How does the C++ linker know which .lib contains which functions?

Question

For example in Boost. I set an include directory in MSVC++2010 to the Boost root directory and have an #include <boost/regex.hpp> in my source code. I set a library directory to boost\stage\lib but there are hundreds of files in there – several for each Boost library and these for boost::regex:

libboost_regex-vc100-s-1_46.lib
libboost_regex-vc100-mt-gd-1_46.lib
libboost_regex-vc100-mt-1_46.lib
libboost_regex-vc100-mt-s-1_46.lib
libboost_regex-vc100-mt-s.lib
libboost_regex-vc100-s.lib
libboost_regex-vc100-mt.lib
libboost_regex-vc100-mt-gd.lib

How does MSVC know which of all lib files is the right one? If it scans all of them for the right function signatures, does that mean that 2 different lib's compiled from two different sources (not linked to each other) which happen to define functions with identical names and parameters cannot be in one lib folder?

And how does it know which is right among all those different regex .lib's? And then, each file with 1_46 in its filename seems to be identical to the respective file without, can I safely delete one of the two?

Answer 1

The boost libraries use some dark magic to select the libraries to link from the headers and compiler options. I don't really know all the gory details, but you can take a look at the boost/config/auto_link.hpp header for extra information.

In particular, this seems to be an important piece of the puzzle:

#  pragma comment(lib, BOOST_LIB_PREFIX BOOST_STRINGIZE(BOOST_LIB_NAME) "-" BOOST_LIB_TOOLSET BOOST_LIB_THREAD_OPT BOOST_LIB_RT_OPT "-" BOOST_LIB_VERSION ".lib")

Answer 2

Most lib files have a Table Of Contents. The linker searches this table when it is looking for a symbol. If the symbol is not found it moves on to the next library, and so on, until all libraries have been searched.

Some linkers may decide to build a table of contents from all the libraries. This table would contain the symbol name and the library it is associated with. This speeds up the searches for symbols.

The search order depends on the manufacturer of the linker. There is no standard nor requirements for this. A linker may search by first come, first served as specified on the command line; last library specified or some other method. Check the documentation for the criteria.

Also search the web for name mangling. This is a technique that compilers use to resolve symbol naming conflicts.

Lastly, linkers may include all the functions in a library even if only one is used. Some linkers only include the code for the function. Depends on the manufacturer of the Linker. For example, does the linker include the entire I/O library when resolving puts or does it just include the necessary functions? Including the whole library speeds up build time but makes the executable huge. Including only necessary code slows the build process but shrinks the executable size.

In general, the Linking Phase is one of the faster parts of the translation process. If you are worried about build times, start the build at the end of the day or go for a walk after the build is started. ;-)