Developers tool choices used to be made with a stopwatch, back when compile times were measured in minutes and desktop PC speed was the clients limiting resource. Todays performance bottlenecks often reside at a distance, whether in network bandwidth limits, in opaque run-time environments or in server-side code.
Should developers still use tools that let them write the fastest-possible code, even if that means spending costly programmer hours to enforce good behavior? Or should they take advantage of the higher productivity that comes from Javas elimination of many common sources of programmer error?
Borland Software Corp.s JBuilder (reviewed above) was one of the first developer tools to exploit what could be known about a Java application, parsing code dynamically for feedback during development. Oracle Corp.s JDeveloper now gives JBuilder a run for its money in this regard, but so does Microsoft Corp.s performance-oriented C#. The choice of productivity or performance is no longer a choice between Java and everything else.
What remains is the tension between writing code that takes corners on two wheels, with facilities such as unsafe pointer operations, and using tools that do more to keep the code on the road—while avoiding the problem that "it is very easy to write slow Java programs," as warned by software performance guru Dov Bulka, performance architect of IBMs Domino-Go Web server, in the preface to his book "Java Performance and Scalability, Volume 1: Server-Side Programming Techniques." Bulka particularly warns developers to "keep a watchful eye on the number of objects our code generates on the performance-critical path."
Overall, hardware performance trends put eWeek Labs on the side of making code do the right thing before worrying about how fast it does it. Developers should take the time to learn what makes code slow, but its better to have a list of performance improvements to make in code that works than to wonder why it falls apart so quickly.