Unix domain sockets: How to name them? And where to put them?

Question

I am using unix sockets for inter-process communication in a C++/gtkmm project.
Are there naming conventions and conventions about where to place sockets in the filesystem? In all example code I found, dummies like /home/user/mysocket were being used.
If there are not any conventions: Would it be sensible to put the sockets in /tmp/ and compose the names of the name of the program which uses the concerning socket, and its PID? (The latter also depends on my use case, of course; from that point of view, it seems sensible.)

Answer 1

@Franek:  KDE applications place their sockets (through KDE API, of course) in "/tmp/ksocket-foo/bar__0", where "foo" is the user name, "bar" the application name and "0" an increasing counter (in case an application creates more than one socket).  Emacs places its server socket in a similar manner "/tmp/emacs1000/server", where "1000" is the UID.
I'd use emacs' name scheme, e.g. "/tmp/bar1000/what_the_socket_is_for", where "bar" is the application name again. 
However, do not place sockets directly in "/tmp/". For one thing, it just clutters this global directory, but more importantly it's unsafe, as every other user can inspect these sockets.  Create a directory within "/tmp/" and restrict its access rights to the owner (mode 700).  Thus other users can't inspect the sockets within that directory.  Of course, the access rights of the sockets themselves must also be limited.
Last edited by lunar (2011-10-12 18:59:32)