat, but at some point I had some

issues. And the editor was getting quite old. I had lots of cross platform issues. In particular

on macOS and Linux. It wasn't working that well. And iterating on the software of the

quite slow, C++. And it was limited to old UI components, and the UX would need some

enhancements. And the entry barrier for new contributors was quite high. C++, again. So,

it was time for me to react. And I was wondering, would web technologies help? I have been doing

a lot of React before at work. And I was like, okay. That's perfect for making complex UI.

I was iterating. Can we have hundreds of elements, dynamic panels, 2D and 3D visualizations,

and all of the things that make a real application, you know? Can we make an ambitious application

using React? And the same goes for JavaScript. It's perfect for most applications. Or is

it really? Can we reuse an existing native codebase? Should we? Or if we have advanced

situations or computations, what do we do? Or if we need consistent performance? Still,

let's look at the quick post of GDevelop. There is the core of the software, composed

of the C++ classes and defining what a game is, what is an object, and how an event is

composed of conditions and actions. And then all the tools that are transpiring to the

events to JavaScript that you get a running game. And on the top of this there was an

interface in C++ using WX widgets. And there was the interaction with the device. So, my

first idea is, okay, let's replace the interface with React. Because I have an idea that it

could be better. And in a fortified system, I could write adapters for Node.js or even

in the browser simulator file system. We could run this on the JS or a browser or on electrons

so that we have a browser in the JS and you can package the application. And for the central

part, we could use WebAssembly to expose the code to the rest of the JavaScript application.

So, a few notes about WebAssembly. It's a kind of a language that is running in your

browsers. And it's a bit like assembly, meaning that it's not one that you write to, but what

you're compiling to. And at the end, it's simply like a language. Browsers are able

to generate really fast execution of any program that is retaining WebAssembly. And it is running

on actually most recent browsers including on iOS and Android. So, how do we write WebAssembly?

Well, there was  the first solution was to use Emscripten. Which is what I did. It's

a project with a C++ WebAssembly compiler. It's a C++, running through a compiler, and

translating it to WebAssembly. And there are solutions for any language you can compile

to WebAssembly. Or even things like AssemblyScript which allows you to create WebAssembly using

a syntax that is really  that looks like TypeScript. So, it's really interesting. In

my case, I have been using Emscripten. And I will show you the way it works. So, using

Emscripten, you can download it and sort it. It's pretty easy. And then getting EMCC, the

C++ to WebAssembly compiler. And you can run it on a C or C++ file. And in my case, as

I have a large codebase, I can replace the build to chain that was using the compiler,

and instead of at the end having a binary, I have a WebAssembly. And so, we have the

WebAssembly, which is the C++ codebase compiled to WebAssembly. And we have to expose the

existing classes in C++ so that we could view them in JavaScript. So, let's take an example.

For example, we have a class that is an objects container. So, it's  think of it as the list

of objects that you've seen. So, we could have objects and get some objects inside.

And then we can have a layout. It's a bit like a level, again. Which is iterating from

an objects container so it can have a list of objects. And there are additional, name,

to change the name. So, also, we have this. We ought to write bindings. Bindings in my

case I have been using WebIDL. There are a few ways to write it. But in my case, use WebIDL,

we use something that is the interface of your C++ class. You have to write it for all

the classes in your codebase. And this will generate new code that will be doing the bridge

between JavaScript and the WebAssembly module. So, the way it works once you have compiles

the WebAssembly module plus the bindings is that you have a module that you can instant

ship into the JavaScript engine. So, on the browser or in the JS. And when it's an instant

change, here it's called GD. You have this module and you can start creating new classes,

new instances of your classes just like usual JavaScript. And you can call them and it's

working more or less automatically out of the box because the primitive types of your

functions are automatically converted. So, depends on the language that you're using.

C++ or another language. But in my case, C++ and JavaScript are translated to C++. Same

for numbers and for strings. It's also more or less automatic. So it's quite convenient.

If you are passing objects to the parameters, then they are converted to a pointer or a

reference in C++. And you can debug using the input and output stream that's converted

to console.log in JavaScript. With this, we have something that is running in the browser.

Here is the Chrome debugger. And you can see on the top, they also output  it's outputted

by the C++ codebase. It's working. And then I can create a new object. So, yeah, it's

working. And it's quite nice to have, you know, with V C++ codebase you have been writing

since a few years. There are a few things to know. First is the memory management requires

care. Because when you're creating a new object, it's quite interesting because it's not garbage

collected. We mean that. If I'm looking at what this layout object is, it's a pointer

to some place in the WebAssembly memory. So, you can think of the WebAssembly memory as

a huge array that is containing bytes. When you're creating a new object, it's already

creating some bytes in this memory. Meaning that if you just drop the reference to the

object in the JavaScript world, the WebAssembly object will still be living in memory. So,

you have to explicitly call GD, to destroy, to remove this content from memory. So, in

the case of using React or any component based framework, if you're creating a framework

when the component is mounted, you have to remember to delete it or you can use an effect

hook if you're using React. So, know that the output files are quite large. For GDevelop,

it's at least 3 megabytes when compared to the WebAssembly module. And honestly, I don't

really care because I'm making a kind of desktop application, a rich application. So, knowing

that I will package one thing and make it a few hundred megabytes. 300 megabytes, more

or less. And something else to know, having a complete test set is really, really useful

because without the kind of heroes that you will be getting when something is going wrong

in the WebAssembly module. And for example, you have the wrong type of parameter. Or if

you forgot parameter, then instead of getting a number of boolean or a pointer in the C++

world, we get undefined which is translated, I guess, zeros. Everything will be broken.

And same thing if you're using a deleted object. Trying to iterate on something that was removed

from memory, and it will crush, basic lip. This being said, we have something that is

working in the browser. So, my next change was, oh, I can create an interface on top

of it is that as good or even better as the one that I use to do in the native world using

C++ before. So, let's see how to do it. Again in my case I was using React. And my change

was to make all of these examples. For example, context menus, trace objects, lots of forms

and buttons and properties. All of this in  in the browser. So, we go through a few buttons.

An example that I have been using in the hope that it will  the first advice I have is

to find a component library, material UI, React Toolbox or Blueprint. These are examples

of React's component libraries. And the things I was looking at was the extensive list of

high quality components because I don't want to spend time redoing the basic widgets. I

want to build an application, not to make a design system. And good theming support

and accessibility. And some good documentation. Because the world interface will be standing

on this component library. It better be good. I went with Material UI. It was the library

with the most extensive support. And it was allowing me to design quickly. And once I

had it, the first thing I had to get was to display a large list of objects. When they

say large list, in the case of drag and drop, a large list of objects that are living in

the game and can be a few hundred objects. So, this originally here was to use categorization.

So, for example, using react virtualized and react sortable hoc, and you can create a list.

That means that instead of having in the DOM  in the browser to have a button for 300 DOM

elements for every object, then you will only have the object that are on the screen in

the DOM. So, 10 or maybe 20 objects. The way it works is that instead of having a list

of objects that is just mapping over an array, you're converting your list to be using the

list component that is referred to by the list and using this property. And you can

say this is the function that you will call to identify every line that is to display

on the screen. And then using React sortable hoc, you can add on top of the list. And you

are using sortable elements and you get a list that is virtualized and you can reorder

and drag and drop. That was the first way to get things that  not  but working well

even with large lists and big games. And the next thing is that I wanted to have patterns.

And it was as on the screen. As I'm doing a game, people want to customize their workspace.

And so, for this I have been using a tiling window library which is called a react mosaic.

There are a few different libraries like this. But this is the one I've used and it's  I

mean, it's working out of the box. So, check it out. Then I've been trying to remake the

events. It's more or less large trees. A large trees arrays with events. An event is composed

of conditions and actions. But it can also have sub events like blocks of code in a traditional

programming language. So, here, again, the solution was to use a virtualization. Previously

I have been using React sortable tree which is a library to make a tree of nodes which

you can reorder. And the interesting thing about this library is it's using React virtualized

under the hood. So, quickly, every node that you see on the screen here is actually a line

in the list. And the lines that are in between the node is just a fake scaffolding that is

making like  that is making the tree look like a tree and another a list. So, it's a

customized version of what I have been using. And then at some point I wanted to remake 

it was the first thing that I did. And at this moment, I wanted to display the screen

so it can be composed of individual tiles and objects to be displayed. And at this moment

you can forget the DOM and go to be using WebGL. Not using WebGL directly, but a library