README-emscripten.md

Emscripten

The state of things

(As of October 2024, but things move quickly and we don't update this document often.)

In modern times, all the browsers you probably care about (Chrome, Firefox, Edge, and Safari, on Windows, macOS, Linux, iOS and Android), support some reasonable base configurations:

• WebAssembly (don't bother with asm.js any more)
• WebGL (which will look like OpenGL ES 2 or 3 to your app).
• Threads (see caveats, though!)
• Game controllers
• Autoupdating (so you can assume they have a recent version of the browser)

All this to say we're at the point where you don't have to make a lot of concessions to get even a fairly complex SDL-based game up and running.

RTFM

This document is a quick rundown of some high-level details. The documentation at emscripten.org is vast and extremely detailed for a wide variety of topics, and you should at least skim through it at some point.

Porting your app to Emscripten

Many many things just need some simple adjustments and they'll compile like any other C/C++ code, as long as SDL was handling the platform-specific work for your program.

First: assembly language code has to go. Replace it with C. You can even use x86 SIMD intrinsic functions in Emscripten!

Second: Middleware has to go. If you have a third-party library you link against, you either need an Emscripten port of it, or the source code to it to compile yourself, or you need to remove it.

Third: If your program starts in a function called main(), you need to get out of it and into a function that gets called repeatedly, and returns quickly, called a mainloop.

Somewhere in your program, you probably have something that looks like a more complicated version of this:

void main(void)
{
    initialize_the_game();
    while (game_is_still_running) {
        check_for_new_input();
        think_about_stuff();
        draw_the_next_frame();
    }
    deinitialize_the_game();
}

This will not work on Emscripten, because the main thread needs to be free to do stuff and can't sit in this loop forever. So Emscripten lets you set up a mainloop.

static void mainloop(void)   /* this will run often, possibly at the monitor's refresh rate */
{
    if (!game_is_still_running) {
        deinitialize_the_game();
        #ifdef __EMSCRIPTEN__
        emscripten_cancel_main_loop();  /* this should "kill" the app. */
        #else
        exit(0);
        #endif
    }

    check_for_new_input();
    think_about_stuff();
    draw_the_next_frame();
}

void main(void)
{
    initialize_the_game();
    #ifdef __EMSCRIPTEN__
    emscripten_set_main_loop(mainloop, 0, 1);
    #else
    while (1) { mainloop(); }
    #endif
}

Basically, emscripten_set_main_loop(mainloop, 0, 1); says "run mainloop over and over until I end the program." The function will run, and return, freeing the main thread for other tasks, and then run again when it's time. The 1 parameter does some magic to make your main() function end immediately; this is useful because you don't want any shutdown code that might be sitting below this code to actually run if main() were to continue on, since we're just getting started.

Another option is to use SDL' main callbacks, which handle this for you without platform-specific code in your app. Please refer to the wiki or docs/README-main-functions.md in the SDL source code.

There's a lot of little details that are beyond the scope of this document, but that's the biggest initial set of hurdles to porting your app to the web.

Do you need threads?

If you plan to use threads, they work on all major browsers now. HOWEVER, they bring with them a lot of careful considerations. Rendering must be done on the main thread. This is a general guideline for many platforms, but a hard requirement on the web.

Many other things also must happen on the main thread; often times SDL and Emscripten make efforts to "proxy" work to the main thread that must be there, but you have to be careful (and read more detailed documentation than this for the finer points).

Even when using threads, your main thread needs to set an Emscripten mainloop (or use SDL's main callbacks) that runs quickly and returns, or things will fail to work correctly.

You should definitely read Emscripten's pthreads docs for all the finer points. Mostly SDL's thread API will work as expected, but is built on pthreads, so it shares the same little incompatibilities that are documented there, such as where you can use a mutex, and when a thread will start running, etc.

IMPORTANT: You have to decide to either build something that uses threads or something that doesn't; you can't have one build that works everywhere. This is an Emscripten (or maybe WebAssembly? Or just web browsers in general?) limitation. If you aren't using threads, it's easier to not enable them at all, at build time.

If you use threads, you have to run from a web server that has COOP/COEP headers set correctly or your program will fail to start at all.

If building with threads, __EMSCRIPTEN_PTHREADS__ will be defined for checking with the C preprocessor, so you can build something different depending on what sort of build you're compiling.

Audio

Audio works as expected at the API level, but not exactly like other platforms.

You'll only see a single default audio device. Audio recording also works; if the browser pops up a prompt to ask for permission to access the microphone, the SDL_OpenAudioDevice call will succeed and start producing silence at a regular interval. Once the user approves the request, real audio data will flow. If the user denies it, the app is not informed and will just continue to receive silence.

Modern web browsers will not permit web pages to produce sound before the user has interacted with them (clicked or tapped on them, usually); this is for several reasons, not the least of which being that no one likes when a random browser tab suddenly starts making noise and the user has to scramble to figure out which and silence it.

SDL will allow you to open the audio device for playback in this circumstance, and your audio callback will fire, but SDL will throw the audio data away until the user interacts with the page. This helps apps that depend on the audio callback to make progress, and also keeps audio playback in sync once the app is finally allowed to make noise.

There are two reasonable ways to deal with the silence at the app level: if you are writing some sort of media player thing, where the user expects there to be a volume control when you mouseover the canvas, just default that control to a muted state; if the user clicks on the control to unmute it, on this first click, open the audio device. This allows the media to play at start, and the user can reasonably opt-in to listening.

Many games do not have this sort of UI, and are more rigid about starting audio along with everything else at the start of the process. For these, your best bet is to write a little Javascript that puts up a "Click here to play!" UI, and upon the user clicking, remove that UI and then call the Emscripten app's main() function. As far as the application knows, the audio device was available to be opened as soon as the program started, and since this magic happens in a little Javascript, you don't have to change your C/C++ code at all to make it happen.

Please see the discussion at #6385 for some Javascript code to steal for this approach.

Rendering

If you use SDL's 2D render API, it will use GLES2 internally, which Emscripten will turn into WebGL calls. You can also use OpenGL ES 2 directly by creating a GL context and drawing into it.

If the browser (and hardware) support WebGL 2, you can create an OpenGL ES 3 context.

Calling SDL_RenderPresent (or SDL_GL_SwapWindow) will not actually present anything on the screen until your return from your mainloop function.

Building SDL/emscripten

SDL currently requires at least Emscripten 3.16.0 to build. Newer versions are likely to work, as well.

Build:

This works on Linux/Unix and macOS. Please send comments about Windows.

Make sure you've installed emsdk first, and run source emsdk_env.sh at the command line so it finds the tools.

(These cmake options might be overkill, but this has worked for me.)

mkdir build
cd build
emcmake cmake ..
# you can also do `emcmake cmake -G Ninja ..` and then use `ninja` instead of this command.
emmake make -j4

If you want to build with thread support, something like this works:

mkdir build
cd build
emcmake cmake -DSDL_THREADS=On ..
# you can also do `emcmake cmake -G Ninja ..` and then use `ninja` instead of this command.
emmake make -j4

To build the tests, add -DSDL_TESTS=On to the emcmake cmake command line. To build the examples, add -DSDL_EXAMPLES=On to the emcmake cmake command line.

Building your app

You need to compile with emcc instead of gcc or clang or whatever, but mostly it uses the same command line arguments as Clang.

Link against the libSDL3.a file you generated by building SDL.

Usually you would produce a binary like this:

gcc -o mygame mygame.c  # or whatever

But for Emscripten, you want to output something else:

emcc -o index.html mygame.c

This will produce several files...support Javascript and WebAssembly (.wasm) files. The -o index.html will produce a simple HTML page that loads and runs your app. You will (probably) eventually want to replace or customize that file and do -o index.js instead to just build the code pieces.

If you're working on a program of any serious size, you'll likely need to link with -s ALLOW_MEMORY_GROWTH=1 -s MAXIMUM_MEMORY=1gb to get access to more memory. If using pthreads, you'll need the -s MAXIMUM_MEMORY=1gb or the app will fail to start on iOS browsers, but this might be a bug that goes away in the future.

Data files

Your game probably has data files. Here's how to access them.

Filesystem access works like a Unix filesystem; you have a single directory tree, possibly interpolated from several mounted locations, no drive letters, '/' for a path separator. You can access them with standard file APIs like open() or fopen() or SDL_IOStream. You can read or write from the filesystem.

By default, you probably have a "MEMFS" filesystem (all files are stored in memory, but access to them is immediate and doesn't need to block). There are other options, like "IDBFS" (files are stored in a local database, so they don't need to be in RAM all the time and they can persist between runs of the program, but access is not synchronous). You can mix and match these file systems, mounting a MEMFS filesystem at one place and idbfs elsewhere, etc, but that's beyond the scope of this document. Please refer to Emscripten's page on the topic for more info.

The easiest (but not the best) way to get at your data files is to embed them in the app itself. Emscripten's linker has support for automating this.

emcc -o index.html loopwave.c --embed-file ../test/sample.wav@/sounds/sample.wav

This will pack ../test/sample.wav in your app, and make it available at "/sounds/sample.wav" at runtime. Emscripten makes sure this data is available before your main() function runs, and since it's in MEMFS, you can just read it like you do on other platforms. --embed-file can also accept a directory to pack an entire tree, and you can specify the argument multiple times to pack unrelated things into the final installation.

Note that this is absolutely the best approach if you have a few small files to include and shouldn't worry about the issue further. However, if you have hundreds of megabytes and/or thousands of files, this is not so great, since the user will download it all every time they load your page, and it all has to live in memory at runtime.

Emscripten's documentation on the matter gives other options and details, and is worth a read.

Debugging

Debugging web apps is a mixed bag. You should compile and link with -gsource-map, which embeds a ton of source-level debugging information into the build, and make sure the app source code is available on the web server, which is often a scary proposition for various reasons.

When you debug from the browser's tools and hit a breakpoint, you can step through the actual C/C++ source code, though, which can be nice.

If you try debugging in Firefox and it doesn't work well for no apparent reason, try Chrome, and vice-versa. These tools are still relatively new, and improving all the time.

SDL_Log() (or even plain old printf) will write to the Javascript console, and honestly I find printf-style debugging to be easier than setting up a build for proper debugging, so use whatever tools work best for you.

Questions?

Please give us feedback on this document at the SDL bug tracker. If something is wrong or unclear, we want to know!