Design FAQ for the Proposed Javatm Preferences API

Here are a few frequently asked questions about the proposed Preferences API that provide a rudimentary design rationale for the API:

  1. How does this API relate to Properties?

    It is intended to replace most common uses of Properties, rectifying many of its deficiencies, while retaining its light weight. When using Properties, the programmer must explicitly specify a pathname for each properties file, but there is no standard location or naming convention. Properties files are "brittle", as they are hand-editable but easily corrupted by careless editing. Support for non-string data types in properties is non-existent. Properties cannot easily be used with a persistence mechanism other than the file system. In sum, the Properties facility does not scale.
     

  2. How does this API relate to JNDI?

    Like JNDI, it API provides back-end neutral access to persistent key-value data. JNDI, however, is far more powerful, and correspondingly heavyweight. JNDI is appropriate for enterprise applications that need its power. This API is intended as a simple, ubiquitous, back-end neutral preferences-management facility, enabling any Java application to easily tailor its behavior to user preferences and maintain small amounts of state from run to run.
     

  3. Why do all of the get methods require the caller to pass in a default?

    This forces the authors of applications to provide reasonable default values, so that applications have a reasonable shot at running even if the repository is unavailable.
     

  4. How was it decided which methods should throw BackingStoreException?

    Only the methods whose semantics absolutely require the ability to communicate with the backing store throw this exception. Typical applications will have no need to call these methods. As long as these methods are avoided, applications will be able to run even if the backing store is unavailable, which was an explicit design goal.
     

  5. Why doesn't this API provide stronger guarantees concerning concurrent access by multiple VMs? Similarly, why doesn't the API allow multiple Preferences updates to be combined into a single "transaction", with all or nothing semantics??

    While the the API does provide rudimentary persistent data storage, it is not intended as a substitute for a database. It is critical that it be possible to implement this API atop standard preference/configuration repositories, most of which do not provide database-like guarantees and functionality. Such repositories have proven adequate for the purposes for which this API is intended.
     

  6. Why does this API have case-sensitive keys and node-names, while other APIs playing in a similar space (such as the Win32 Registry and LDAP) do not?

    In the Java programming universe, case-sensitive String keys are ubiquitous. In particular, they are provided by the Properties class, which this API is intended to replace. It is not uncommon for people to use Properties in a fashion that demands case-sensitivity. For example, Java package names (which are case-sensitive) are sometimes used as keys. It is recognized that this design decision complicates the life of the systems programmer who implements Preferences atop a backing store with case-insensitive keys, but this is considered an acceptable price to pay, as far more programmers will use the Preferences API than will implement it.
     

  7. Why doesn't this API use the Java 2 Collections Framework?

    This API is designed for a very particular purpose, and is optimized for that purpose. In the absence of generic types (see JSR-14), the API would be less convenient for typical users, and would lack compile-time type safety, if were forced to conform to the Map API. Further, it is not anticipated that interoperability with other Map implementations will be required (though it would be straightforward to implement an adapter class if this assumption turned out to be wrong). The Preferences API is, by design, so similar to Map that programmers familiar with the latter should have no difficulties using the former.
     

  8. Why don't the put and remove methods return the old values?

    It is highly desirable that both of these methods be executable even if the backing store is unavailable. This would not be possible if they were required to return the old value. Further, it would have negative performance impact if the API were implemented atop some common back-end data stores.
     

  9. Why does the API permit, but not require, stored defaults?

    This functionality is required in enterprise settings for scalable, cost-effective administration of preferences across the enterprise, but would be overkill in a self-administered single-user setting.
     

  10. Why doesn't this API contain methods to read and write arbitrary serializable objects?

    Serialized objects are somewhat fragile: if the version of the program that reads such a property differs from the version that wrote it, the object may not deserialize properly (or at all). It is not impossible to store a serialized objects using this API, but we do not encouraged it, and have not provided a convenience method.
     

  11. Why is Preferences an abstract class rather than an interface?

    It was felt that the ability to add new methods in an upward compatible fashion outweighed the disadvantage that Preferences can not be used as a "mixin (that is to say, arbitrary classes cannot also be made to serve as Preferences objects. Also, this obviates the need for a separate class for the static methods. (Interfaces cannot contain static methods.)
     

Copyright 2000 Sun Microsystems, Inc., 901 San Antonio Road, Palo Alto, California 94303 U.S.A. All rights reserved.