Kathey Marsden wrote:
>
> This might be reaching too far, but in playing with this this week,
> (before hitting the error log show stopper)
> I was thinking that rather than just providing samples on how to load
> Derby in a classloader, it would be really
> nice to offer some sort of API to make it really simple to load derby
> given an URL[] and an easy way to get
> DataSources from it. Might it be possible to integrate this into
> your an API somehow, perhaps with a set method
> that takes the URL string specification of the directory where Derby
> is installed and a DataSourceFactory that is
> initialized with the server instance?
>
I think this would be useful if we can find a way around the chicken/egg
problem: how does the class containing that API get loaded and from
which classloader?
Perhaps we could have a very small and stable bootstrap API jar that
contained this factory code and which could be used to bootstrap any
version of the engine. Something like:
public abstract class DataSourceFactory {
/** Locate the factory as a JAR service */
public static DataSourceFactory getFactory(String uri) {...}
/** return true if an impl can handle the requested url */
public abstract boolean canHandle(String uri) {...}
/** return a configured DataSource */
public abstract DataSource getDataSource() {...}
/** set a property for this source *//
public abstract void setProperty(String name, String value) {...}
/* other common initialization methods ... */
}
/**
* Implementation that returns a DataSource for an embedded engine
*/
public class EmbeddedDataSourceFactory extends DataSourceFactory {...}
The DataSourceFactory base class would use the JAR service location
mechanism to find an implementation that could handle the supplied URI -
note the URI would identify the implementation and not the actual
database (basically just the protocol bit from the Driver URL).
[[ Random thought, there's nothing here that's Derby specific, perhaps
we should suggest adding something like this to javax.sql? ]]
This still leaves it up to the host environment to add the
implementations to the classpath so that getFactory can locate them
using getResource(). I would argue that is the right separation of concerns.
--
Jeremy
