Suppose that I don't have room in /usr for /usr/local but I do have some room in /export. So I make a directory called /export/usr.local and it has a subdirectory called bin. And I do a "ln -s /export/usr.local /usr/local". Now I do a "cd /usr/local" then I do a "cd bin". At this point my view is that I am in /usr/local/bin and a shell with a built-in pwd command may return that path. But /usr/bin/pwd will return /export/usr.local/bin. With more symbolic links in a path, there can be more aliases like this. The value returned by /usr/bin/pwd is the physical path. With good permissions on each directory leading to the current directory, /usr/bin/pwd can return the physical path. A built-in pwd in the same shell that navigated to the current directory may be able to return the particular logical path used to arrive at the current directory. Finding all logical paths to a particular directory would be rather daunting.

Writing a program like /usr/bin/pwd is not very easy in unix. You can stat the . directory to get the inode of the current directory. Then you can open .. and stat each file in the parent directory until you find the same inode. You walk up the chain one .. at a time repeating this process until . and .. are the same inode which means you have reached /. (Actually it is harder than that because you might traverse a mount point.) This is why putting an NFS mounted filesystem in / is very unwise... every /usr/bin/pwd must contact the NFS server to perform the stat.