Overview

1
Overview

• Some hyperlinking history
• XML-Linking goals and status
• XML Base
• The XPointer language
• The XLink language
• Controlling traversal behavior

2
Some Hyperlinking History

3
History repeats itself

• In word processing
• Early systems knew text structure
• WYSIWYG took over but skipped structure
• As it scaled, people started to fake structure
• Software gradually built structure all over again

• In hypermedia
• Many early systems had linking structure
• HTML ltAgt took over but skipped structure
• As projects scale, people are faking link
  structure
• So

4
Hypermedia in the 60s (!)

• For example, FRESS (Brown, 1969)
• 800,000 lines of 370 assembler -- still runs
• Grandfather of most Word processors
• Teaching databases typesetting
• Some features still not on Web
• Links were structured w/ types, keywords
• Links were truly bidirectional
• Links overviews
• Supported document structures / ranges
• Out-of-line links later (1983, Intermedia)

5
HTML and hypermedia

• WWW is powerful because of Internet, URIs, and
  documents linking is weak
• HTML linking strengths
• Easy syntax lta href"http//foo.net/x.html"gt
• Trivial to implement
• Can go to few specific targets (ID only)
• lta href"http//foo.net/x.htmlsec2"gt
• lta name"sec2"gt
• Anchors move along with editing
• (unlike, say, byte offsets)

6
Anatomy of an HTML link
lta href"http//z.uk/x.htmch2"gtInfo.lt/agt

Locallink-end
Locator
Link
lt!DOCTYPE lthtmlgtlta name"ch2"gtlt/h1gt...lt/htm
lgt
Remotelink-end
7
HTML Linking Limitations

8
1 Limited elements

• People cant create their own subtypes
• What if you need several sub-types of link?
• REL/REV attributes non-standardized
• CLASS attributes dont make it
• Semantics/processing second-class
• Maybe no HTML element to start from
• Its like C without subclasses
• CSS cannot make elements into links

9
2 Link behavior tied to type

• A means (usually)
• Replaces document in same window
• Activated by user
• IMG means (usually)
• Embeds target inline
• Activated upon loading
• Graphics only
• Why cant I have other combinations?
• CSS doesn't have what youd need
• Browsers cant embed HTML/XML yet

10
3 No real type system

• Few have tackled the link type problem
• Trigg, Bieber, DeRose, Connolly
• REL/REV provide a hook, at least
• No standard/portable set of types
• Need an orderly but extensible system
• Most proposed type-sets are grab-bags
• Mixing rhetorical, topological, and syntactic
  types

11
4 A privileged end

• The link is physically at one of its ends
• Infeasible to allow annotation
• Thus no links from read-only documents
• Nearly all documents, for any one person
• So no annotations that readers of the annotated
  document can see
• Doesnt allow link databases, so
• Cant buy or sell link-sets
• Cant filter or select links on demand
• Cant share many users links

12
5 Few possible destinations

• Can point to a whole document
• Can point to HTML lta namegt if it existslta
  href"http//xyz.com/foo.htmlid37"gt points
  to lta name"id37"gt
• Author controls what others can reference
• You can only quote paras 1, 12, or 20.
• Can't do some links at all
• Impossible to supply handles for links that
  partly overlap
• Impractical to supply handles even at word level

13
6 No aggregate destinations

• No multi-ended links in HTML
• No multiple locations for a single end
• No dynamic document assembly
• Some uses for multi-ended links
• Back-of-book indexes
• Pull up multiple commentaries at once
• Editing after multiple document reviews
• Parallel views of editions and translations

14
7 One way A or IMG to somewhere

• Why cant I ask who links to me?
• (without doing a global search)
• (even for a limited domain)
• Why cant I make a two-way link?
• Why cant I make multi-ended links?
• Go back ? 2-way link
• Only works after youve gone the one way
• You cant start at the other end

15
Overview

• Some hyperlinking history
• XML-Linking introduction
• Goals
• History
• Pointing vs. Linking
• XML Base
• The XPointer language
• The XLink language
• Controlling traversal behavior

16
XML-Linking goals end user

• Links from un-writable documents
• Which is most of the Web, for any person
• Perhaps the most important single feature
• -gtBidirectional and multi-ended links
• -gtAnnotations and annotation sharing
• Dynamic updates, patches, highlighting
• Precise link attachment in any media
• Large sets/databases of managed links
• An entirely new market for links per se
• Anyone can publish/sell their commentary

17
The XML-Linking effort

• Started 1/1996 in the XML WG
• Separate Working Group 1998
• Specs
• XPath complete 11/1999 (joint w/ XSL WG)
• XLink complete 6/2001
• XML Base complete 6/2001
• XPointer being refactored
• first 3 documents out soon

18
Pointing vs. linking

• In HTML, many things are combined
• lta href"eg.org/foo"gtwowlt/agt
• Technically
• "eg.org/foo" is a pointer (namely a URI)
• The abstract connection itself is the link
• The ltagt element is a link representation
• "wow" is the local anchor
• Anchors are also called link-ends
• Data at eg.org is the remote anchor
• HTML specifies the link behavior

19
Why pick all this apart?

• Engineering/theory
• Design each one better
• Ease implementation
• More easily combine them in new ways
• Practically
• Multiple locations can be linked at once
• Link-ends can be organized and labeled
• Meta-information can be added
• Links can live outside of linked docs

20
XML-Linking specifications

• XPath expressions on infoset nodes
• REC http//www.w3.org/TR/xpath
• XPointer XPath ranges, in URIs
• CR http//www.w3.org/TR/WD-xptr
• Xpointer Schemes failover for fragment Ids
• Xpointer NS scheme (declare namespaces)
• Xpointer Element scheme
• XLink gather locations to make links
• REC http//www.w3.org/TR/xlink/
• (XML Base)

21
XPointer locators
ltxmlgt ltxref target"http//z.com/foo.xmlid('p3
7')"gtSee Section 1.lt/xrefgt
A way of locating data in XML structure used to
attach link end(s) to data
A pointer identifies or locates some part of a
document -- this is only the yellow part above
22
XLink connections
Someplace
Someplace

• Describes a relationshipof referenced
  location(s),
• To each other
• To descriptions
• XLink providessome key ones

A link connects data and meta-data portions,
including their relationship -- really just the
lines
role
role
role
A link may be expressed at a unique source end,
or out in a link database
Someplace
Someplace
Someplace
23
Overview

• Some hyperlinking history
• XML-Linking introduction
• XML Base
• The XML Pointer Language (XPointer)
• Fragment identifiers
• XPointer schemes
• XPointer syntax/semantics
• The XML Link language (XLink)
• Controlling traversal behavior

24
XPointer

• Locates parts of XML resources
• Even things without IDs
• Even things that aren't whole nodes
• XPointer adds (beyond XPath)
• Way to refer to point and range selections
• Way to use inside URI fragment identifiers
• Typically, a browser might load a document and
  scroll to/highlight the part

25
Anatomy of a URI reference
URI reference

URI
http//example.com/foo.htmbing
domain
path
fragment identifier
scheme
XPointer defines this part
26
Fragment identifiers

• Part of URIs after ""
• Says where in document is actual target
• Separate form for each media type
• Identifiers for graphics ? for text
• IETF MIME definition specifies form
• HTML
• To scroll to lta name"coyote"gt
• http//example.com/hello.htmlcoyote

27
The XPointer scheme system

• A way to have several pointing mechanisms in one
  fragment ID
• Each scheme in a fragment is tried in sequence
• scheme1(args) scheme2(args)
• unimplemented schemes and pointers that fail to
  locate, are simply ignored.
• recent changes
• Things are more split by scheme. Full XPointer
  delayed (again)
• Last call drafts imminent

28
The 3 XPointer/XPath schemes

• Bare names
• An XML "name" finds element with that ID
• Only exception to the scheme system
• element() scheme
• Counts Stepwise down through elements /1/4/27/2
• May start with an ID element(intro/4/3/2)
• xmlns() binds namespaces for schemes
• xpointer scheme (awaits disposition of comments)
• Full Xpointers, xpath and more
• For now, the only "scheme" is "xpointer"

Name Letters, digits, hyphen, underscore,
period.
29
All work on XML tree

• Address Infoset, not markup
• Count nodes of different types
• Root, elements, attributes, text chunks
• Comments, PIs, namespace dcls
• Can also work down in character data
• Note Most of XPointer XPath
• Adds ranges
• Adds way to put into fragment identifier

30
XPointer's 2 parts

• Provide 'scheme' mechanism
• Identify media-specific pointer types
• Allow multiple ones to co-exist
• Pointing methods for XML
• Point to ranges, sets, id's, coords
• Point descriptively

31
XPointer schemes

• XML medias type may need pointer types
• pngRect(0,10 100,200)
• vrml(camera1,2,3 light4,50,500)
• map(W0?10/ N51?30)
• Schemes label fragment identifier types
• scheme1(args) scheme2(args)
• Escape any extra ( ) -- tlg('(apax')
• element(), xmlns() is the first scheme

32
Multiple schemes in a URL?

• When a server responds to a URI, it
• Checks what media the client can handle
• Picks one of those to send
• content negotiation
• If a visually-impaired user clicks
• lta href"http//www.example.com/foo.gif gif(0,0
  1,1) xpointer(id(chap1))"gt
• The server may fall back to an XML file
• The client tries fragment identifiers
  left-to-right, and uses the first one that works

33
Full XPointers

• The content of the 'xpointer' scheme
• Provides way to walk around trees
• Provides way to select nodes/ranges

34
XPath/XPointer expressions

• XPointers produce a location set
• Location node point range(XPath only
  produces nodes)
• Locating is stepwise
• A step generally look along some axis
• Candidates are then filtered by predicates
• Special functions locate strings, ids, etc.
• Each step operates on a context, and
• Step are separated by a slash
• id(foo) / childSEC3 / childLIST4

35
Points and Ranges
Hello, world.

• Point
• What you get by click-selection
• Gap before/after node or char
• Range
• What you get by drag-selection
• From a start point to an end point
• Not generally a WF XML subtree
• May partially contain some elements
• ltpgtHello, world.lt/pgtltpgtHi, backlt/pgt
• Crucial for creating hypertext links
• How often do you click/drag exactly one entire
  element?

Hello, world.
36
Absolute linking functions

• origin()
• Locates where traversal came from
• So if paragraphs P1 and P2 are linked, and the
  user clicks on P1 to follow a link, P1 is
  origin
• Can make abstract links, "next chapter"...
• here()
• Locates the link representation itself
• With one-way links like HTML ltagt, here origin
• Can make a web-ring of one-ended links

37
here() and origin()
href"xpointer(here()/ following-siblingchapter
)"

origin()/ancestor2
Assuming you clicked on the title
Assuming you clicked on the xref
href"..."
38
Summary axes and functions

• root( ), id( )
• parent, self, child
• ancestor, ancestor-or-self
• descendant, descendant-or-self
• preceding-, following-sibling
• preceding, following
• attribute, namespace
• here( ), origin( )

Absolute
Relative
Absolute
39
More about ranges

• Nodes are great handles for linkingbut most
  selections are not whole nodes
• Typical link creation
• Select something do "make link"
• A click indicates a point, not a node
• A drag indicates a range, not a node
• A range is not a set of nodes (!)
• It is one object, not a huge number
• Characters are not nodes in DOM
• Range does not have properties of node

Defined in the DOM 2 spec
40
string-range

• string-range(n,"string", offset,length)
• Locates selections (think cursor)
• Finds nth occurrence of string provided
• Offset counts positions before characters
• 1 is before first character -1 before last
• Length specifies length of result
• In the cards are dealt string-range(1,"cards"
  ,1,1) c(1,"cards",1,0) point before
  c(1,"cards") point before c(1,"cards",3,2)
  rd(1,"cards",-1,5) s are

41
range-to

• Range-to(xptr)
• From start of context to end of xptr
• The xptr would typically be an ID or relative
• id('sec2')/range-to(id('sec4'))
• id('fn37')/range-to(following-sibling3)
• A typical form
• xpointer( id('sec2.1')/4
  range-to(id('sec2.1')/2))

42
Some range applications

• A typical hypertext interface
• User selects some text
• (a selection is hardly ever a node)
• "make link", "make annotation"
• Application generates an XPointer
• Application saves links in a link-db
• Great for referring to "milestones" or other
  non-hierarchical units

43
XPointer new datatypes

• Extending nodes
• Points
• Ranges
• Locations
• The supertype of the extensions.
• Location sets
• set of locations

44
Reducing pointer/link fragility

• Consider what will break an XPointer
• If you just used offsets (easy but weak) the
  slightest edit
• If you describe paths through the tree, only
  changes on the path
• If you use paths with element type names, only
  ancestors or pr-siblings of same type
• If you use IDs, only a specific change to
  that ID
• When you can describe the true intent, they can
  re-attach when possible
• Use IDs, or relative from an ID

45
XPointer robustness

• Absolute robustness is not possible
• Servers go down
• Pages are deleted or moved
• Author can mangle documents any time
• Author can even change all IDs any time
• But, how you construct pointers counts
• IDs are good until explicitly changed
• Software can manage IDs to be persistent
• These are not true of counters, offsets, etc.

46
Overview

• Some hyperlinking history
• XML-Linking goals and status
• XPath and XPointer in general
• The XML Pointer language (XPointer)
• The XML Linking Language (XLink)
• Terminology
• Inline and out-of-line links
• Arcs and behavior
• Linksets and link databases

47
Remember...

• In style
• XPath identifies set of nodes
• XSLT itself does transforms on them
• Similarly in linking
• XPointer identifies sets of locations
• XLink connects and describes them
• These are very separate ideas

48
XLink is a language that...

• Lets you invent your own linking elements and
  their meanings
• In keeping with XML approach overall
• Lets you create link databases
• Links become first-class objects in the model
• Provides some basic traversal behavior
• E.g., Open the target in a new window
• The rest is left to a style mechanism such as XSL

49
XLink terminology

• Linking element
• Identifies, connects, and describes anchors
• Locator
• Locates some link end (anchor)s data
• Link end or anchor
• A data portion reachable as part of a link
• Arc
• Explicit connection between two link ends
• Resource
• Anything you can point at on the Web
• Using an arc is called Traversal

50
What links do with link-ends

• A link identifies where its ends are
• Using some kind of locators
• URIXPointer will be the locator for XML
• URIscheme()scheme() in general
• A link attaches metadata to each end
• Its formal role in relation to the other ends
• A title by which to refer to it (say, in menus)
• Some traversal behaviors
• Arcs to say which traversals happen
• Link itself can also have type, other info

51
Inline links

• Linking element itself (better, the origin() end)
  is one of the links ends

52
Out-of-line links

• Linking element itself isn't automatically made
  into one of its own resources

Requires that there be a way to find link
databases in the first place
53
Anatomy of an XML link
Link need not beat a link-end
lthtmlgtKnuths right.lt/htmlgt
ltlink type"annotated-reference"gt ltloc
role"ref" href"xptr.xmlchild(2,div)"gt ltloc
role"src" href"knut73.texs4.2.2"gt ltloc
role"com" href"http//x.com/note.html"gt ltlinkgt
Each link-end can be described
Link may have any number of ends
lt!DOCTYPE spec...ltspecgtltdivgtlt/divgtltdivgt ltheadgt.
..lt/headgt...
Link-ends need not be XML
\ 4.2.2 A tree is a set of nodes where each
node has one parent, except for a root node,
which has none.
Link-ends need not be marked up
54
XLink and namespaces

• Xlink should be usable with any schema
• And have any old name
• Xlink elements are in a namespace
• http//www.w3.org/1999/Xlink
• Apply this in usual namespace way
• ltxlinklocator xmlnsxlink "http//www.w3.org/19
  99/Xlink"gt
• Namespaces can't rename, so true name goes on
  xlinktype attribute
• This is much like architectural forms

55
Simple and extended links

• A simple link is a
• Starter kit of basic linking functionality
• Only two resources are ever involved
• It is always inline
• An extended link is a
• Generalized, all-purpose linking construct
• Any number of resources can be involved
• It might be inline or out of line
• It enables link databases

56
The one-ended link
This is really important
Hilite'red'

• A link can identify and describe
• Can say the end "is a typo" or "important" or
  "highlighted in red"
• A one-ended link still describes an end
• Can attach tags or attributes to R/O data
• ltclaim href"hamletid(soliloquy)" isacrux/gt
• The description isnt properly an end
• If it were, youd expect to navigate there
• If it were, it would need a role and title
• Critical for adding "standoff" markup

57
Arcs

• Arcs specify traversal rules
• Multi-ended links may restrict travel among their
  endpoints
• Restrictions generic or app-specific
• Arcs enable the description of both
• An arc is a pair of roles, plus metadata
• Enables traversal between ends with the given
  roles
• May be multiple locators per role (useful for
  document assembly, multiple-choice travel)

58
Example vehicle annotations
Warning explosive
Warningtoxic
gasoline
fuel-type
warning
warning
ARCS vehicle ? fuel-type fuel-type ? warning
Link body 1
vehicle
vehicle
vehicle
59
How to detect links

• Could have any name and content at all
• ltfootnotegt, ltcriticismgt,
• xlinktype attribute marks linking elements for
  applications to find
• lt!ELEMENT footnote EMPTYgtlt!ATTLIST footnote
  xlinktype CDATA FIXED "simple"
  xlinkhref CDATA REQUIREDgt
• For example ...has studied the issue.ltfootnote
  href"http//www.doctools.com" /gt

Defaultvalue forattribute
60
Types of links

• simple easy, basic linking element
• Creates link solely by means of attributes
• So allows only one non-local end (cf ltAgt)
• extended
• Creates link using attributes / subelements
• Each sub-element represents an end
• This allows structured or multi-lingual titles
• Both can have
• Machine-readable role attribute
• Human-readable title attribute

61
Simple links

• Like HTML's ltagt
• Has two participating resources
• Linking element itself (if in-line)
• Resource pointed to by the href locator attribute
• Uni-directional
• From simple
• To remote resource

62
Simple link declaration
Content may be an end

• lt!ELEMENT simple ANYgt
• lt!ATTLIST simplexlinktype CDATA FIXED
  "simple"xlinkhref CDATA REQUIRED xlinkr
  ole CDATA IMPLIEDxlinktitle CDATA IMPLIEDgt

Identifies this as a link
Locatesresource
Describeother end
63
Extended links

• Can be in- (like ltagt) or out-of-line
• Have any number of resources
• Each resource gets its own child locator
  element(s)
• Link ends are distinguished by roles
• Each locator has an href attribute
• Multi-directional
• Can start traversal from any resource
• Can control allowed traversals via arcs
• Application must be informed about the link
• E.g., user subscribes to link database XYZ

64
The extended link container

• lt!ELEMENT extended (locatorarctitle)gt
• lt!ATTLIST extendedxlinktype CDATA FIXED
  "extended"xlinkrole CDATA IMPLIEDxlinktitl
  e CDATA IMPLIEDgt
• Other content/attributes may be included if
  needed

Allows for the links ends
Identifiesas a link
Can supplydefaultsfor ends
65
An extended links ends
Identifies this as a locator

• lt!ELEMENT locator EMPTYgt
• lt!ATTLIST locatorxlinktype CDATA FIXED
  "locator" xlinkhref CDATA REQUIRED
• xlinkrole CDATA IMPLIEDxlinktitle CDATA
  IMPLIEDgt
• Link end locators may also have content and/or
  application-specific attributes

Locatesresource
Describethe end
66
Declaring Arcs
Identifies this as an arc

• lt!ELEMENT go ANYgt
• lt!ATTLIST goxlinktype (arc) FIXED "arc"
• xlinkfrom NMTOKEN IMPLIEDxlinkto NMTOKEN I
  MPLIEDxlinkshow (newreplaceembedunknown)
  IMPLIED xlinkactuate (onLoadonRequestunknown)
  IMPLIED xlinkarcrole NMTOKEN IMPLIEDxlinkt
  itle CDATA IMPLIEDgt

Which roles to connect
Traversal semantics
Describethe arc
67
Arcs and Traversals

• Traversal is split into
• Behavior
• Author's intention for behavior of a link.
• Input to style mechanism
• Not a presentation command
• Actuation
• Defines the event that triggers a link
• Events are very generic, intentionally

68
Two kinds of behavior policies

• show attribute
• new to traverse and provide new context
• replace to display in existing context
• embed to display in the body of the initiating
  resource
• Some semantic details are left unspecified
  combining multiple ends, style inheritance, etc.
• actuate attribute
• onRequest to require external request
• onLoad to traverse when link processed

69
Replacement behavior

• replace plus onRequest
• Like clicking on the HTML ltAgt element
• replace plus onLoad
• Like a redirect when the link is processed, the
  other resource replaces the original one in which
  the link was found

70
New behavior

• new plus onRequest
• Like clicking on the HTML ltAgt element, except
  that it opens up a new window
• new plus onLoad
• Useful for revealing commentary when the link is
  processed, a new window opens up in addition to
  the original one

71
Embedding behavior

• embed plus onRequest
• Useful for expanding a thumbnail or when
  auto-loading of images is off when the user
  clicks, the content appears in place
• embed plus onLoad
• Like the HTML ltIMGgt element or an external entity
  reference the content automatically appears in
  place when the link is processed
• This is transclusion
• True transclusion requires that the transcluded
  data also provide access to original context

72
Simple link syntax

• lt!ELEMENT defref (PCDATA)gtlt!ATTLIST
  defref xlinktype CDATA "simple" xlinkhref
  CDATA REQUIRED xlinkrole CDATA
  "mynscooking"gtltpgtuse any number of ltdefref
  xlinkhref"filterdef"gt filters lt/defrefgt
  (about 1 tsp)lt/pgt

Note can add "FIXED" before default value so it
will be the only value allowed.
73
Extended links

• lt!ELEMENT term-def (locator)gt
• lt!ATTLIST term-def xlinktype CDATA
  "extended"gt
• lt!ELEMENT locator EMPTYgt
• lt!ATTLIST locator
• xlinktype CDATA "locator"
• xlinkhref CDATA REQUIRED
• xlinkrole (mynsmentionmynsdef) REQUIREDgt
• ltterm-defgtltlocator href"myterm1"
  role"mention"/gtltlocator href"myterm2"
  role"mention"/gtltlocator href"definition
  role"def"/gt
• lt/term-defgt

74
Link databases let you

• Attach descriptive information from afar
• Annotate other people's stuff
• Maintain links more easily
• When a destination changes, you dont have to
  touch documents with links to it
• Engage in online commerce in links
• Express, package, and sell point-of-view
• Collect out of line links as databases

75
External Linksets

• Users will have persistent linkdbs
• Subscriptions, interest groups, private,...
• Document can specify relevant link dbs
• Linked by special type of extended link
• Included within regular documents too
• LinkDBs enable link management
• Needed to author using external links
• Example Public annotations on.

76
Declaring an External Linkset

• lt!ELEMENT xls (linkbase)gt
• lt!ATTLIST xls
• xmlnsxlink CDATA FIXED "http//www.w3.org/
  1999/xlink"
• xlinktype (extended) "extended"
• xlinkrole NMTOKEN "xlinkextended-linkset"
• xlinktitle CDATA IMPLIEDgt
• lt!ATTLIST linkbase
• xlinktype (locator) "locator"gt

77
An external Linkset Instance

• ltxlsgtltlinkbase xlinkhref"linkset1.xml"
  /gtltlinkbase xlinkhref"linkset2.xml"
  /gtltlinkbase xlinkhref"linkset3.xml" /gt
• lt/xlsgt

78
Notes on implementation

• Best thing for any standard
• Lots of implementations
• Attention to conformance/interoperability
• IE does not yet have these
• Encourage implementors
• XPointer easy on top of XPath DOM
• XLink easy for inline, trickier for good LDBs
• Proxy server can merge document links
• Client app can do likewise
• Encourage distributed linkbase dev

79
Some Resources

• On the Web Consortium site
• W3C XML site www.w3.org/XML/
• Other specs www.w3.org/TR/
• Elsewhere on the Web
• XML FAQ www.ucc.ie/xml/
• XML Developers list www.lists.ic.ac.uk/hypermail
  /xml-dev/
• TEI etext.virginia.edu/
• Robin Covers definitive site www.oasis-open.org
  /cover/