Working with vendor code in C can get very tricky, especially when you except breaking changes to occur. Especially when you have multiple binaries depending on that vendor code, updating at different times, necessitating different live versions. Let’s explore.


Assume you’re working with an external vendor, who is providing you with code for a wonderful function getFoo:

// foo.h version 1.2.3

int getFoo();

// foo.c version 1.2.3

int getFoo() {
	sleep(1000); // TODO improve performance
	return 42

You use this function in many of your products - for example, in your best-selling barApp application:

// barApp.c

#include <stdio.h>

int main() {
	printf("%d\n", getFoo());
	return 0;

So barApp, and other applications, would want to use a foo library. It makes sense to provide this function in a shared library ( However, this library will change in the future, in several ways:

  1. Binary-compatible changes
  • Performance improvements (sleep will be removed)
  • Additional functionality will become available (new functions)
  1. Binary-incompatibile changes - at the very least, recompilation will be necessary
  • For C, this is usually caused by changes to macros
  • For C++, a plethora of reasons: Virtual function reimplementation, function inlining, new private data members…
  1. Source-incompatible changes - these will require you to change your source code (in barApp):
  • Functions (which you use) being removed or renamed
  • Semantic changes - getFoo could return 43

This gets even more complicated due to the fact that barApp is an operational, mission-critical application for your organization. Developers may need to hotfix older versions of barApp, which use older versions of libfoo. The build servers and developer boxes will need to be able to have multiple versions of libfoo installed simultaneously.

Compiling, installing, and using a shared library properly

First, the upstream vendor should compile with an SONAME, like so:

gcc -shared -Wl,-soname, -o foo.c
objdump -x | grep SONAME
# SONAME     

The guarantee the upstream vendor should give is this: As long as SONAME doesn’t change, binary compatibility will be retained.

Now, you (or, preferably, your package manager) should install the package on your machine like so:

mkdir -p /usr/include/foo1
cp foo.h /usr/include/foo1
cp /usr/lib
ldconfig -v | grep libfoo
# ->

Now, traditionally another symlink -> would be created, so you could compile barApp with -lfoo. However, here’s an alternative:

gcc -I/usr/include/foo1 barApp.c -o barApp
ldd barApp
# =>  (0x00007fff8edfe000)
# => /usr/lib/ (0x00007fb367cce000)
# => /lib/x86_64-linux-gnu/ (0x00007fb367906000)
# /lib64/ (0x00007fb367ef2000)

Now barApp is compiled, and looks for - it will find it thanks to the symlink created by ldconfig, and use


Binary-compatible updates

Suppose a new, compatible, faster version of libfoo is released - say version \1.3.0, which has removed that pesky sleep. Well, just place it in /usr/lib and rerun ldconfig.

cp /usr/lib
ldconfig -v | grep libfoo
# -> ->

The symlink has been updated, and now all applications (barApp, for example) which were linked against will have improved performance.

Incompatible updates

Suppose a new, incompatible version 2.0.0 of libfoo is released, which would force the newer barApp2.0 to be recompiled against the new, different headers. No problem:

mkdir -p /usr/include/foo2
cp foo.h /usr/include/foo2
cp /usr/lib
ldconfig -v | grep libfoo
# -> ->
# -> ->
gcc -I/usr/include/foo2 barApp2.0.c -o barApp2.0

Both versions of libfoo are installed simultaneously, and do not conflict.

Final thoughts

The Debian policy guide states that -dev packages should include the symlink. However, this would cause a conflict between the -dev packages for two different generations of libfoo. I am curious as to how this problem is solved “in the wild”, as I’m sure Debian have good reasons for suggesting this.