Problem from SGML 93 by David Sklar.

Consider a DTD that allows FIGURE elements to be placed anywhere in
the body of the document.  Some figure elements have caption
attributes, but not all of them do:

   <!ELEMENT ROOT   (FRONTMATTER,BODY,ENDMATTER)>
   <!ELEMENT FIGURE - O (HOTSPOT*)>
   <!ATTLIST FIGURE
             caption  CDATA  #IMPLIED
             filename CDATA  #REQUIRED>
   
Create a transformation that appends an element called FIGREVIEW to
the end of the ENDMATTER's children list:

   <!ELEMENT FIGREVIEW - - (FIGURE+)>

FIGREVIEW simply contains copies of all the FIGURE elements found in
the body of the document, sorted lexicographically based on the
figure's caption.  (Assume all text is ISO 8859-1.)  Uncaptioned
figures should not appear at all in the FIGREVIEW section.

Note that each FIGURE element is actually the root of a subtree of
arbitrary size (I intentionally don't show the content model for
HOTSPOT); make sure the entire subtree of each captioned FIGURE is
duplicated in the FIGREVIEW section.  (I am interested in knowing if
that is possible, in your technology, without full knowledge of the
content model of HOTSPOT.)