On Tue, 26 Oct 2004, Stefano Mazzocchi <[EMAIL PROTECTED]> wrote: > Niclas Hedhman wrote:
>> Ant projects are treated according to a template of classpath >> injection, BUT some projects do their own downloads, and I wonder >> if there are some that actually bypasses the Gump intentions. > > Yes, this is where I think Maven surpasses Ant in design, in respect > to Gump integrability: ant is a lot less declarative. which means > it's much easier for people to do things that we don't have control > over from the outside. > > Stefan, thoughts? I trust build.sysclasspath a lot 8-) Seriously, in all built-in Ant tasks you'll have a very hard time to defeat Gump's intentions. And custom tasks will have to explicitly avoid Ant's own infrastructure (like AntClassLoader) for that. We have a bunch of project pulling down dependencies from all over the place and the result is just ignored by Ant in Gump. The main exception is the manifest's Class-Path attribute, but as long as you use URLClassLoader (and why would you want to rewrite that) there is nothing much you can do. This is not a comment on whether Ant or Maven is better designed for any purpose. I just want to say that it is really hard to bypass Gump's control of the classpath using Ant. I'm not sure how mandatory/optional support for jar overrides in Maven is. >> Perhaps it could be tested by setting a security policy for Ant >> which disallowed network connections. > > Wouldn't that make projects fail? Certainly. We've seen this a couple of times. Many <javadoc> tasks will fail, for example, since the <link> to Sun javadoc sites and the javadoc executable tries to download the package index files from there. Cactus and Axis need network access (at least to localhost) for tests ... Stefan --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
