Referential Integrity
Now that we know we can accommodate all our data, let's look at whether this is really the best way to do it.
Anyone who works with relational databases is probably screaming "Finally!" This structure makes a lot of sense from an XML standpoint, where products are embedded within vendors, and so on, but in a typical enterprise application, this is not the way it's done.
We'll be dealing with traditional databases in Chapter 8, "Updating Inventory: SQL Databases and SAX," but let's take a moment to look at the implications or this comment method of data storage.
In a relational database, types of information are stored together, with keys that link them together. For instance, vendors and products would probably be stored in a structure similar to Figure 3.5.
){kind=link}
Figure 3.5 A typical relational database structure links data in one table to data in another.
In this case, vendors and products are separate database tables holding information on vendors and products, respectively. Each record, or item, is noted by a unique identifier, the ID. We can also tie the two tables together by requiring the vendor_id to match an id in the vendor table. This way, we know that each product belongs to a legitimate vendor.
But why bother doing this, when we can just put everything in one place, as we've done here? That is the basic question behind relational databases.
Previously, databases have been, essentially, flat files—meaning the data was stored together, as we have it here. There were several problems with this. For one thing, information was repeated more often than it needed to be. For instance, a flat file database table of Crazy Marge's data would look like Figure 3.6.
){kind=link}
Figure 3.6 A flat file of Crazy Marge's data would include a lot of duplicated information.
Notice that a lot of information, such as the vendor_name and advertisement, is listed on every line, even though it's exactly the same. Also, if we wanted to add a new pricetype, we would need to add it to every single row, even if it didn't apply to that particular product.
By separating out information, we can eliminate a lot of these problems, as illustrated in Figure 3.7.
){kind=link}
Figure 3.7 Breaking information into several tables makes it much easier to manage.
This way, the information becomes much cleaner and easier to manage. A change to a vendor's advertisement has to be made just once, not once for every product.
The key is to make sure that information is never lost because of bad data. If, perhaps, we specified a product's vendor_id of marge, that product would be lost to us, because it would never appear with an existing vendor. The idea is to implement it in such a way that the referring data (in this case, the vendor_id) always matches the appropriate key (in this case, the id). This is known as referential integrity.
We can implement a structure like this using DTDs.
We're not going to actually use the product to do this, however. Although it is likely that a relational database would be set up that way, we are, in fact, dealing with XML. In short, just because we can do something, doesn't mean that we should. Programming for databases is very different from programming for XML. Moving product out from under vendor would be a nightmare for us later, so, instead we'll examine this issue by moving the specials.
In Listings 3.14 and 3.15, we're pulling the special out from under vendors and dropping it into the root element but linking it back to its vendor using IDs.
Listing 3.14 products.xml: Moving Specials Out to the Root Element
0:<?xml version="1.0"?> 1:<!DOCTYPE products SYSTEM "products.dtd"> 2: 3:<products> 4: 5:<vendor webvendor="full" id="conners"> 6: <vendor_name>Conners Chair Company</vendor_name> ... 68:</vendor> 69: 70: <vendor webvendor="partial" id="wally"> 71: <vendor_name> 72: Wally's Wonderful World of Furniture 73: </vendor_name> ... 148:</vendor> 149: 150:<vendor webvendor="no" id="marge"> 151: <vendor_name>Crazy Marge's Bed Emporium</vendor_name> 152: <advertisement> 153: <ad_sentence> ... 218: </giveaway_item> 219: <giveaway_desc> 220: with free delivery -- we'll take your 221: old mattress as a trade in! 222: </giveaway_desc> 223: </giveaway> 224: </product> 225: 226:</vendor> 227: 228:<special specialtype="weekly" vendor_id="marge"> 229: This week only: Round beds with rotating motors 230: starting at a price that will make your head spin. 231: Just talk to Crazy Marge, she'll tell you all about it! 232:</special> 233: 234:</products>
Listing 3.15 products.dtd: Accommodating the Changes in the DTD
0:<!ELEMENT products (vendor | special)+> 1: 2:<!ELEMENT vendor (vendor_name, advertisement?, suite?, product*)> 3:<!ATTLIST vendor webvendor CDATA #REQUIRED> 4: 5:<!ELEMENT special (#PCDATA)> 6:<!ATTLIST special specialtype CDATA #FIXED 'weekly'> 7: 8:<!ELEMENT vendor_name (#PCDATA)> ...
Considering the magnitude of the structural change we just made, we really didn't have to change the DTD all that much. The first thing we did, on line 0, was to accommodate the changes to the root element. Note that syntax is important here. The way we have it written, we can have one or more vendors or specials, in any order. We could have written this definition as follows:
<!ELEMENT products (vendor+, special+)>
This would have required us to list all the vendors first, and then all the specials. Although this is probably closer to what would happen in this situation, we don't want to require it.
Now that we've got the files reorganized, we need to tell the DTD that we want to enforce the referential integrity. This means that any value in vendor_id must match an existing vendor id attribute. We do this using two new datatypes, ID and IDREF, as in Listing 3.16.
Listing 3.16 products.dtd: Adding IDs and IDREFs
<!ELEMENT products (vendor | special)+>
<!ELEMENT vendor (vendor_name, advertisement?, suite?, product*)>
<!ATTLIST vendor webvendor CDATA #REQUIRED
id ID #REQUIRED>
<!ELEMENT special (#PCDATA)>
<!ATTLIST special specialtype CDATA #FIXED 'weekly'
vendor_id IDREF #REQUIRED>
...
We're actually doing two things with this listing. First, we're creating a primary key on vendor called id by setting it to type ID. A primary key is a value that must be present and must be unique. No two vendors may have the same value for id. Second, we're creating an attribute of type IDREF. This means that its values will have to match an existing value of an ID attribute (in this case, the vendor id attribute).
The Second Limitation: Keys
Now, this is extremely handy and goes a long way toward emulating the relational database structure, but it doesn't go quite far enough.
For instance, it would be natural (and in fact, expected) to put a primary key on the id attributes for product and suite, but this causes two problems.
First, whereas primary keys are most commonly numeric, IDs in XML must be Names—they must start with a letter. There's no particular reason for this except that it's what's specified in the Recommendation.
Second, you'll notice that there's no way to tell an IDREF which ID it's supposed to be referencing. In fact, each document has only one "pool" of IDs. For instance, if we specified that the id attribute for products was an ID, we'd find that a product and a vendor can't have the same ID because they would no longer be unique (even though they're different elements). What's more, a special could have a vendor_id that matches the product but not a vendor, and the processor will not see it as a problem.
The Third Limitation: Same Name, Different Element
Another limitation of DTDs stems from the fact that each element is declared in a vacuum, so to speak, without reference to the elements that contain it. For instance, we have products that are part of the root element, and we have products that are part of a suite. It is not inconceivable that these two products might have different requirements, but within the structure of DTDs, the only way to define them differently would be to give them different names.
Fortunately, all three of these limitations can be overcome using XML Schema.