Comparison of dynamic linking approaches? The Unix/ELF and Windows/PE dynamic linking differ in several interesting ways.
The ELF scheme uses a single name space per program, while the PE scheme uses a name space per library. An ELF executable lists the symbols it needs and the libraries it needs, but it doesn’t record which symbol is in which library. A PE file, on the other hand, lists the symbols to import from each library. The PE scheme is less flexible but also more resistant to inadvertent spoofing. Imagine that an executable calls routine AFUNC which is found in library A and BFUNC which is found in library B. If a new version of library A happens to define its own BFUNC, an ELF program could use the new BFUNC in preference to the old one, while a PE program wouldn’t. This is a problem with some large libraries; one partial solution is to use the poorly documented DT_FILTER and DT_AUXILIARY fields to tell the dynamic linker what libraries this one imports symbols from, so the linker will search those libraries for imported symbols before searching the executable and the rest of the libraries. The DT_SYMBOLIC field tells the dynamic linker to search the library’s own symbol table first, so that other libraries cannot shadow intra-library references