The Emscripten runtime environment is different to that expected by most C/C++ applications. Emscripten works hard to abstract and mitigate these differences, so that in general code can be compiled with little or no change.
This article expands on some of the differences and the resulting API Limitations, and outlines the few changes you may need to make to your C/C++ code.
Emscripten implements the Simple DirectMedia Layer API (SDL) for the browser environment, which provides low level access to audio, keyboard, mouse, joystick, and graphics hardware. Applications that use SDL typically require no input/output changes to run in the browser.
In addition, we have more limited support for glut, glfw, glew and xlib.
Applications that do not use SDL or the other APIs can use the Emscripten-specific APIs for input and output:
A lot of C/C++ code uses the synchronous file system APIs in libc and libcxx to access code in the local file system. This is problematic because the browser prevents code from directly accessing files on the host system, and because JavaScript only supports asynchronous file access outside of web workers.
Emscripten provides an implementation of libc and libcxx and a virtual file system so that normal C/C++ code can be compiled and run without change. Most developers need only specify the set of files to be packaged for preloading into the virtual file system at runtime.
备注
Using a virtual file system bypasses the limitations listed above. The file data is packaged at compile time and downloaded into the file system using asynchronous JavaScript APIs before the compiled code is allowed to run. The compiled code then makes “file” calls that are really just calls into program memory.
The default file system (MEMFS) stores files in-memory, so that any changes are lost when the page is reloaded. If file changes need to be stored more permanently then developers can mount the IDBFS file system, which allows data to be persisted in the browser. When running code in node.js, developers can mount NODEFS to give code direct access to the local file system.
Emscripten also has an API to support asynchronous file access.
For more information and examples see 文件和文件系统.
The browser event model uses co-operative multitasking — each event has a “turn” to run, and must then return control to the browser so that other events can be processed. A common cause of HTML pages hanging is JavaScript that does not complete and return control to the browser.
Graphical C++ apps typically run in an infinite loop. Within each iteration of the loop the app performs event handling, processing and rendering, followed by a delay (“wait”) to keep the frame rate constant. This infinite loop is a problem in the browser environment because there is no way for control to return to the browser so other code can run. After a period the browser will notify the user that the page is stuck and offer to halt or close it.
Similarly, JavaScript APIs like WebGL can only run when the current “turn” is over, and will automatically render and swap buffers at that point. This contrasts with OpenGL C++ apps where you would need to swap the buffers manually.
The standard solution for this problem is to define a C function that performs one iteration of your main loop (not including the “delay”). For a native build this function can be called in an infinite loop, leaving the behaviour effectively unchanged.
Within Emscripten compiled code we use
emscripten_request_animation_frame_loop() to get the environment to call
this same function at the proper frequency for rendering a frame (that is, if
the browser renders at 60fps, it will call this 60 times per second). The
iteration is still run “infinitely” but now other code can run between
iterations and the browser does not hang.
Typically you will have a small section with #ifdef __EMSCRIPTEN__ for the two cases. For example:
#include <emscripten.h>
#include <emscripten/html5.h>
#include <stdio.h>
// Our "main loop" function. This callback receives the current time as
// reported by the browser, and the user data we provide in the call to
// emscripten_request_animation_frame_loop().
EM_BOOL one_iter(double time, void* userData) {
// Can render to the screen here, etc.
puts("one iteration");
// Return true to keep the loop running.
return EM_TRUE;
}
int main() {
#ifdef __EMSCRIPTEN__
// Receives a function to call and some user data to provide it.
emscripten_request_animation_frame_loop(one_iter, 0);
#else
while (1) {
one_iter();
// Delay to keep frame rate constant (using SDL).
SDL_Delay(time_to_next_frame());
}
#endif
}
备注
A more full-featured API is provided in
emscripten_set_main_loop(), which lets you specify the
frequency at which to call the function, and other things.
备注
are just rendering a single frame and halting. You should also note:
The current Emscripten implementation of SDL_QUIT will work if you use emscripten_set_main_loop(). As the page is shut, it will force a final direct call to the main loop, giving it a chance to notice the SDL_QUIT event. If you do not use a main loop, your app will close before you have had an opportunity to notice this event.
There are limitations to what you can do as the page shuts (in onunload). Some actions like showing alerts are banned by browsers at this point.
Another option is to use Asyncify which will
rewrite the program so that it can return to the browser’s main event loop
by just calling emscripten_sleep(). Note that this rewriting causes size and speed overhead
while emscripten_request_animation_frame_loop / emscripten_set_main_loop
as described earlier do not.
When an Emscripten-compiled application is loaded, it starts by preparing data in the preloading phase. Files you marked for preloading (using emcc --preload-file, or manually from JavaScript with FS.createPreloadedFile()) are set up at this stage.
You can add additional operations with addRunDependency(), which is a counter of all dependencies to be executed before compiled code can run. As these are completed you can call removeRunDependency() to remove the completed dependencies.
备注
Generally it is not necessary to add additional operations — preloading is suitable for almost all use cases.
When all dependencies are met, Emscripten will call run(), which proceeds to call your main() function. The main() function should be used to perform initialization tasks, and will often call emscripten_set_main_loop() (as described above). The main loop function will be then be called at the requested frequency.
You can affect the operation of the main loop in several ways:
emscripten_push_main_loop_blocker() adds a function that blocks the main loop until the blocker completes.
This is useful, for example, to manage loading new game levels. After a level completes, you can push blockers for each action involved (unpacking the file, generate the data structures, etc.) When all the blockers have completed the main loop will resume and the game should run the new level. You can also use this function in conjunction with emscripten_set_main_loop_expected_blockers() to keep the user informed of progress.
emscripten_pause_main_loop() pauses the main loop, and emscripten_resume_main_loop() resumes it. These are low level (less recommended) alternatives to the blocker functions.
emscripten_async_call() lets you call a function after some specific interval. This will use requestAnimationFrame (by default) or setTimeout if a specific interval was requested.
The browser execution environment reference (emscripten.h) describes a number of other methods for controlling execution.
In both asm.js and WebAssembly Emscripten represents memory in a way similar to native architectures. Pointers represent offsets into memory, structs use the same amount of address space as they normally do, and so forth.
In WebAssembly, this is done using a WebAssembly.Memory which was designed for that purpose. In asm.js, Emscripten uses a single typed array, with different views providing access to different types (HEAPU32 for 32-bit unsigned integers, etc.).
Emscripten experimented with other memory representations in the past, ending up on the “typed arrays mode 2” approach for JS and then asm.js, as described above, and then WebAssembly implemented something similar.