RMIClassLoaderSPI, was: Re: getting a basic class loader

Alan Bateman Alan.Bateman at oracle.com
Mon Dec 7 14:01:27 UTC 2015

On 07/12/2015 13:13, Peter Firmstone wrote:
> Just a quick one, now that all modules have URI annotations, is there a tool or utility to find the correct ClassLoader during deserialisation and unmarshalling with RMI?
> We've got our own RFC3986 compliant URLClassLoader, that overrides CodeSource to ensure that URL isn't used as a key in SecureClassLoader's cache.
> The benefit is CodeSource identity is based on a normalized RFC3986 compliant address and Certificates, not on a DNS resolved IP address.
> Will there be new ClassLoader api?
> We also mostly thread confine class loading (1 thread per ClassLoader) to avoid ClassLoader monitor contention, which reduces  class loading overhead to less than 1% cpu.  Previously class loading was hierarchical, now I'm guessing that ClassLoading isn't hierarchical with Jigsaw?
> Will there be an api method to find ClassLoaders using their URI annotation in java 9?
In the current proposal then modules are not required to have a 
"location" or be located by a URI. There are modules generated 
dynamically for things like proxies and dynamic languages that don't 
have locations for example. There isn't anything in the API to locate 
class loader objects by URI.

Alex's "Under the Hood" talk is a good talk to go through and understand 
the relationship between class loaders and modules (or module layers) in 
the current proposal. In general, Jigsaw places very few restrictions on 
how modules are mapped to class loaders and those restrictions are 
really just stem from fundamental constraints on how class loading works.

As regards the built-in class loaders (boot, extension, application) 
then they continue to do hierarchical/parent-delegation when loading 
types in unnamed modules. They use direct delegation when loading types 
in named modules. This may or may not be interesting to what you are 
doing as the built-in class loaders are not URLClassLoader objects. Also 
the CodeSources are unlikely to have a host component in the URI.


More information about the jigsaw-dev mailing list