Is lib{library name}.a / .so a naming convention for static libraries in Linux?

Question

I've had to do some minor programming on a Ubuntu system recently (at which I am an extremely low-level beginner) and I'm really just getting familiar with makefiles.

I noticed that the arguments to tell the linker which libraries to include were always -l{library name} where the corresponding library would be something called "lib{library name}.a" in the /usr/lib folder.

I am wondering: is that a convention? I would have thought I would need to type -llibNAME to find a library called libNAME.a, but it seems to assume a lib prefix.

Is this always the case? Can I name a library without using a lib prefix?

Answer 1

name.a is a static library (a because it's an archive of objects).

name.so is a dynamic library (so because it's a shared object, also sometimes known as a DSO, for Dynamic Shared Object).

The -lfoo linker switch traditionally assumes a name of the form libfoo.{so,a}, and searches for it on the library path. You can also pass a library name to the linker directly (without using the -l switch), but in that case you have to pass the path to the library explicitly.

As @geekosaur noted, if you open a shared object at runtime, dlopen() takes the full filename.

Answer 2

You can name one any way you want, but ld's -l assuming a lib prefix applies to both static and shared libraries and goes back a long way; you'd need to name it explicitly to use one without the lib prefix.

This is actually useful even on modern systems: a name libfoo.so can be identified as a link-time library, while foo.so indicates a shared object implementing a runtime plugin. Or subsystem-specific prefixes in place of lib to identify plugins for particular subsystems; see for example pam_*.so and nss_*.so.