bug#20845: 25.0.50; bookmark.el, handling of fictitious `buffer' property

[Drew Adams]

> > It should be sufficient to communicate the buffer
> > _name_ as the property value.  And that's something
> > savable and retrievable - it's useful generally.
> >
> > IOW, it's OK to have `buffer' bookmark property,
> > but I think its value should be a buffer name.
> 
> Would that potentially be a problem if the buffer is renamed?

Yes. But it's not much different from the problem
that arises if a file name is changed.

And no different from the case where a buffer
(not a buffer name) no longer exists (which can
happen even if a buffer object is used in a live,
not persisted, `bookmark-alist').

We can try to jump through hoops for this.  It's
a choice how many hoops we want to jump through.

We could `condition-case'-handle the case where
`bookmark-default-handler' is used, and try to
deal with errors such as those.  We already do
that for condition `bookmark-error-no-filename'.