Did you actually look at the njar code? It is extremely short and simple and I think would answer most of your questions. Essentially it hides and caches unpacking a nested jar.
Aside from registering the njar handler, I don't think we are doing anything special to make multiply nested njar urls work. I believe more visible unpacking for jsps is needed only because the compiler needs files rather than urls. david jencks On 2002.05.20 10:39:58 -0400 Frederick N. Brier wrote: > I was trying to read through the postings regarding njar. It sounded > like > a lot of work and is an impressive technical achievement, but like Mark, > I > believe KISS is good. Further, it goes against speed and efficiency. > Disk > space is cheap. Memory is faster than disk is faster than compressed > disk > (or unzipping archive files on demand). All the archives (in lib and > deploy) on a server are less than 20MB and most are .jar(s). Even with > more archives, fully expanded, it is not going to be more than 50MB on > machines that have 80+GB drives. Is there some sort of cache mechanism > that keeps the njar protocol from repeatedly unzipping the archive to a > temporary location each time a InputStream is opened to it? Granted that > > these things usually only happen at startup, but is it going to happen > each > time a file is read such as Castor's database.xml and mapping.xml. And > there could be multiple MBean(s) in a .sar. Wouldn't that slowdown > startup? > > I am currently concerned about whether a relative file resource can be > resolved to a fully expanded URL with the njar: protocol > specified. However, even if it can, a second point is many libraries > check > the protocol of a URL or URI. For example, I am not sure how Xerces > would > react to nested protocols as required by "njar". The specific classes in > > question are org.apache.xerces.util.URI and > org.apache.xerces.impl.XMLEntityManager's method expandSystemId() [line > 1071]. Since Castor is built on top of Xerces, this would cause problems > > for Castor. I don't know how many other libraries attempt to draw > information from a URL, except that the URL exposes methods to decompose > it, so that it is a legitimate approach. Further, the java.net.URL class > > does not seem to provide any methods for understanding nested protocols. > > Again, if I am missing the boat here, please free to let me have both > barrels between the eyes. > > Fred. > > > _______________________________________________________________ > Hundreds of nodes, one monster rendering program. > Now that's a super model! Visit http://clustering.foundries.sf.net/ > > _______________________________________________ > Jboss-development mailing list > [EMAIL PROTECTED] > https://lists.sourceforge.net/lists/listinfo/jboss-development > > _______________________________________________________________ Hundreds of nodes, one monster rendering program. Now that's a super model! Visit http://clustering.foundries.sf.net/ _______________________________________________ Jboss-development mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/jboss-development
