Your repository implementation should be distributed in a JAR file. The produced JAR should be self contained, meaning that all of the runtime classes should be bundled in it. It should not rely on manifest header like Class-Path and others. The build system will not take those into account when loading your classes.
Your package should contain a class file which is an implementation of
SakerRepositoryFactory. This class will be loaded first when the repository implementation is used.
Your archive may be multi-release, the build system will load the classes accordingly. However, module related information on JDK 9+ may not be taken into account.
Given a properly packaged archive, it can be used in the following way on the command line:
The above example will load the repository factory with the class name
example.ExampleRepositoryFactory from the archive located in the current process working directory with the name
myrepository.jar. The repository will have the identifier
myrepo during the build execution. See Task repositories for more information.
Note that the
-repository-class option was necessary for the build system to locate the entry point for the repository. It can be omitted if the archive declares the
saker.build.runtime.repository.SakerRepositoryFactory service using the
ServiceLoader facility of the Java runtime.
This way, the
-repository-class option may be omitted.
Although declaring the service can be convenient, users should be noted that the
ServiceLoader can find other classes too which are not present in the distributed JAR. This can happen in various cases such as when the user specifies a class path for the Java process that already has some repository service declaration. In order to promote determinism and reproducability, users are recommended to use the