Spring's migration to Maven has sparked some some interesting discussions. First Sylvain argues that Ant+Ivy is better. Then Carlos responds with the argument that its often better to hide details from beginners.

The most interesting comment comes from Leo Simmons, posting to the end of Carlos's page. His point is not that there's anything wrong with keeping stuff hidden, as long as it is accessible when you need it -because in any big project, you are going to need it. Leo doesnt think that Maven exposes its internals very well, for that moment when overriding the defaults becomes mandatory. I don't have enough Maven experience to comment there.

Scaling up builds is a really tough problem. You look at the fun Geronimo or Cocoon had, its about as pleasant as doing TSA-approved security checks non stop for a month. Big projects are complex, and Maven doesn't work out the box with them. Neither does Ant. I think in our SmartFrog build we've kind of reached the limit of Ant, and that is using the maven library tasks. I want to start using the Ivy tasks soon to see if they are easier to use.

The other thing I'm thinking of is doing more deployment-time build steps. Certainly EARs and WARs should be created on demand, with custom metadata XML files, targeted at specific systems. I dont think the other parts of the build (javac, jar, rmic) should be delayed until then, but certainly the final assembly step is a target machine/appserver customisation, which should be held off until you know the details.