The org.apache.velocity.tools.view.servlet.WebappLoader class ensures that all the paths it can load from end with the '/' character. When loading a template from disk in getResourceStream(), WebappLoader ensures that the file to be loaded does not start with a '/' character, so the file always loads properly.
However, in isSourceModified(Resource resource) and getLastModified(Resource resource), the same sanity check is not made on the path of the resource, so you can have a path like /myApp//some/file.vm. The result of this, in my environment(s) at least, is that the isSourceModified() check always returns true because the file cannot be found. With caching turned on, this leaves you with with the Least Efficient Cache Ever, since it always re-reads any template requested after the modificationCheckInterval.
I haven't looked at it carefully, but I think FileResourceLoader may have the same behavior.
An interesting side effect of this unruly behavior is, I think, a race-condition bug in the Template class. In the process() method, the first thing it does is set its "data" member, the AST of the template, to null. However, another thread can get that same data member while it's uninitialized. This leads to the "java.lang.Exception: Template.merge() failure. The document is null, most likely due to parsing error." error. A google search for "Template.merge() failure" turns up a couple web applications that, I think, are falling prey to this problem as well. This problem is easy to duplicate:
1. Enable resource loader caching
2. Set the modificationCheckInterval to something small, like 10 seconds.
3. Hammer your application with requests.
I'm not intimately familiar with the internals of velocity, so I'm not sure what the fix should be, or even where it should go. FWIW, I ended up writing my own ResourceLoader that expects a leading '/' on all template names. Thoughts?