On Fri, 24 Aug 2012, Philipp Schmidt wrote:
Hi, I have multiple classes that I need to hide some functionality. Said functionality is in part provided by code that is being generated from files outside of my control. Since they provide different services that all utilize a specified format many class and function names are the same and even have the same interface. Yes, the best solution would be to resolve that, but sadly that is not possible. So I somehow have to work around that. In order to seperate them I put them all in seperate shared libraries. So now I have some libraries, libA, libB etc. that have nonambiguous interfaces. In the implementation file (not the header) each library includes the according header files from the generated code and operates on data calling the generated functions. So nothing is visible to the outside and they all compile fine. In the main program (successfully linked against all libraries) I now generate classes from them. Now the call to a function from the libA-class that uses the foo()-function from the generated code for libA works as expected, but when getting to the function in libB that calls the foo() function from the generated code for libB it segfaults. The backtrace revealed that the class from libB actually tries to call the foo()-function from libA, where of course some symbols are different. So far so logical. But how can I get my code to work so that the foo() function used in libB is actually the one that I included and compiled into libB? Can I somehow tell the gcc linker to first look in the library code where the function is being executed?
Depending on what exactly you can change, you may be interested in -fvisibility (or #pragma gcc visibility), or linker maps/scripts.
-- Marc Glisse
