tomato-testing/docs/README-emscripten.md
Green Sky 644725478f Squashed 'external/sdl/SDL/' changes from ec0042081..399bc709b
399bc709b build-scripts.pl: Added add-source-to-projects.pl
ac6827187 Visual-WinRT: dos2unix the project files to match other Visual Studio targets.
34719cba9 Fixed crash in hid_init() if the HIDDeviceManager isn't available
2e92e94eb Make sure we update device->sample_frames in SDL_AudioDeviceFormatChangedAlreadyLocked()
9964e5c5b wayland: Don't retrieve the drag offer strings with every pointer motion event
bac7eeaaa Added missing include
a541e2ac1 audio: Change a few SDL_memcpy calls to SDL_copyp.
54125c140 audio: Only update bound audiostreams' formats when necessary.
e0b0f9a36 testaudio: Fix mouseover testing.
2f3deec24 wayland: Don't process drag & drop events from surfaces not owned by SDL
42bdced05 events: Log file drop position events and print the pointer coordinates
c10d93d3a wayland: Replace magic constant with define
500852153 emscripten: Restore compatibility with existing emsdk releases.
953b55dd6 Use EM_ASM_PTR when the return value is a pointer
a4541a255 audio: SDL_GetAudioStreamQueued now returns bytes, not frames.
703aefbce Sync SDL3 wiki -> header
99421b64d linux: Add portal drag and drop
952c5059b Remove stray Â
eebd5d18a linux: Handle upower's UP_DEVICE_STATE_PENDING_CHARGE, PENDING_DISCHARGE
f8fdb20d8 audio: Destroy all existing SDL_AudioStreams on shutdown.
62d445997 audio: Removed declarations of functions that don't exist anymore.
b656720bc loopwave: Use SDL_GetAudioStreamQueued() for more accurate results.
34b931f7e audio: Added SDL_GetAudioStreamQueued
23206b9e3 audio: Added SDL_EVENT_AUDIO_DEVICE_FORMAT_CHANGED
c7e6d7a1f audio: Changed debug logging output.
87ec6acf2 audio: Added a FIXME
ac88ffb7e audio: don't allocate buffer in SDL_SetAudioPostmixCallback for NULL callback.
2a950f6ae audio: Replace some SDL_memcpy calls with SDL_copyp.
0dc0434a3 audio: Fixed race condition in subsystem shutdown.
23f60203a audio: precalculate if we can use simple copies instead of the full mixer.
36b0f1141 audio: Optimize setting device formats during audio thread iteration.
4c3e84897 testspriteminimal: make standalone by embedding icon.bmp
2a01f9dcb tests: plug leaks when running with --trackmem
f42bbeca2 SDL_test: track stack frames of allocations on Windows
12c0be028 SDL_test: clear text cache on exit event
b4bfb1831 SDL_test: free state before logging allocations
248b1edd3 SDL_test: destroy windows in SDL_CommonQuit
98da2dd30 SDL_test: don't warn about expected allocations when running with --trackmem
6a381567b Support audio rate conversion up to 384KHz
b2b548a1f Don't hang if IAudioRenderClient_GetBuffer() fails indefinitely
a3a5e1728 Fixed build warning '=': conversion from 'Uint32' to 'Uint16', possible loss of data
6d3e21c27 Fixed android build warnings
fca2f5318 Fixed warning: this function declaration is not a prototype
a72dfa6a5 Fixed sensor timestamp units for third-party PS5 controllers
f6756047a Fixed error: array subscript 2 is above array bounds of ‘const Uint8[2]’
7059a55cc Fixed sensor timestamp calculation for third-party PS5 controllers
c0443e5d1 Fixed crash in SDL_IMMDevice_FindByDevID()
fde8499f6 Use around 20ms for the audio buffer size
e5739d7d1 video: Remove SDL_GetFocusWindow()
39c2f9737 Fix NULL dereference in SDL_OpenAudio
9a23d0e3f Added new audio files to the Xcode project
a62e62f97 Refactored SDL_audiocvt.c
31229fd47 include: Added a note about SDL's iOS app delegate functions.
65aaf3a9a x11: Always update clipboard owner
f622f21e6 Fixed build
5774c9638 Prefer hidraw over libusb when libusb whitelisting is not enabled
9301f7ace hidapi/libusb: only enumerate each interface once
859dc14ad Replaced SDL_GetGamepadBindForAxis() and SDL_GetGamepadBindForButton() with SDL_GetGamepadBindings()
9e50048ab Revert "Removed SDL_GamepadBinding from the API"
9f17d1a9d Don't reference the same function in "see also"
86505ea63 fix SDL_AudioStreamCallback documentation
d885d5c31 Sync SDL3 wiki -> header
2f43f7bc5 audio: Allow querying of device buffer size.
cf9572113 audio: Added a hint to let apps force device buffer size.
47d8c77c6 audio: Choose better default sample frame counts.
8b26e95f9 audio: Change SDL_AudioStreamCallback
9da34e8fb docs: Updated README-emscripten.md.
fd1c54a00 detect fanatec steering wheels
cb4414608 docs: Whoops, this got added by the wiki bridge by accident!
cd633b9a8 Renamed SDL_IsAudioDevicePaused() to SDL_AudioDevicePaused()
c6cad07fa Sync SDL3 wiki -> header
a6e52f9e4 Sync SDL3 wiki -> header
2de2e9d03 Fix flickering of window when using desktop-fullscreen and borderless window on multiple monitors on Linux.  Closes #8186.
723835d16 Windows: fix for client rect resizing larger each time we came from exclusive fullscreen -> windowed on a monitor with HiDPI set.  The problem was we were using the monitor DPI rather than the window DPI so AdjustWindowRectExForDpi was giving us an incorrect size which would be too large for the client rect.  Closes #8237.
ce27363df wikiheaders: Sort undocumented functions.
e22282b09 Added README about transparent windows in Win32
1d1c6e630 Turn off COREAUDIO debug logging by default
52efefca0 wayland: Fix drag offer leak
3a992af44 audio: Added a postmix callback to logical devices.
7207bdce5 render: Enable clipping for zero-sized rectangles
22d81fb3e cmake: use MSVC_RUNTIME_LIBRARY to force MT
a2e17852d cmake: make sure SDL_GetPrefPath is run before testfilesystem
2fb266e0a ci: run tests in parallel
ad1313e75 testaudio: Patched to compile.
5747ddc01 testaudio: Clean up some messy memory management.
fafbea1ce audio: Move internal float32 mixing to a simplified function
116b0ec97 include: minor tweak to audio API documentation
fb1377035 include: Replaced old Bugzilla URL.
38c8fc05c audio: Remove ChooseMixStrategy.
b00cbd76a wikiheaders.pl: create Unsupported.md file with list of functions undocumented in either the headers or the wiki
37e1fc3b5 wayland: Ensure that the toplevel window is recreated when switching decoration modes
f2ca9a615 Added SDL_AUDIO_FRAMESIZE
53122593f Added SDL_AUDIO_BYTESIZE
544351c98 Sync SDL3 wiki -> header
2e7d2b94e Clarify that SDL_BlitSurface() ignores the width and height in dstrect
a2c1984d3 Detect Simagic wheel bases as wheels (#8198)
1d8dfbb22 avoid type redefinition errors after PR/8181
266b91d2f Detect Logitech G923 Playstation as wheel G923 have two different versions - Xbox version is already present in the wheel list, but not the PS version.
cde67ea49 Detect Logitech PRO Racing Wheel for Xbox (PC mode) as wheel Logitech PRO Racing Wheel have two different versions - for Playstation and Xbox. Vendor + Product ID for Playstation version already present in SDL sources, but not an Xbox version
3a932141e Restore audio format binary compatibility with SDL 2.0
e85206ffd wikiheaders.pl: add --rev= option to pass revision string
233789b0d Audio types have the same naming convention as other SDL endian types, e.g. [S|U][BITS][LE|BE]
36b5f3e35 Sync SDL3 wiki -> header
0e552761b Renamed AudioStreamSpeed to AudioStreamFrequencyRatio
47bcb078f Fixed some incorrect SDL_AUDIO_F32 uses
2833f2e7b Fixed OOB access in audio_convertAccuracy test
8387fae69 Sync SDL3 wiki -> header
832181345 docs: Add note about Wayland application icons
825d34475 Make sure that the same timestamp is used for all PS5 events from the same packet
9c1430324 Removed SDL_dataqueue
28b28bd8f Added audio_formatChange test
a59152688 Try and avoid overflow when handling very large audio streams
5394a805f Improved testaudiostreamdynamicresample
e55844274 Added SDL_(Get|Set)AudioStreamSpeed
43c3c5736 Track the formats of data in an SDL_AudioStream
337fed3df Tweaked ResampleFrame_SSE Use _mm_unpack(lo|hi)_ps instead of _mm_shuffle_ps
fd7cd91dc audio: Mix multiple streams in float32 to prevent clipping.
9097573e3 audio: Choose a mixing strategy on each iteration.
bbe2e012a Don't provide the SDL3 header path
c17a35f09 Fixed typo
4f72255eb Fixed README.md link
e0ab59754 Simplified SDL_main.h migration notes
d44bde61e Added SDL migration information to the top level README.md
6ff31e10c metal: Add hint to select low power device instead of the default one (#8182)
8a8aed477 Make sure that we process touch events that position the mouse
f84c87f20 Sync SDL3 wiki -> header
a7eea9997 macOS: Don't raise the parent top-level window when raising a child window, only raise the child window to the top of the parent
a5e721479 Add SDL_WINDOW_NOT_FOCUSABLE flag to set that the window should not be able to gain key focus
b385dc3b6 n3dsaudio: Patched to compile.
4e0c7c91f audio: PlayDevice() should return an error code.
a94d724f1 wayland: add SDL_VIDEO_DRIVER_WAYLAND_DYNAMIC_EGL
da5d93d3d wayland: don't define SDL_VIDEO_DRIVER_WAYLAND_DYNAMIC_* macro's
f002f7d12 ci: build emscripten with Debug buid type
3699b12ed audio: Fixed some "is_*" variables to be cleaner and/or more specific.
2471d8cc2 audio: Fixed logic error in SDL_OpenAudioDeviceStream.
1b03a2430 testsurround: fix order of arguments of callback
82db2b58f Renamed audio stream callback and moved the userdata parameter first
5bdad5210 Sync SDL3 wiki -> header
58c859f64 audio: Rename SDL_GetAudioStreamBinding to SDL_GetAudioStreamDevice.
efd2023a7 audio: Fixed documentation.
1e775e0ee audio: Replace SDL_CreateAndBindAudioStream with SDL_OpenAudioDeviceStream.
bd088c2f9 Revert "Clarify whether an audio function expects a physical or logical device ID"
82e481b52 Added --randmem test parameter
ea68bb802 Add some additional checks to audio_convertAudio
f8286df16 Fixed ResampleFrame_SSE doing unnecessary work
b1d63be53 Fixed audio_resampleLoss test
c191d6c30 Better Win32 transparent window support
923d612ca hidapi: sync macOS code with mainstream.
363f4fa9c avoid type redefinition errors after commit ee806597b9.
615824a80 Updated documentation now that SDL_GetAudioDevices() has been split into separate functions for output and capture devices
506a133d8 Clarify whether an audio function expects a physical or logical device ID
3b1d1e4e3 hidapi: sync the hidraw changes with mainstream
f617918e0 cmake: check linkage to libusb too, instead of libusb.h presence only.
041dbd6b5 Fixed GetResamplerAvailableOutputFrames Non-euclidean division is a pain
b49d0a607 x11: Avoid including full Vulkan headers.
4d2f9f3a3 yuv_rgb: Comment out unused code.
3c3486e2a wayland: Don't include full Vulkan headers when not necessary.
f066bbe98 x11: Don't include system headers twice.
d86d02bbb updated dynapi after SDL_GDKGetDefaultUser addition
4355f9cec Fixed warning C4389: '!=': signed/unsigned mismatch
5755de07a Fixed build warnings
0f80d47bb Fixed thread-safety warning
ee806597b Removed SDL_vulkan_internal.h from SDL_sysvideo.h
34860b932 Fixed testautomation --filter pixels_allocFreeFormat
6f8a6a31c gdk: GetBasePath should be a UTF8 version of Win32 GetBasePath
e30e5c77e Sync SDL3 wiki -> header
c0cd8c814 gdk: Add SDL_GDKGetDefaultUser, SDL_GetPrefPath implementation
106abce69 Refactored GetAudioStreamDataInternal buffer handling The final conversion step should now always go straight into the output buffer.
e44f54ec5 Avoid using hex-floats
5b696996c Added ResampleFrame_SSE
958b3cfae Tweaked and enabled audio_convertAudio test
7dbb9b65b audio_convertAccuracy: Shuffle the data in case of a bad SIMD implementation
f6a4080ff audio_resampleLoss: Add support for multiple channels
4f894e748 audio_resampleLoss: SDL_GetAudioStreamData now returns the correct length
ab83f75bb Make sure GetAudioStreamDataInternal is called with a valid length
6a73f74b6 Rebuild full ResamplerFilter (left wing + right wing) at runtime
0c15ce006 Add a missing int cast
b74ee86b1 Optimized ResampleAudio, with special cases for 1 and 2 channels This would also benefit from some SIMD, since it's just a bunch of multiply-adds
fba6e1e3d Removed ResamplerFilterDifference It takes 1 extra multiply to calculate the correct interpolation, but I think the improvement in cache locality (and binary size) outweighs that.
9f7a22fa4 Removed 64-bit handling from AudioConvertByteswap
1f5327a9f Removed future_buffer, left_padding, and right_padding from SDL_AudioStream
71ad52d6d Lowered SDL_GetAudioStreamData to 32 KB No particular reason for this number, but 1 MB was a bit silly
69aec8c91 Fixed the report format for the Razer Wolverine V2 Pro
7c2669c9d Accept key events from any source
1e9d31448 Updated to Android minSdkVersion 19 and targetSdkVersion 34
8924d0d92 Added missing function prototype for SDL_WriteS64BE()
845f3c745 Fixed mismatch between stdlib calloc() and SDL free()
fb7921173 emscriptenaudio: Fire the capture silence_callback at an interval.
5191b2054 emscriptenaudio: Don't bother undefining things about to be unreachable.
fd75a4ca0 emscriptenaudio: Deal with blocked audio devices better.
981b8a337 emscriptenaudio: Remove unnecessary functions.
c7588e426 Transparent window for Win32 + OpenGL (#8143)
f9581178d cmake: fixed a typo.
e6c878824 Fixed ResampleAudio interpolation factor calculation
498363863 Misc audio tweaks/cleanup
72d9d53de Invert the inner ResampleAudio loops to avoid doing unnecessary work
88123a510 The history buffer should always have the maximum possible padding frames
96e47f165 Clamp results of GetResampler(AvailableOutput|NeededInput)Frames
d2b9c8b80 Fixed maths in testaudiostreamdynamicresample (and just show the actual scale)
14e38b17d Removed assertions from inner ResampleAudio loop
9d413dfdc The history buffer doesn't need to be so large
2788e848f Allow resampling less than 1 frame of input
383084e0a Pre-calculate resampling rate, and use it instead of .freq in most places
40a6a445c Update resample_offset inside ResampleAudio
47fea7f06 Used fixed-point arithmetic in ResampleAudio
7bb4e806e Clear resample_offset in SDL_ClearAudioStream, not SetAudioStreamFormat Not entirely sure if ClearAudioStream is the right place, but SetAudioStreamFormat was the wrong place
b9541b9ea Improved ResampleAudio * filterindex2 was off-by-one * Generate ResamplerFilter using doubles * Transpose ResamplerFilter to improve access patterns
cdaa19869 Track offset within the current sample when resampling
d60ebb06d mouse: Ensure that the dummy default cursor is removed from the cursor list
e58c2731f mouse: Free the default cursor when destroyed
789ce17e1 audio: Don't resample in chunks for now.
cbab33482 audio: Don't call SDL_AudioStream callbacks for empty data sets.
3e1ae0c86 Clearified the libusb whitelist default logic
f4520821e Removed some unnecessary integer casts
0989b7e86 Avoid using designated initializers
c6c1e673c Optimized SDL_Convert_*_to_*_Scalar
f97b920b3 Optimized SDL_Convert_*_to_*_SSE2 Some of the SDL_Convert_F32_to_*_SSE2 do not explicitly clamp the input, but instead rely on saturating casts. Inputs very far outside the valid [-1.0, 1.0] range may produce an incorrect result, but I believe that is an acceptable trade-off.
300d1ec3e Added audio_convertAccuracy test
32cecc2ea Fixed assertion in audio_convertAudio
33f11e21e Removed assertions in AudioConvert(To|From)Float
c2f388fd8 cmake: add SDL_HIDAPI_LIBUSB_SHARED option + test on ci
371cc2d17 wayland: Remove unnecessary flag and state settings
fe85e6e75 cocoa: Send a maximized event instead of restored if a deminiaturized window is zoomed
ddddcb78c cocoa: Use the close method to hide a miniaturized window
be8c42cfd Clarify that a window being 'hidden' means that it is unmapped/ordered out
a44338cbc Fix typo in SDL_audiocvt.c
f464eb2c5 SDL_hidapi.c: change 'use_libusb_whitelist_default' into a macro.
6607a3cfa Disable cache in python http server
181d5d285 hidapi: Enable libusb support by default.
f0f15e365 hidapi: Use a whitelist for libusb when other backends are available
c3f7a7dc4 Convert audio using SDL_AUDIO_F32SYS format instead of SDL_AUDIO_F32
796713b9d xxd.py: always write \n line endings
723bcd0a8 SDL_TriggerBreakppoint for riscv arch (both 32/64) version.

git-subtree-dir: external/sdl/SDL
git-subtree-split: 399bc709b7485bab57880f8261f826f29dc0d7b2
2023-09-23 18:45:49 +02:00

14 KiB

Emscripten

The state of things

(As of September 2023, 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, you probably need this in at least one of your source files:

#ifdef __EMSCRIPTEN__
#include <emscripten.h>
#endif

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

Third: 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.

Fourth: You still start in a function called main(), but 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.

There's a lot of little details that are beyond the scope of this document, but that's the biggest intial 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 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 capture 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 https://github.com/libsdl-org/SDL/issues/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.

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

First: do you really need to build SDL from source?

If you aren't developing SDL itself, have a desire to mess with its source code, or need something on the bleeding edge, don't build SDL. Just use Emscripten's packaged version!

Compile and link your app with -sUSE_SDL=2 and it'll use a build of SDL packaged with Emscripten. This comes from the same source code and fixes the Emscripten project makes to SDL are generally merged into SDL's revision control, so often this is much easier for app developers.

-sUSE_SDL=1 will select Emscripten's JavaScript reimplementation of SDL 1.2 instead; if you need SDL 1.2, this might be fine, but we generally recommend you don't use SDL 1.2 in modern times.

If you want to build SDL, though...

SDL currently requires at least Emscripten 3.1.35 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.

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 SDL/build/libSDL3.a file you generated by building SDL, link with -sUSE_SDL=2 to use Emscripten's prepackaged SDL2 build.

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 -sALLOW_MEMORY_GROWTH=1 -sMAXIMUM_MEMORY=1gb to get access to more memory. If using pthreads, you'll need the -sMAXIMUM_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_RWops. 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!