Dynamic Linking

备注

This documentation is somewhat outdated and is in the process of being refreshed.

Emscripten supports linking object files (and ar archives that contain object files) statically. This lets most build systems work with Emscripten with little or no changes (see 构建项目).

In addition, Emscripten also has support for a form of dynamic linking of WebAssembly modules. This can add overhead, so for best performance static linking should still be preferred. However, this overhead can can be reduced with the use of certain command line flags. See below for details.

Background

Before we get to dynamic linking, let’s talk about static linking. Emscripten’s linking model is a little different than most native platforms. To understand it, consider that native linking models work in a setting where the following facts are true:

  1. The application runs directly on the local system, and has access to local system libraries, like C and C++ standard libraries, and others.

  2. Code size is not a big concern. In part this is because the system libraries already exist on the system, so “hello world” in C++ can be small, even if it uses a large amount of iostream code in the C++ standard library. But also, code size is perhaps a matter that influences cold startup times, in that more code takes longer to load from disk, but the cost is general not significant, and modern OSes mitigate it in various ways, like caching apps they expect to be loaded.

In Emscripten’s case, code is typically going to run on the web. That means the following:

  1. The application is running in a sandbox. It has no local system libraries to dynamically link to; it must ship its own system library code.

  2. Code size is a major concern, as the application’s code is being downloaded over the internet, which is many orders of magnitude slower than an installed native app on one’s local machine.

For that reason, Emscripten automatically handles system libraries for you and automatically does dead code elimination etc. to do the best possible job it can at getting them small.

An additional factor here is that Emscripten has “js libraries” - system libraries written in JavaScript. Such system libraries are the way we access APIs on the web. It’s also a convenient way for people to connect compiled code and handwritten code on the same page. This is another reason for Emscripten to handle system libraries in a special way, and in particular, in a way that lets it strip out as much of those js libraries as it can, leaving only what is actually used, and again, that works best in the context of statically linking a standalone app with no external dependencies.

Overview of Dynamic Linking

Emscripten’s dynamic linking is fairly simple: you build several separate code “modules” from your source code, and can link them at runtime. The linking basically connects up the undefined symbols in each module with the defined symbols in the others, in the simplest of ways. It does not currently support some corner cases.

System libraries do utilize some more advanced linking features that include such corner cases. For that reason, Emscripten tries to simplify the problem as follows: There are two types of shared modules:

  1. Main modules, which have system libraries linked in.

  2. Side modules, which do not have system libraries linked in.

A project should contain exactly one main module. It can then be linked at runtime to multiple side modules. This model also makes other things simpler. For example, only the singleton main module includes the JavaScript environment and side modules are pure WebAssembly modules.

The one tricky aspect to this design is that a side module might depend on a system library that the main module did not depend on. See the section on system libraries, below, for how to handle that.

Note that the “main module” doesn’t need to contain the main() function. It could just as easily be in a side module. What makes the main module the “main” module is that there is only one main module, and only it has system libraries linked in.

(Note that system libraries are linked in to the main module statically. We still have some optimizations from doing it that way, even if we can’t dead code eliminate as well as we’d like.)

Practical Details

If you want to jump to see running code, you can look in the test suite. There are test_dylink_* tests that test dynamic linking in general, and test_dlfcn_* tests that test dlopen() specifically. Otherwise, we describe the procedure now.

Load-time Dynamic Linking

Load-time dynamic linking refers to the case when the side modules are loaded along with the main module, during startup and they are linked together before your application starts to run.

  • Build one part of your code as the main module, linking it using -sMAIN_MODULE.

  • Build other parts of your code as side modules, linking it using -sSIDE_MODULE.

For the main module the output suffix should be .js (the WebAssembly file will be generated alongside it just like normal). For the side module the output will be just a WebAssembly module we recommend the output suffix .wasm or .so (which is the shared libraries suffix used by UNIX systems).

In order to have the side modules loaded at startup you need to tell the main module about their existence. You can do this by specifying them on the command line when you link the main module. e.g.

emcc -sMAIN_MODULE main.c libsomething.wasm

At runtime, the JavaScript loading code will load libsomthing.wasm (along with any other side modules) along with the main module before the application starts to run. The running application then can access code from any of the modules linked together.

Runtime Dynamic Linking with dlopen()

Runtime dynamic linking can be performed by the calling the dlopen() function to load side modules after the program is already running. The procedure begins in the same way, with the same flags used to build the main and side modules. The difference is that you do not specify the side modules on the command line when linking the main module; instead, you must load the side module into the filesystem, so that dlopen (or fopen, etc.) can access it (except for dlopen(NULL) which means to open the current executable, which just works without filesystem integration). That’s basically it - you can then use dlopen(), dlsym(), etc. normally.

Code Size

By default, main modules disable dead code elimination. That means that all the code compiled remains in the output, including all system libraries linked in, and also all the JS library code.

That is the default behavior since it is the least surprising. But it is also possible to use normal dead code elimination, by building with -sMAIN_MODULE=2 (instead of 1). In that mode, the main module is built normally, with no special behavior for keeping code alive. It is then your responsibility to make sure that code that side modules need is kept alive. You can do this either by adding to EXPORTED_FUNCTIONS or tagging the symbol EMSCRIPTEN_KEEPALIVE in the source code. See other.test_minimal_dynamic for an example of this in action.

If you are doing load time dynamic linking then any symbols needed by the side modules specified on the command line will be kept alive automatically. For this reason we strongly recommend using MAIN_MODULE=2 when doing load time dynamic linking.

There is also the corresponding -sSIDE_MODULE=2 for side modules.

System Libraries

As mentioned earlier, system libraries are handled in a special way by the Emscripten linker, and in dynamic linking, only the main module is linked against system libraries. When linking the main module it is possible to pass the side modules on the command line, in which case any system library dependencies are automatically handled.

However when linking a main module without its side modules (Usually with -sMAIN_MODULE=1) it is possible that required system libraries are not included. This section explains what to do to fix that by forcing the main module to be linked against certain libraries.

You can build the main module with EMCC_FORCE_STDLIBS=1 in the environment to force inclusion of all standard libs. A more refined approach is to name the system libraries that you want to explicitly include. For example, with something like EMCC_FORCE_STDLIBS=libcxx,libcxxabi (if you need those two libs).

Miscellaneous Notes

Dynamic Checks

Native linkers generally only run code when all symbols are resolved. Emscripten’s dynamic linker hooks up symbols to unresolved references to those symbols dynamically. As a result, we don’t check if any symbols remain unresolved, and code can start to run even if there are. It will run successfully if they are not called in practice. If they are, you will get a runtime error. What went wrong should be clear from the stack trace (in an unminified build); building with -sASSERTIONS can help some more.

Limitations

  • Chromium does not support compiling >4kB WASM on the main thread, and that includes side modules; you can use --use-preload-plugins (in emcc or file_packager.py) to make Emscripten compile them on startup [doc] [discuss].

  • EM_ASM code defined within side modules depends on eval support are is therefore incompatible with -sDYNAMIC_EXECUTION=0.

  • EM_JS functions defined in side modules are not yet supported.

Pthreads support

Dynamic linking + pthreads is is still experimental. As such, linking with both MAIN_MODULE and -pthread will produce a warning.

While load-time dynamic linking works without any complications, runtime dynamic linking via dlopen/dlsym can require some extra consideration. The reason for this is that keeping the indirection function pointer table in sync between threads has to be done by emscripten library code. Each time a new library is loaded or a new symbol is requested via dlsym, table slots can be added and these changes need to be mirrored on every thread in the process.

Changes to the table are protected by a mutex, and before any thread returns from dlopen or dlsym it will wait until all other threads are sync. In order to make this synchronization as seamless as possible, we hook into the low level primitives of emscripten_futex_wait and emscirpten_yield.

For most use cases all this happens under hood and no special action is needed. However, there there is one class of application that currently may require modification. If your applications busy waits, or directly uses the atomic.waitXX instructions (or the clang __builtin_wasm_memory_atomic_waitXX builtins) you maybe need to switch it to use emscripten_futex_wait or order avoid deadlocks. If you don’t use emscripten_futex_wait while you block, you could potentially block other threads that are calling dlopen and/or dlsym.