You would break a "book" into a number of HTML+ nodes (files)
with a "contents" node that points to the others and so on
hierarchically as needed. Entities *wouldn't* be pulled across
as you describe. Instead the user is free to follow embedded
links (as now). You can also define navigational links for
properties such as"prev/next chapter", "table of contents"
and "glossary". These links may be defined explicitly or
implicitly as described in the HTML+ spec. The advantage of
implicit links is that they avoid having to insert links
into all the constituent nodes and further allow given nodes
to occur in more than one "book".
> Also, the external files would not each be parsable SGML,
> right? They would be SGML fragments because their parent
> section containers would be in the master file. So they
> couldn't be accessed directly. I'm looking for random
> access into pieces of a large document.
Wrong! The HTML+ spec ensures that each node can conform to
the DTD regardless of its position in the wider hierarchy.
So random access to nodes is OK.
> I'm hoping the hierarchical relationships between sections
> can still be encoded in SGML, using a flatter style of
> DTD like HTML
Indeed they can.
Dave Raggett
(wishing he had made this section of the draft spec clearer!)