Remove `localhost:9090` fallback for default gRPC channel

[therepanic]

#400 (comment)

So, I did a little research on this issue.

If we refer to the autoconfiguration logic we moved to Spring Boot, we'll see that with default we'll use the default "static://localhost:9090".

In general, I suggest we think about this. I can suggest the following:

We could force a custom target to be specified, whether as a property or an annotation parameter, and if it's not specified, we would return an error log. After that, we could remove the hardcoded code from Spring Boot's autoconfiguration.

This would be consistent with the logic behind, for example, @EnableJpaRepositories. After all, when we use this annotation, we don't connect to jdbc:h2:mem:default by default, and we see an error.

This would be the right thing to do, I think, and I can open an issue about it and we can figure out a solution.

If we're happy with this setup, I can submit a PR to the Spring Boot repository where I'll fix the autoconfiguration itself. After this, the user will be given a specific message that he should specify the address in the default channel.

[dsyer]

As things stand if you use @ImportGrpcClients and not Spring Boot you get a default target of "default", so the DefaultGrpcChannelFactory will try to connect to that and get a IllegalArgumentException ("Could not find a NameResolverProvider for static://default"). The "static://" prefix is added by the VirtualTargets.DEFAULT and it was copied from the Spring gRPC community project, but it's not a thing in the Google grpc-java library. Nothing in spring-grpc-core actually understands it. We should probably make the default "localhost:9090" (with no schema added by the VirtualTargets) in this project? WDYT?

Spring Boot uses its own VirtualTargets where the default target is "static://localhost:9090", and the "static:" prefix is stripped out by Boot so that grpc-java can just use the target as an authority string (i.e a hostname with optional port). As long as no-one is complaining specifically about this in Spring Boot, it's probably OK to leave it as is. But maybe they should consider dropping the "static:" prefix as well, unless it's there for a reason.

[therepanic]

I agree, the intended mechanism won't work without Spring Boot. I created a PR to change this behavior: #404.

I also assume that tcp: as well as static: are redundant?

[dsyer]

Yes, tcp: and static: are made up schemas (as is inprocess:). The google library has a NameResolverProvider for dns: and unix:. I think that might be it.