Discussion: How to Structure Integrations

We have several integrations of Helidon with other technologies, for both Helidon SE and Helidon MP. I have built another one—​for Micrometer support—​and before I recreate the PR for it Tomas suggested in a Slack DM that we have a broader discussion for how to handle these integrations.

So far, we have used various approaches for how to structure the Helidon SE and Helidon MP modules for these integrations, as Tomas summarized:

Summary	Location for SE module	Location for MP module	Adv	Disadv
First-class citizen	helidon/xxx	helidon/microprofile/xxx	Follows our familiar approach of putting SE modules as children of the top-level and modules for Helidon MP under microprofile.	Not really appropriate if xxx is not truly a MicroProfile technology. e.g., Micrometer
Integration-level citizen, two modules	helidon/integrations/xxx/xxx	helidon/integrations/xxx/cdi	Follows our familiar approach which separates the Helidon SE and Helidon MP code.	Seems slightly odd to repeat the technology name in the SE module location.
	helidon/integrations/xxx/base	helidon/integrations/xxx/cdi
Integration-level citizen, one module	helidon/integrations/xxx		Reduces the number of modules in the system.	Looks slightly odd to have annotations in a module that is usable from Helidon SE. CDI behavior of xxx is triggered somewhat indirectly by the app developer declaring a dependency on some other module that has a harder dependency on CDI.

The "Integration-level citizen, one module" approach (e.g, Neo4J) declares a dependency on CDI using

<scope>provided</scope>
<optional>true</optional>

To follow the principle that no modules should be on the runtime classpath unless they are used, other dependencies—​such as for a module that defines annotations—​would be declared the same way in our integration module. Further, the developer would need to declare a dependency to make sure that such modules will be present at runtime.