🚀 go-pugleaf

Here's a straw-man proposal for an upgrade to SGML for handling overlapping
markup streams.  I won't claim it's very pretty, but it has advantages:

   * It requires *no* changes to the syntax of SGML documents, but
     only an additional keyword permitted in ATTLIST declarations in DTDs.

   * A parser that doesn't know about it could ignore the keyword and
     produce acceptable output (such as the identical ESIS).

   * Unbounded numbers and types of overlaps can be encoded inline.

   * The element relationships are pretty human-readable and findable.

   * A parser that does know about the construct can at least validate
     that all tags for overlapping elements balance.

The mechanism is somewhat similar to Eliot's recent proposal; it is
also among several approaches discussed over the last
four years by the relevant Text Encoding Initiative committees, and for
slightly longer by the CHUG Hypermedia Working Group.

Here it is, building on my previous example:

   <!ELEMENT speech.start - - EMPTY>
   <!ATTLIST speech.start connect  STARTID #REQUIRED>
   <!ELEMENT speech.end   - - EMPTY>
   <!ATTLIST speech.end   connect  ENDID   #REQUIRED>

   <speechstart speaker='Sir John' connect=chunk5>
   <poetry-line>Greetings, O brave defender of our right,</poetry-line>
   <poetry-line>'Till now, I thought you knew not how to Write</poetry-line>
   <poetry-line>but heavy Morals.
   <speechend connect=chunk5>

   <speechstart speaker='Melissa' connect=chunk7>
   These my Pens imploy</poetry-line>
   <poetry-line>for all my business was to pall Your Joy:</poetry-line>
   ...
   <speechend connect=chunk7>

Almost exactly as with ID and IDREF attributes, there is a validity constraint:

   * The value of a STARTID attribute must be unique among all STARTID
     attribute values in a document instance.
     
   * The value of a ENDID attribute must be unique among all ENDID
     attribute values in a document instance.

   * Every STARTID attribute value in a document instance must also occur
     as exactly one ENDID attribute value *later* in the same document.
   
This is slightly easier to check than ID/IDREF matching, and guarantees a
certain level of integrity.  The semantics of the overlap are in the
application's domain (say, whether it should be highlighted, editable,
whether there are cross-constraints between overlaps, etc).

A slight improvement to this construct would have you declare both attributes
for the *same* GI, and have the constraint that the elements having
matching STARTID/ENDID must be of the same GI.  I actually like this larger
extension much better, because it strengthens validation and allows using
the GI to actually reflects the types of overlapping elements (rather than
permitting only a naming convention such as "speech.start" and "speech.end").

I also see it as a plus that current parsers can handle documents using
either construct without even being upgraded, simply by letting the user
change the DTD to read "ID" for "STARTID" and "IDREF" for "ENDID" (although
this weakens validation and expressiveness, documents can be created and
meaningfully parsed right now).

Anybody see problems with this idea?  It does not attempt to address
Darrell's case of truly discontiguous elements, only the case of contiguous
but overlapping elements.  With a little tuning would this make sense for
some of the applications we've been discussing?


Steve DeRose