Although it doesn't make much difference in my subjective
testing < 2 seconds (I'm not even sure that number is real
since I only ran it once), I committed it anyway.
Maybe somebody else will see some benefit?
See
http://jira.jboss.com/jira/browse/JBCL-12
which will be in the next classloader release.
Why don't we rather do *real* VFS cache, instead of each problem /
project dealing with this on its own?
Cache seems a natural fit for AOP.
Perhaps a new module in VFS project with caching aspects - e.g. aspects
per VirtualFile instance (on getChild, getChildren, ...), global
VFS-->URL-->VirtualFile, ...
And then user could choose how he wants his VFS cache to function,
selecting his set of aspects.
Since classloading now depends on VFS, and AOP depends on classloading,
we should probably ship this pre-prepared, having a couple of different
'flavors'.