by Angel Leon. March 17, 2015;
Last update on December 14, 2023
Updated on February 27, 2023
Updated August 29, 2019.
On the compilation phase, you will usually need to specify the different include paths so that the interfaces (.h, .hpp) which define structs, classes, constants, and functions can be found.
With gcc and llvm include paths are passed with -I/path/to/includes, you can pass as many -I as you need.
In Windows, cl.exe takes include paths with the following syntax:
/I"c:\path\to\includes\ you can also pass as many as you need.
Some software uses macro definition variables that should be passed during compile time to decide what code to include.
These compilation-time variables are passed using -D,
e.g. -DMYSOFTWARE_COMPILATION_VARIABLE -DDO_SOMETHING=1 -DDISABLE_DEPRECATED_FUNCTIONS=0
These compilation time flags are by convention usually put into a single variable named CXXFLAGS, which is then passed to the compiler as a parameter for convenience when you're building your compilation/make script.
When you compile your .c, or .cpp files, you will end up with object files.
These files usually have .o extensions on Linux, on Windows they might be under .obj extensions.
You can create an .o file for a single or for many source files.
When you have several .o files, you can put them together as a library, a static library. In Linux/Mac these static libraries are simply archive files, or .a files. In windows, static library files exist under the .lib extension.
They are created like this in Linux/Mac:
ar -cvq libctest.a ctest1.o ctest2.o ctest3.o
libctest.a will contain ctest1.o,ctest2.o and ctest2.o
They are created like this on Windows:
LIB.EXE /OUT:MYLIB.LIB FILE1.OBJ FILE2.OBJ FILE3.OBJ
Shared or dynamic libraries, such as .so in Linux, .dylib in Mac, and .dll in Windows, are critical for reducing executable sizes and memory consumption, but they differ in their implementation and behavior.
In Linux, .so files are created like this:
gcc -Wall -fPIC -c *.c
gcc -shared -Wl,-soname,libctest.so.1 -o libctest.so.1.0 *.o
-Wallenables all warnings.-cmeans compile only, don't run the linker.-fPICmeans "Position Independent Code", a requirement for shared libraries in Linux.-sharedmakes the object file created shareable by different executables.-Wlpasses a comma separated list of arguments to the linker.-sonamemeans "shared object name" to use.-o <my.so>means output, in this case the output shared library
In Mac, .dylib files are created similarly:
clang -dynamiclib -o libtest.dylib file1.o file2.o -L/some/library/path -lname_of_library_without_lib_prefix
DLLs in Windows function differently from shared objects in Unix/Linux systems. They tie both exports and imports to specific DLL names, creating a more rigid binding. This difference is essential for understanding how symbols and libraries are managed in different environments.
In Windows, .dll files are created like this:
LINK.EXE /DLL /OUT:MYLIB.DLL FILE1.OBJ FILE2.OBJ FILE3OBJ
When linking with MSVC, unlike Unix/Linux linkers where you point directly to the .so, you link against an import library (.lib) or in rare cases the .exp. These import libraries contain import thunks and a symbol index, facilitating the linking process. The .lib format is used by both MSVC and GCC on Linux, but with different contents (COFF files in MSVC, ELF .o files in GCC).
When linking your software you may be faced with a situation on which you want to link against several standard shared libraries.
If all the libraries you need exist in a single folder, you can set the LD_LIBRARY_PATH to that folder. By common standard all shared libraries are prefixed with the word lib. If a library exists in LD_LIBRARY_PATH and you want to link against it, you don't need to pass the entire path to the library, you simply pass -lname and you will link your executable to the symbols of libname.so which should be somewhere inside LD_LIBRARY_PATH.
Tip: You should probably stay away from altering your LD_LIBRARY_PATH, if you do, make sure you keep its original value, and when you're done restore it, as you might screw the build processes of other software in the system which might depend on what's on the LD_LIBRARY_PATH.
If you have some other libbar.so library on another folder outside LD_LIBRARY_PATH you can explictly pass the full path to that library /path/to/that/other/library/libbar.so, or you can specify the folder that contains it -L/path/to/that/other/library and then the short hand form -lbar. This latter option makes more sense if the second folder contains several other libraries.
Sometimes you may be dealing with issues like undefined symbol errors, and you may want to inspect what symbols (functions) are defined in your library.
On Mac there's otool, on Linux/Mac there's nm, on Windows there's depends.exe (a GUI tool that can be used to see both dependencies and the symbol's tables. Taking a look at the "Entry Point" column will help you understand clearly the difference between symbols linking to a shared library vs symbols linking statically to the same library)
See shared library dependencies on Mac with otool
otool -L libjlibtorrent.dylib
libjlibtorrent.dylib:
libjlibtorrent.dylib (compatibility version 0.0.0, current version 0.0.0)
/usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 120.0.0)
/usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1213.0.0)
See shared symbols with nm (Linux/Mac)
With nm, you can see the symbol's name list.
Familiarize yourself with the meaning of the symbol types:
T(text section symbol)U(undefined - useful for thoseundefined symbolerror),I(indirect symbol).
If the symbol is local (non-external) the symbol type is presented in lowercase letters, for example a lowercase u represents an undefined reference to a private external in another module in the same library.
nm's documentation says that if you're working on Mac and you see that the symbol is preceeded by + or - it means it's an ObjectiveC method, if you're familiar with ObjectiveC you will know that + is for class methods and - is for instance methods, but in practice it seems to be a bit more explicit and you will often see objc or OBJC prefixed to those methods.
nm is best used along with grep ;)
Find all Undefined symbols
nm -u libMacOSXUtilsLeopard.jnilib
_CFRelease
_LSSharedFileListCopySnapshot
_LSSharedFileListCreate
_LSSharedFileListInsertItemURL
_LSSharedFileListItemRemove
_LSSharedFileListItemResolve
_NSFullUserName
_OBJC_CLASS_$_NSArray
_OBJC_CLASS_$_NSAutoreleasePool
_OBJC_CLASS_$_NSDictionary
_OBJC_CLASS_$_NSMutableArray
_OBJC_CLASS_$_NSMutableDictionary
_OBJC_CLASS_$_NSString
_OBJC_CLASS_$_NSURL
__Block_copy
__NSConcreteGlobalBlock
__dyld_register_func_for_add_image
__objc_empty_cache
__objc_empty_vtable
_calloc
_class_addMethod
_class_getInstanceMethod
_class_getInstanceSize
_class_getInstanceVariable
_class_getIvarLayout
Linking is simply "linking" a bunch of .o files to make an executable.
Each one of these .o's may be compiled on their own out of their .cpp files, but when one references symbols that are supposed to exist in other .o's and they're not to be found then you get linking errors.
Perhaps through forward declarations you managed your compilation phase to pass, but then you get a bunch of symbol not found errors. Make sure to read them slowly, see where these symbols are being referenced, you will see that these issues occur due to namespace visibility in most cases.
Perhaps you copied the signature of a method that exists in a private space elsewhere into some other namespace where your code wasn't compiling, all you did was make it compilable, but the actual symbol might not be visible outside the scope where it's truly defined and implemented.
Function symbols can be private if they're declared inside anonymous namespaces, or if they're declared as static functions.
An example:
Undefined symbols for architecture x86_64:
"FlushStateToDisk(CValidationState&, FlushStateMode)", referenced from:
Network::TxMessage::handle(CNode*, CDataStream&, long long, std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> >&, bool, bool) in libbitcoin_server.a(libbitcoin_server_a-TxMessage.o)
Here, when I read the code of Network::TxMessage::handle(...) there was a call to FlushStateToDisk, which was declared in main.h, and coded in main.cpp. My TxMessage.cpp did include main.h, the compilation was fine, I had a TxMessage.o file and a main.o, but the linker was complaining.
The issue was that FlushStateToDisk was declared as a static, therefore only visible inside main.o, once I removed the static from the declaration and implementation the error went away and my executable was linked. Similar things happen when functions are declared in anonymous spaces in other files, even if you forward declare them on your local .h
In other cases your code compiles and you get this error linking because your library can't be added using -lfoo, and adding its containing folder to -L doesn't cut it, in this case you just add the full path to the library in your compilation command: gcc /path/to/the/missing/library.o ... my_source.cpp -o my_executable
DO NOT EXPORT CFLAGS, CPPFLAGS and the like on your .bash_profile/.bashrc, it can lead to unintended building consequences in many projects. I've wasted so many hours due to this mistake.
Nice article, but how can I find in which library of my Linux, my undefined function exists?
I might use ceil and not know which library I must link with - llibraryname