Towards XML Transclusions
Angelo Di Iorio, Silvio Peroni, Fabio Vitali John Lumley, Tony Wiley
Dept. of Computer Science HPLabs Bristol
University of Bologna Filton Road, Stoke Gifford
Mura Anteo Zamboni, 7 Bristol BS34 8QZ
40127, Bologna, Italy United Kingdom
{diiorio | speroni | fabio}@cs.unibo.it {john.lumley | anthony.wiley}@hp.com
ABSTRACT Transclusions were the core idea of Xanadu. Xanadu doc-
The idea of transclusion has been at the same time the uments were built on-the-fly from fine-grained references so
strength and weakness of Xanadu: some people considered that users could access, modify and reuse any fragment from
it as an extremely powerful mechanism to get any version any document in a safe and controlled way. The implemen-
of any fragment of any document in a global shared docu- tation of transclusions relies on external referencing to lo-
ment space, others as a very complex solution too difficult to cations in a text data stream, through complex addressing
be actually implemented and delivered. We believe transclu- mechanisms. Mark-up information, links and metadata are
sions are still worth implementing and would allow designers expressly distinct from the flow of text in order to guarantee
to build very sophisticated hypermedia applications. On the flexibility and expressiveness.
other hand, we are aware that the original design of Xanadu
cannot be implemented without uprooting current systems, The world of XML (and SGML) relies on a completely differ-
protocols and technologies - in primis the World Wide Web ent strategy: mark-up is embedded, documents are strictly
and XML. In fact, there is a great distance between the hierarchical and assertions about text fragments are made
original data model of transclusions - strongly based on ex- by wrapping them with elements, enriched by attributes.
ternal referencing mechanisms - and the XML data model Ted Nelson himself pointed out disadvantages of such an
- strongly based on hierarchical structures and embedded approach in relation with transclusions[12]. The three objec-
markup. tions he raised can be summarized as: (1) SGML approach
This paper investigates to what extent the concept of tran- interposes a ‘forced’ structure between users and actual con-
sclusion can be shaped for the world of XML, and stud- tent while editing, (2) SGML approach only supports well-
ies simplified models for building functionalities inspired by formed inclusions and does not allow users to change in-
Xanadu. Particular attention is given to the support for cluded content, (3) SGML approach does not support over-
tracing fragments provenance in multi-source documents and lap and non-hierarchical relationships.
for synchronizing distributed content through transclusions.
The paper also traces a roadmap to actually implement tran- Thus, these two positions seem to be irreconcilable. The
sclusions for XML - identifying three incremental steps - and goal of this paper is to investigate whether and how tran-
briefly describes some experimental prototypes. sclusions can be implemented for XML documents and trace
a possible course towards that goal.
Categories and Subject Descriptors
H.5.4 [ INFORMATION INTERFACES AND PRE- The preliminary step is to understand what we mean by
SENTATION (I.7)]: Hypertext/Hypermedia ‘XML transclusions’. Our goal is not to re-implement a re-
vised version of Xanadu based on XML technologies, rather
to support users in creating and editing composite XML
documents that make some Xanalogical functionalities pos-
1. INTRODUCTION sible. In particular, we are interested in: tracing fragment
The concept of transclusion is rooted in the early days of
provenance and remote manipulation. First of all, compos-
hypertext[13]. A transclusion is a very advanced inclusion,
ite XML documents would benefit from rich information
whose content is not actually copied but stored as a virtual
about the origin of each fragment. That makes it possible to
reference to the original source. There is only one copy of
identify single contributions in collaboratively edited docu-
each fragment in the whole document space and transcluded
ments, to display multiple changes in a single document, to
data is permanently connected to the original.
go back to original resources and navigate documents in a
free and powerful way. The permanent connection between
transcluded content and original sources would also make
possible sophisticated forms of editing. Changes to remote
documents could be propagated through transclusions or -
the other way round - local modifications could actually up-
date remote resources. Yet, such scenarios also require other
tricky issues to be addressed such as content merging, syn-
chronization, access permissions, reliability and so on.
23
The core of this paper discusses three approaches for imple- The design and implementation of CB-OHS follows the ‘phi-
menting XML transclusions, in section 4. These solutions losophy of the primacy of structure over data’[16]. This is
use different syntaxes and are progressively expressive and a key feature of the Open Hypermedia approach, strongly
powerful. Before that, we briefly review the recent litera- related with the idea of transclusions: the separation be-
ture about transclusions and XML. The paper also mentions tween the relationships and the information they relate to.
some prototypes we are developing. It makes possible to handle data separately, to create links
pointing to read-only reasorces, to create multiple and over-
lapped links on the same content and so on.
2. TRANSCLUSIONS AND XML
The two most recent research efforts for implementing tran-
sclusions are outside the XML universe. Nelson and his 3. A CASE-STUDY: COLLABORATIVE RE-
team proposed Transliterature[11] a revision of the original VIEWS
Xanadu project built on newer technologies. Transliteral The idea of transcluding fragments from and to XML doc-
documents are dynamically built on top of transcluded frag- uments is still under-developed. Hereinafter we discuss a
ments, so that rich and fine-grained connections between road-map to make that development possible, by using a
documents are permanently available. Two prototypes are case study throughout the paper.
worth mentioning: Transquoter, that allows users to hide,
highlight and surf multi-source quotations and XanaduS- Let us suppose we are building a system for supporting mul-
pace, that provides users a 3D view of the overall document tiple users to write collaborative reviews about movies. It
space and shows some of the advanced surfing and display- would be useful to let reviewers quote fragments from other
ing functionalities envisioned for such inter-connected doc- reviews, keep trace of their source and maintain a ‘live’ chan-
uments. nel between connected fragments. In fact, a review process
can be improved by quoting opposite criticism, by letting
Kolbitsch at al.[8] investigated transclusions for HTML, pre- reviewers access related reviews, by automatically updating
senting a prototype that allows users to select content from distributed reviews with new material and by fostering dis-
web pages and transclude them into new documents. Tran- cussion among reviewers. It is worth noting that such an
sclusions directives are stored as in-line elements and very idea is both rooted in the early days of hypermedia (the
complex URLs. Interesting issues are still open about con- essence itself of hypermedia is the inter-connection between
tent addressing and merging, especially considering that web documents) and central in the recent trend of the World
pages may change often and may have complex (and badly- Wide Web (one of the milestones of the so-called Web 2.0 is
formed) internal structures. just collaboration).
The closest solution to transclusions in the XML universe Consider now that reviewers are writing their comments
is XInclude[10]. XInclude is a W3C standard for merging about ‘Australia’, starring Nicole Kidman and Hugh Jack-
XML documents, by writing inclusion directives and retriev- man1 . The film is one of the most controversial of the early
ing other (parts of) documents. The focus of XInclude is 2009. David wrote a quite positive review: “Baz Luhrmann’s
on well-formed XML fragments. Although even text frag- Australia is good, but not a masterpiece”.2 Brad has an op-
ments can be included (parse="text"), XInclude does not posite opinion: “A major miscalculation if there ever was
allow users to include bad-formed fragments or ranges. That one”3 . Assume that Brad replied to David quoting his orig-
makes it impossible to implement fully-fledged transclusions. inal note. In fact, a third reviewer - say Mike - might be
Simplified forms of transclusions are possible through XIn- interested in quoting both these opinions, even by citing a
clude anyway. XIPr[17] is an implementation of XInclude different comment by a fourth reviewer (for instance, Re-
in XSLT 2.0, very efficient and simple to be integrated in becca saying “I actually regret having seen the film through
other XML applications. XSLT technologies could be also to the end.”4 ). Figure 1 shows a possible view of such a
used straightforwardly for simplified transclusions: a general composite review, highlighting multiple contributions5 .
approach is discussed in [5].
The multi-contribution view is only one of the applications
XLink[3] could also be cited as a way to transclude pieces of such an advanced quoting. Users might also be given the
of content. The @show and @actuate attributes allow users possibility to surf to the original review, in order to col-
to define hot-links, that are similar to transclusions apart lect more information about the movie. Moreover, perma-
from implementation details. On the other hand, such a nent connections between fragments could be enriched with
solution is only partial and XLink does not seem to succeed metadata (stating, for instance, that the remote review is
as expected. ‘positive’ or ’negative’), so that advanced search could be
performed over the network of documents. Note also that
XLink is a relative recent effort, rooted in a very long ex- 1
perience in the hypermedia research. Since the early 90s http://www.australiamovie.com/
2
researchers of the Open Hypermedia community have been http://www.theaustralian.news.com.au/story/0,25197,24670334-
designing systems that let users to add sophisticated links to 601,00.html
3
a wide range of documents, merging original content and ex- http://www.ropeofsilicon.com/article/movie-review-
australia-2008
ternal interventions. Chimera[1], Microcosm[6] and DLS[2] 4
http://movies.about.com/od/australia/fr/austral-
are only few representatives of these systems. More recently review.htm
Tzagarakis et al.[16] presented CB-OHS (Component-Based 5
The picture, shown in the following page, contains the final
Open Hypermedia Systems), an extensible infrastructure for rendering of the document rather than its internal (XML)
managing complex and distributed hypermedia elements. markup
24
clude inclusions. Figure 2 and 3 show simplified source codes
for Mike’s review (including Brad’s and Rebecca’s reviews)
and Brad’s one (including David’s).
Just read Brad’s review: "
Figure 2: XInclude instructions in mike.xml
Figure 1: A possible view of Mike’s review, high-
lighting multiple contributions
I have to fully disagree with David. How
can you say that ‘
Figure 3: XInclude instructions in brad.xml
Last, but not least, connections across fragments can be
used as ‘channels to make documents communicate’. Such
an approach can be considered as a simplified form of tran- An XInclude processor transforms mike.xml into a final doc-
sclusions: transclusion as a ‘live and bidirectional channel’ ument where all inclusions are resolved and content frag-
between pieces of content. The term ‘channel’ stresses the ments are merged together. The (partial) result we want
fact that such a mechanism would allow automatic updates to achieve enriches that document with detailed informa-
of content, in both directions: from a remote review to all re- tion about the origin of each fragment. It is crucial that all
views including that fragment and from a transcluded frag- inclusions traversed to reach the content are stored in the
ment to the original review it belongs to. Yet, the auto- document.
matic update of content opens tricky issues of synchroniza-
tion, conflict resolution, priority, content merging and so on. XInclude supports two types of inclusions: (i) text fragments
On the other hand, it opens fascinating perspectives for col- (parse="text" and xpointer addresses a sequence of char-
laborative documents accessing and editing. acters) or (ii) XML elements (parse="xml" and xpointer
addresses a well-formed XML fragment). The second case
The natural candidate to mark-up these composite docu- obviously requires that the source documents already con-
ments is XML. It is standard, universally supported and tain the XML elements to be included. In the example, all
very powerful. Surprisingly enough, none of the XML appli- quotations are encoded through the DocBook quote element
cations we are aware of provides users all these Xanalogical in brad.xml, rebecca.xml and david.xml.
functionalities together. The problem is paradoxically the
same hierarchical structure of XML, that makes impossible A first solution consists of introducing in the output new ele-
to address and overlap fine-grained transclusions[12]. The ments wrapping the included content. For instance, we could
question is now at a different level: to what extent XML use a new element decorated with metadata
transclusions can be implemented? Is it possible to create a about the inclusion. With that information, we can build
fully-fledged Xanalogical environment even for XML docu- applications to let users access metadata, surf to original
ments and use it for ‘collaborative reviews’ ?6 content, identify contributions, update content, etc.
There are two main drawbacks to such an approach. Firstly,
4. CHARTING A COURSE FOR XML TRAN- any tool processing the final document has to know about
SCLUSIONS the element xi:included. It is then impossible to directly
Intermediate results are also interesting, leading to a full im- reuse legacy tools such as DocBook converters or renderers.
plementation of XML transclusions. This section first dis- Tools that directly process children of unknown elements
cusses some partial objectives - that can be achieved today (basically ‘ignoring’ the presence of xi:included) could be
- and then focuses on long-term developments. used as well, but we cannot take such a ‘transparent’ be-
havior for granted. A second issue is related to the way
users define inclusions. Consider a fifth reviewer quoting
4.1 Step 1: Embedding simplified forms of XML fragments from (the expanded view of) mike.xml : how can
transclusions she/he set the xpointer attribute of inclusion instructions?
An extension of XInclude can support simplified forms of She/he has to also consider the xi:included elements and
XML transclusions. Let us encode the use-case reviews as write very complex expressions, unless XPointer addresses
DocBook documents, whose quotations are actually XIn- are filtered to ‘ignore’ those elements. In any case, a tan-
6 gled and error-prone management of internal addresses is
Note that many other scenarios would benefit such col- required. A detailed discussion about these issues can be
laborative reviews approach. Just think about the creation
of multi-source reports in a company that collect informa- found in[7], along with a possible solution based on Archi-
tion from existing resources, internal documents and exter- tectural Forms[15]. A running prototype is also presented
nal publications. in the same paper.
25
The basic idea of Architectural Forms is to extend the at- may happen for different reasons: errors in the document
tribute set for an element to express semantic information. address or in the internal location of a fragment, temporary
Since attributes (that belong to a different namespace) can or permanent disconnectivity, permission issues and so on.
be added to SGML and XML elements without impacting Solutions exploiting XML comments or decoration of pre-
the document’s integrity and without interfering with other ceding/following siblings might be investigated but seem to
applications, Architectural Forms can be exploited to mix be tortuous and awkward.
information in a transparent manner.
The second - and more important - issue is that such a model
The same approach can be applied to inclusions. The idea does not allow a lot of very common inclusions. Consider,
presented in [7] is to process all inclusions and ‘embed’ in- for instance, a reviewer interested in quoting only few words
formation in qualified attributes (belonging to a reserved of an existing quotation, or a sentence that partially spans
namespace), so that the main structure of the XML docu- over two paragraphs, or an interval that spans over a bad-
ment is not altered. The presence of these attributes does formed XML fragment. Such selections can be supported
not impact the basic processing, parsing and integrity of the only by exploiting external referencing mechanisms, that go
document. On the other hand, full expressive information beyond the scope of the strictly hierarchical organization of
about inclusions is available. Figure 4 shows such a solution XML.
applied to the use case.
Just read Brad’s review: "I have A further step to overcome the aforementioned limitations
to fully disagree with David. How can you say that consists of storing data about transcluded content in exter-
‘ Australia is good,but not information in the document itself. The prototype presented
a masterpiece
’? I would rather define it in [7] exports a module handling such externalization, al-
‘a major miscalculation if there ever was one’ though only some transclusions are actually supported. Fig-
". And Rebecca’s review is even worse: she ure 5 shows the example document mike.xml represented
actually "regret having seen the film through
to the end
".
about nested inclusions and spread that information all over
the XML tree. Consider, for instance, the quotation ‘Aus- Just read Brad’s review: "I have to fully
tralia is good, but not a masterpiece’: the document knows disagree with David. How can you say that
that it has been included from brad.xml but it was actually ‘Australia is good, but not a masterpiece’?I would
originated in david.xml. Thus, all copies of a given fragment rather define it ‘a major miscalculation if there
are very well connected in a complex network of inclusions, ever was one’". And Rebecca’s review is even worse:
available to both users and applications. she actually "regret having seen the film through
to the end".
In practice, xi:inclusion_history was never given great
importance and it was never really supported by the XIn- Figure 5: Resolving and externalizing inclusions in
clude implementations. In fact, the prototype presented in brad.xml
[7] uses a different syntax. Syntactic details are not relevant
here: what is really important is that record of inclusions are The content - even the transcluded one - is stored as a plain
hidden within documents and can be activated on-demand. stream of text while the ted:out-of-band data-structure
That information can be exploited to build the aforemen- contains pointers to text fragments and metadata about
tioned simplified form of XML transclusions, allowing users each fragment. Several advantages of such a representa-
to navigate transclusion metadata, to retrieve original con- tion can be outlined. First of all, it contains rich infor-
tent, to highlight multiple contributions and to automati- mation about the nesting of transclusions as well as rich
cally update content. metadata about original sources, without interfering with
users and applications that interact with the document. In
There many reasons why such a solution is only a partial fact, all transclusions data are stored externally and can be
step towards our goal. First of all, it only applies to the ignored by transclusion-unaware processors. The position
inclusion of well-formed XML, since qualified attributes - of the ted:out-of-band data-structure is also worth dis-
meant to be added to the included infoset - can be only cussing: it can be placed within the XML document itself
added on elements. One of the consequences is that no in- (for instance, as a first child of the root) or even in a com-
formation about ‘dangling’ inclusions is eventually stored in pletely external file. The first approach produces a single
the document. The term ‘dangling inclusion’ indicates an self-contained resource that can be moved from one system
inclusion that failed because content was not available. It to another without requiring any management of relative
26
links; on the other hand, it adds extra structures to the orig- us to exploit legacy tools of the Semantic Web community
inal XML tree. The second approach is completely trans- and to create sophisticated services of searching and rea-
parent and decoupled from the document tree-structure but soning over transcluded content. Nevertheless, the adoption
it requires applications to process links and set of resources. of RDF leaves still open all the very complex issues related
Both these approaches are valid, according to users needs to consistent update and manipulation of externalized tran-
and preferences. sclusions.
However, the main strength of such an externalized solution We finally propose a further generalization of RDF to in-
is its capability of defining transclusions that span over text tegrate in a single approach advantages of both embedded
fragments, ranges or even overlapped transcluded content. markup (a la XML, presented as a first step of our roadmap)
While Architectural Forms only work for transcluding well- and external annotations. A detailed discussion of that
formed XML fragments - as information is embedded into proposal - called EARMARK (Extreme Annotational RDF
attributes of the transcluded elements - external pointers Markup) - can be found in [14]. EARMARK derived from
may refer to any location in the document, through proper the analysis of some limitations of RDF. The main prob-
XPointer expressions. For instance, it is possible to iden- lem is that RDF assertions need URIs to address resources:
tify a transcluded fragment by simply indicating a pointer statements only apply to whole documents, fragments pro-
to the start- and end- offset of that fragment, and without vided with an identifier, or fragments addressable by some
altering the document itself. Note also that this approach al- URI schema. The XPointer standard (actually, the full
lows users to store information about dangling transclusions XPointer schema of XPointer [4]) makes it possible to re-
(whose content is not available because of network errors or fer to arbitrary pieces of text and XML documents. The
simply because of different regimes in accessing content) by adoption of that standard in conjunction with RDF would
simply adding a reference to the point where the transcluded solve all the URI-related issues. The problem is that the
content was supposed to be. There is no ‘intruder’ structure XPointer full schema has never been confirmed by the W3C
in the XML tree but very rich information is available about and its approval seems remote.
nested or empty transclusions.
The second interesting aspect of RDF is the fact that all
On the other hand, such an approach opens very tricky URIs are considered ‘opaque’ strings regardless of the type
issues of content merging and manipulation of references. of resource they refer to. It is very difficult to differenti-
The main problem is related to the ‘bad-formedness’ of in- ate an assertion about a text fragment from an assertion
cluded fragments. External references - and in particular about an element, to differentiate classes of assertions and
XPointer expressions - allow users to also include any piece relations among assertions. The solution we propose, EAR-
of an XML tree. Consider, for instance, a user selecting the MARK, defines an ontologically sound model that formal-
last sentence of a paragraph and the first one of the follow- izes the concepts of the XPointer schema, and a natural way
ing paragraph (a possible HTML source code: ‘concluding to express externally both RDF assertions and embedded
paragraphAnd starting a new one’). Several ques- markup constructs in the same framework. We also imple-
tions arise from that simple selection: which is the best way mented a first prototype translating documents from and to
to include that fragment into a new document? Is it a plain EARMARK format, RDF and embedded XML. Although
string (without the p elements), or does it generate two para- the implementation of EARMARK is at a very early stage,
graphs or is it a completely different structure? The adop- such a ‘semantic layer on top of transclusion’ is very flexible
tion of an externalized approach require us to address these and promising.
issues and to build a general and flexible document archi-
tecture for manipulating fine-grained fragments. The con-
sistence of external data structures is another very difficult
5. CONCLUSIONS
Transclusions have contributed to shape the most successful
issue to face. Consider as example the implementation of
hypermedia projects, although they were never fully imple-
an editor for such documents: it requires any edit operation
mented. After several years from their invention - dated
to be propagated to the external data structure in a consis-
back to the early 60s - they still keep the original attractive-
tent way. Even small changes require a complex network of
ness and potentialities. The challenge for the community is
statements and relations to be updated and manipulated.
now to adapt the original Xanalogical design to the current
technologies and systems, in particular to the World Wide
4.3 Step 3: Further generalization and exter- Web and XML.
nalization via RDF and EARMARK
The previous approach can be improved, although it is enough The contribution of this paper is manifold: (i) a preliminary
to provide users with a fully-fledged Xanalogical environ- discussion of simplified forms of transclusions, viable for the
ment. The idea is to exploit existing XML standards: in- world of XML, (ii) a roadmap to actually implement XML
stead of using ad-hoc data structures and operations, XML transclusions and (iii) a brief presentation of early proto-
transclusions could be described through RDF (and OWL) types we are developing in that direction.
statements. RDF is a language for representing informa-
tion about resources in the World Wide Web[9], that allows There are actually other crucial issues not discussed in this
users to make any assertion on any identifiable resource: work: content merging, synchronization and access rights.
statements on any fragment of text, multiple statements on They all require a much deeper analysis and experientation
the same resource and overlapped statements are all possi- we haven’t done yet. Our impression, however, is that ex-
ble in RDF. The additional benefit of RDF lies on its stan- ternalizing transclusions pave the way towards these goals.
dardization: expressing transclusions in RDF would allow In fact, a flexible and fine-grained model makes it possible
27
to integrate multi-source and multi-author fragments. An [16] M. Tzagarakis, D. Avramidis, M. Kyriakopoulou,
orthogonal step will be designing a distributed architecture m. c. schraefel, M. Vaitis, and D. Christodoulakis.
that propagates changes - in both directions - and that han- Structuring primitives in the callimachus
dles access permissions. component-based open hypermedia system. J. Netw.
Comput. Appl., 26(1):139–162, 2003.
In conclusion, a lot fascinating challenges await us on the [17] E. Wilde. XIPr: XInclude Processor.
horizon. It is our intention to foster the discussion within http://dret.net/projects/xipr/, 2007.
the community and to actually integrate Xanalogical func-
tionalities in a renewed generation of Xanadu-like systems.
6. REFERENCES
[1] K. M. Anderson, R. N. Taylor, and E. J. Whitehead,
Jr. Chimera: hypermedia for heterogeneous software
development enviroments. ACM Trans. Inf. Syst.,
18(3):211–245, 2000.
[2] L. Carr, D. De Roure, W. Hall, and G. Hill. The
Distributed Link Service: A Tool for Publishers,
Authors and Readers. The Web Journal, (1), 1995.
[3] S. DeRose, E. Maler, and D. Orchard. XML Linking
Language (XLink) Version 1.0.
http://www.w3.org/TR/xlink/, 2001. W3C
Recommendation.
[4] S. J. DeRose, E. Maler, and R. Daniel. XPointer
xpointer() Scheme. World Wide Web Consortium,
Working Draft WD-xptr-xpointer-20021219, December
2002.
[5] B. DuCharme. Transclusion with XSLT 2.0.
http://www.xml.com/pub/a/2003/07/09/xslt.html,
2003.
[6] W. Hall, H. Davis, and G. Hutchings. Rethinking
Hypermedia: The Microcosm Approach. Kluwer
Academic Publishers, Norwell, MA, USA, 1996.
[7] A. D. Iorio and J. Lumley. From XML Inclusions to
XML Transclusions. To appear in the Proceedings of
ACM Hypertext 09.
[8] J. Kolbitsch and H. Maurer. Transclusions in an
HTML-based environment. Journal of Computing and
Information Technology, 14(2):161–174, 2006.
[9] F. Manola and E. Miller. RDF Primer.
http://www.w3.org/TR/rdf-primer/, 2004. W3C
Recommendation.
[10] J. Marsh, D. Orchard, and D. Veillard. XML
Inclusions (XInclude) Version 1.0.
http://www.w3.org/TR/xinclude/, 2006. W3C
Recommendation.
[11] T. H. Nelson. Transliterature: A Humanist Format for
Re-Usable Documents and Media. http://translit.org/.
[12] T. H. Nelson. Embedded markup considered harmful.
World Wide Web J., 2(4):129–134, 1997.
[13] T. H. Nelson. Xanalogical structure, needed now more
than ever: parallel documents, deep links to content,
deep versioning, and deep re-use. ACM Comput.
Surv., page 33, 1999.
[14] S. Peroni and F. Vitali. EARMARKing documents for
arbitrary, overlapping and out-of-order annotations.
To appear in the Proceedings of ACM DocEng 09.
[15] J. K. Truss. International Organization for
Standardization. A.3 Architectural Form Definition
Requirements (AFDR). In In ISO/IEC 10744:1997,
Annex A, SGML Extended Facilities, pages
1601–1608, 1997.
28