> Apart from that, this case doesn't really fit into any other case that
> handles automatically created buffers.  Although -
> `ediff-dispose-of-variant-according-to-user' is not only about indirect
> buffers, right?  We could marry this case with this existing mechanism.
> Would that make some sense?

Currently, ediff-dispose-of-variant-according-to-user

- kills any indirect buffers with ediff-temp-indirect-buffer non-nil
  (used by region/window comparison), and

- if ediff-keep-variants is nil, asks about killing any unmodified
  buffers.

It's not clear to me how best to fit modified non-file buffers (such as
the FILE=... buffer produced by ediff-current-file) into this picture.
And, as you point out, the user might wish to keep such buffers around.

> And apart from that, adding a STARTUP-HOOKS argument to
> `ediff-current-file' looks acceptable to me, although it would have no
> direct uses in Emacs currently.  I don't like it when users are forced
> to use really bad hacks to change the behavior of such cases.

That indeed seems like the simplest way to allow customization that
doesn't resort to hacks or simply overwriting the function, so please
see the attached patch.