![]() | ![]() |
|
C++ dlopen mini HOWTOAaron Isotton2003-08-12
1. IntroductionA question which frequently arises among Unix C++ programmers is how to load C++ functions and classes dynamically using the dlopen API. In fact, that is not always simple and needs some explanation. That's what this mini HOWTO does. An average understanding of the C and C++ programming language and of the dlopen API is necessary to understand this document. This HOWTO's master location is http://www.isotton.com/howtos/C++-dlopen-mini-HOWTO/. 1.1. Copyright and LicenseThis document, C++ dlopen mini HOWTO, is copyrighted (c) 2002 by Aaron Isotton. Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.1 or any later version published by the Free Software Foundation; with no Invariant Sections, with no Front-Cover Texts, and with no Back-Cover Texts. 1.2. DisclaimerNo liability for the contents of this document can be accepted. Use the concepts, examples and information at your own risk. There may be errors and inaccuracies, that could be damaging to your system. Proceed with caution, and although this is highly unlikely, the author(s) do not take any responsibility. All copyrights are held by their by their respective owners, unless specifically noted otherwise. Use of a term in this document should not be regarded as affecting the validity of any trademark or service mark. Naming of particular products or brands should not be seen as endorsements. 1.3. Credits / ContributorsIn this document, I have the pleasure of acknowledging (in alphabetic order):
1.4. FeedbackFeedback is most certainly welcome for this document. Send your additions, comments and criticisms to the following email address: <aaron@isotton.com>. 2. The ProblemAt some time you might have to load a library (and use its functions) at runtime; this happens most often when you are writing some kind of plug-in or module architecture for your program. In the C language, loading a library is very simple (calling dlopen, dlsym and dlclose is enough), with C++ this is a bit more complicated. The difficulties of loading a C++ library dynamically are partially due to name mangling, and partially due to the fact that the dlopen API was written with C in mind, thus not offering a suitable way to load classes. Before explaining how to load libraries in C++, let's better analyze the problem by looking at name mangling in more detail. I recommend you read the explanation of name mangling, even if you're not interested in it because it will help you understanding why problems occur and how to solve them. 2.1. Name ManglingIn every C++ program (or library, or object file), all non-static functions are represented in the binary file as symbols. These symbols are special text strings that uniquely identify a function in the program, library, or object file. In C, the symbol name is the same as the function name: the symbol of strcpy will be strcpy, and so on. This is possible because in C no two non-static functions can have the same name. Because C++ allows overloading (different functions with the same name but different arguments) and has many features C does not — like classes, member functions, exception specifications — it is not possible to simply use the function name as the symbol name. To solve that, C++ uses so-called name mangling, which transforms the function name and all the necessary information (like the number and size of the arguments) into some weird-looking string which only the compiler knows about. The mangled name of foo might look like foo@4%6^, for example. Or it might not even contain the word "foo". One of the problems with name mangling is that the C++ standard (currently [ISO14882]) does not define how names have to be mangled; thus every compiler mangles names in its own way. Some compilers even change their name mangling algorithm between different versions (notably g++ 2.x and 3.x). Even if you worked out how your particular compiler mangles names (and would thus be able to load functions via dlsym), this would most probably work with your compiler only, and might already be broken with the next version. 3. The Solution3.1. extern "C"C++ has a special keyword to declare a function with C bindings: extern "C". A function declared as extern "C" uses the function name as symbol name, just as a C function. For that reason, only non-member functions can be declared as extern "C", and they cannot be overloaded. Although there are severe limitations, extern "C" functions are very useful because they can be dynamically loaded using dlopen just like a C function. This does not mean that functions qualified as extern "C" cannot contain C++ code. Such a function is a full-featured C++ function which can use C++ features and take any type of argument. 3.2. Loading FunctionsIn C++ functions are loaded just like in C, with dlsym. The functions you want to load must be qualified as extern "C" to avoid the symbol name being mangled. Example 1. Loading a Function main.cpp:
hello.cpp:
The function hello is defined in hello.cppas extern "C"; it is loaded in main.cpp with the dlsym call. The function must be qualified as extern "C" because otherwise we wouldn't know its symbol name.
3.3. Loading ClassesLoading classes is a bit more difficult because we need an instance of a class, not just a pointer to a function. We cannot create the instance of the class using new because the class is not defined in the executable, and because (under some circumstances) we don't even know its name. The solution is achieved through polymorphism. We define a base, interface class with virtual members in the executable, and a derived, implementation class in the module. Generally the interface class is abstract (a class is abstract if it has pure virtual functions). As dynamic loading of classes is generally used for plug-ins — which must expose a clearly defined interface — we would have had to define an interface and derived implementation classes anyway. Next, while still in the module, we define two additional helper functions, known as class factory functions. One of these functions creates an instance of the class and returns a pointer to it. The other function takes a pointer to a class created by the factory and destroys it. These two functions are qualified as extern "C". To use the class from the module, load the two factory functions using dlsym just as we loaded the the hello function; then, we can create and destroy as many instances as we wish. Example 2. Loading a Class Here we use a generic polygon class as interface and the derived class triangle as implementation. main.cpp:
polygon.hpp:
triangle.cpp:
There are a few things to note when loading classes:
4. Frequently Asked Questions
The problem is, as usual, Windows. There is no dlfcn.h header on Windows,and there is no dlopen API. There is a similar API around the LoadLibrary function, and most of what is written here applies to it, too. Alternatively, you can use libltdl (included in libtool) to "emulate" dlopen on a variety of platforms. You should also read section 4, "Dynamically Loaded (DL) Libraries", of the Program Library HOWTO for more techniques to load libraries and create classes independently of your platform. I don't know of any, and I don't think there'll ever be one supporting all of dlopen's options. There are alternatives though: libtltdl (a part of libtool), which wraps a variety of different dynamic loading APIs, among others dlopen and LoadLibrary. Another one is the Dynamic Module Loading functionality of GLib. You can use one of these to ensure better possible cross-platform compatibility. I've never used any of them, so I can't tell you how stable they are and whether they really work. 5. See Also
BibliographyISO14482 ISO/IEC 14482-1998 — The C++ Programming Language. Available as PDF and as printed book from http://webstore.ansi.org/. |