Best practices for command-line builds

Item 2 of the Joel Test is "Can you make a build in one step", but what is the best (or commonly accepted) way to achieve this? What are the pros and cons of using the IDE's command-line interface to do a headless build, as opposed to maintaining a build script that is completely independent of the IDE (e.g. using Ant or Maven in the case of a Java project).

I ask this question because I am experimenting with Maven, and was a bit surprised to find that even when using the m2eclipse plugin it's not really feasible to turn off Eclipse's Java Builder and delegate the whole build process to Maven while working on the project. It seems that if I want to migrate the project to Maven, I'll end up having to maintain two equivalent but different开发者_运维技巧 build processes, for example the command-line build will use the DataNucleus Maven plugin for enhancing JDO classes, while the IDE build will use the DataNuclues plugin for Eclipse; the command-line build will use the Tomcat Maven plugin, while the IDE will use Eclipse's web tools platform. This redundancy seems unfortunate.

I gather that NetBeans uses Ant for its build, which sounds like it would solve this problem. But unfortunately I'm using neither NetBeans nor Ant!

EDITED TO ADD: I found that I can set up a Maven builder in Eclipse with specific goals that call some of the Maven plugins that I'm using (e.g. in my case the goals "datanucleus:enhance process-resources" take care of JDO enhancement, resource copying, and native2ascii conversion). This leaves java compilation up to Eclipse, but still achieves some degree of integration with Maven.

The eclipse builder gives you intermediate and repeated building, but the "Joel Test" is really about being able to go from source to ready-to-deploy something in a single step. If you're using Maven, there's several ways of going about it - including just invoking Maven as a command line script from within Eclipse.

mvn package

Is what I most frequently used for a quick run through the whole build setup. You can extend Maven through the POM and inject some additional mechanisms if you want to. If you get a little more complex, it's often most effective to start using multi-module POM and maven setups to get functional tests integrated and running with the rest of the code.

Basically, let eclipse "do it's thing" with it's builders and take advantage of what it provides, but when you're ready to use the end product have it invoked from the command line through a continuous integration server (Hudson is a nice easy one to get and set up - free too: http://hudson-ci.org/). Presumably if you're looking at using Maven, you'll also have an instance of Archiva or Nexus set up as your DSL for the resulting libraries. You can have hudson invoke "mvn deploy" for regular checkin builds, or "mvn release:prepare && mvn release:perform" for when you're ready to cut the release (separate builds in Hudson work best for this)

You can do a headless build in an IDE. In eclipse you can execute an arbitrary shell command as an "External Tool". This is also true in IDEA and netbeans with a little effort. At the very least, this is a convenient test of the headless build that you should run whenever you make changes to the build configuration.

Also, I'd like to add that the build should be one-step as a minimum requirement. It should also be easy to set up and easy to debug. If it takes longer than an couple of hours to set up a new developers environment, then the one-step process is likely to be less than optimal. This is with the caveat that if you add or replace team members quarterly or annually this is less of an issue.