This is incorrect. Qmail *is* being maintained, just not by DJB.
And DJB refuses to allow Qmail to be updated. Therefore I stick
by my original claim.
Making a call out to a sharred library is no different in principle
than making a call out to an object library.
I think you should look at the qmail source again. It makes a several
fork/exec's on each SMTP transaction. Despite the fork/execs, qmail
is still very fast because it doesn't do any expensive operations such
as your suggested dlopen.
Do the dlopen in the initialization loop and pass it to the child process
along with the socket.
If the O/S supports transfer of sockets then this is almost certain to work.
Calling a handle returned by a dynamic link library is not going to be any
slower than a normal handle call.
This can be made to work if anyone cares to want to make it work.