Have you ever had a debate with someone only to find that they are
missing an important piece of knowledge? I humbly submit that the
concept of "mixed content" in XML is one such concept.
I have been involved in debates about XML processing techniques that
seemed to be going around in circles. More often than not, the
disagreement stemmed from a different conceptual model of XML processing
and, more often than not, that difference revolved around the important
concept of mixed content in XML. If one party to a debate sees it in
their mind-map of XML and the other does not, communication problems are
likely to ensue.
What follows is a debug trace of a conversation bug attributable to
mixed content myopia:
A: You know what XML is?
B: Sure. XML is a way of adding tags to text to denote the text's
meaning. Tags come in pairs -- one at the start and one at the
end.
A: Okay, but what about the text that is not tagged?
B: What do you mean, "not tagged"? If you add tags, you tag up all
the text; there is no text outside of the tagging. Adding tags
allows you to retrieve the values of those tags (i.e., the text
between the start-tag and the end-tag) using APIs.
A: So you think of tag names as field names then?
B: Yes. Furthermore, XML APIs should allow me to read the value of
any particular field.
A: Ah, okay. I see where you are coming from. Oh dear! Is that the
time? Must dash!
The fact is, not all text is guaranteed to be tightly tagged in XML.
Text occurring in, so-called, "mixed content models" can sit outside of
any direct tagging. Further, such freestanding text can be freely
intermixed with tags.
Such mixed content models offer both advantages and disadvantages. Being
a natural model for narrative text offers a big advantage. For example,
the p element in XHTML can contain mixed content as shown in the
following fragment:
<p>This is a para with a <a href="foo">link</a> in it</p>
The value of the a element is the text "link", but what is the value of
the p element? Is it the text before the a start-tag? The text before
and after the a element? Does it include the word "link", which is in
the a sub-element?
We could remove this ambiguity by fully tagging the text like this:
<p><text>This is a para with a </text><a href="foo">link</a><text>
in it</text></p>
The advantage of this style is that each element either contains a
fragment of raw text or a collection of other elements, but never a
- mixture* of both. On the downside, i is more complex and ugly to tag by
hand.
Now you may be inclined to think that some hairs are being split here.
Let us move over from doc-land to data-land and disprove this thought.
In the following XML fragment, what is the value of the <InvoiceNumber>
element?
<Invoice><InvoiceNumber>42</InvoiceNumber><InvoiceValue>$24.42</InvoiceValue></Invoice>
I think you will agree that the value of the InvoiceNumber element is
pretty unambiguously "42". Surely, XML APIs would provide some way to
retrieve this value in a single call to some programmer's library?
Something like GetValue("InvoiceNumber") for example?
This is where mixed content really bites. The most common APIs -- SAX
and DOM -- provide nothing like GetValue. The API cannot know whether or
not InvoiceNumber supports mixed content. If it does, no simple "value"
can be associated with it.
Understandably, database programmers stare blankly at explanations of
mixed content and think "this is Neanderthal isn't it?". They are right,
in part. Yes, the standard APIs not better supporting such a common use
case for XML is absurd; and yes, various attempts have been made to get
around it. My own stuttering attempt was RAX[1]. More recently, a common
API for XML Pull Parsing initiative has been formed [2].
Supporting pull APIs cleanly is made more complicated by the fact that
you cannot know in advance if an XML document will contain mixed
content. It is true that if you are validating the document as you
parse, you can detect the presence of mixed content in the content
models. Without the external indication, the poor parser has no option
but to anticipate the presence of mixed content.
One possible way out of it would be to allow XML instances to declare,
up front, that they do not use mixed content. That way, the parser could
be smart about the API it exposes to the programmer providing pull
features where possible.
Another way out of it -- and this appears to be the one gaining most
ground out there -- is not to bother using horizontal XML APIs for
non-mixed content data at all. XML data binding tools [3] move the
problem up a level by allowing the developer to work at the level of
native objects in her application and let the XML data-binding layer
look after the details of reading/writing objects/records.
NOTES
[1] http://www.xml.com/pub/a/2000/04/26/rax
[2] http://www.xmlpull.org/
[3] http://www.rpbourret.com/xml/XMLDataBinding.htm
