The content- and protocol-handler mechanisms I've introduced can be used by any application that accesses data via URLs. This mechanism is extremely flexible; to handle a URL, you need only the appropriate protocol and content handlers. To extend a Java-built Web browser so that it can handle new and specialized kinds of URLs, you need only supply additional content and protocol handlers. Furthermore, Java's ability to load new classes over the Net means that the handlers don't even need to be a part of the browser. Content and protocol handlers could be downloaded over the Net, from the same site that supplies the data, and used by the browser. If you wanted to supply some completely new data type, using a completely new protocol, you could make your data file plus a content handler and a protocol handler available on your Web server; anyone using a Web browser built in Java would automatically get the appropriate handlers whenever they access your data. In short, Java lets you build automatically extendible Web browsers; instead of being a gigantic do-everything application, the browser becomes a lightweight scaffold that dynamically incorporates extensions as needed. Figure 9.3 shows the conceptual operation of a content handler; Figure 9.4 does the same for a protocol handler.
Sun's HotJava was the first browser to demonstrate these features. When HotJava encounters a type of content or a protocol it doesn't understand, it searches the remote server for an appropriate handler class. HotJava also interprets HTML data as a type of content; that is, HTML isn't a privileged data type built into the browser. HTML documents use the same content- and protocol-handler mechanisms as other data types.
Unfortunately, a few nasty flies are stuck in this ointment. Content and protocol handlers are part of the Java API: they're an intrinsic part of the mechanism for working with URLs. However, specific content and protocol handlers aren't part of the API; the ContentHandler class and the two classes that make up protocol handlers, URLStreamHandler and URLConnection, are all abstract classes. They define what an implementation must provide, but they don't actually provide an implementation. This is not as paradoxical as it sounds. After all, the API defines the Applet class, but doesn't include any specific applets. Likewise, the standard Java classes don't include content handlers for HTML, GIF, MPEG, or other common data types. Even this isn't a real problem, although a library of standard content handlers would be useful. (JDK provides some content and protocol handlers in the sun.net.www.content and sun.net.www.protocol packages, but these are undocumented and subject to change.) There are two real issues here:
The next release of Sun's HotJava Web browser should certainly take full advantage of handlers, but current versions of Netscape Navigator do not. When the next release of HotJava appears, it may resolve these questions, at least on a de facto basis. (It would certainly be in Sun's interest to publish some kind of standard as soon as possible.) Although we can't tell you what standards will eventually evolve, we can discuss how to write handlers for standalone applications. When the standards issues are resolved, revising these handlers to work with HotJava and other Web browsers should be simple.
 Downloadable handlers will be part of HotJava 1.0, though they are not supported by the "pre-beta 1" release. The current release does support local content and protocol handlers. HotJava 1.0 also promises additional classes to support network applications.
The most common Java-capable platform, Netscape Navigator, doesn't use the content- and protocol-handler mechanisms to render Net resources. It's a classic monolithic application: knowledge about certain kinds of objects, like HTML and GIF files, is built-in. It can be extended via a plug-in mechanism, but plug-ins aren't portable (they're written in C) and can't be downloaded dynamically over the Net. Applets running in Netscape can use a limited version of the URL mechanism, but the browser imposes many restrictions. As I said earlier, you can construct URLs relative to the applet's code base, and use the openStream() method to get a raw input stream from which to read data, but that's it. For the time being, you can't use your own content or protocol handlers to work with applets loaded over the Net. Allowing this would be a simple extension, even without content- and protocol-handler support integrated into Netscape itself. We can only hope they add this support soon.