Do we really need such generic stuff? ... IMO explicit is better than
implicit. Why not getting an error when a function, which is referred
from a reST-document disappears in the source? Those errors help
to maintain the consistency of documentation with source-code.

That's a totally different problem.

I know, there are also use-cases where generic is very helpful (e.g.
create a complete API description from the header file, with just
one line in reST). And I know, that we have already generic e.g. the
"export" option of the kernel-doc directive.

Exactly. But now you can either

 * use "export" or "internal" to get *everything*
 * list every single function, and get no warning when there's a
   function you didn't list

This serves to help get a mixture of the two, to be able to group
things but also document everything that got missed as a fall-back.

johannes