Namespace revisit

[tadelesh]

When Go and Rust deal with the namespace in the emitter, they are asking how to determine a root namespace that could let language know where a package should start from. Current TCGC's namespace concept is just to represent TypeSpec namespace or user specific namespace. We do not have root namespace concept. So, we want to know if all languages have such requirement and does TCGC need a rootNamespace config for emitter, and provide getRootNamespaces with calculating longest common prefix for all root clients' namespaces?

[azure-sdk]

No changes needing a change description found.

[pkg-pr-new]

Open in StackBlitz

npm i https://pkg.pr.new/@azure-tools/typespec-client-generator-core@4033

commit: 831c44a

[azure-sdk]

You can try these changes here

🛝 Playground	🌐 Website

[iscai-msft]

love the doc

[weidongxu-microsoft]

When no namespace in tspconfig, Java would generate client and model to their namespace specified by TCGC (we may do some hack for models from TypeSpec or Azure namespace -- here is the only place a rootNamespace may matter -- we need to put them somewhere, if not to typespec.rest package). This part would be fine without a rootNamespace.

The remaining problem I see, is the artifactId in the generated pom.xml.

---

E.g. for source

@service(#{ title: "Pet Store" })
namespace PetStore {
  model Dog { ... }
}

@service(#{ title: "Toy Store" })
namespace ToyStore {
  model Car { ... }
}

The 2 client would be generated under petstore and toystore package. No surprise.
But we are not able to decide whether the artifactId (you can think it as package name) should be petstore or toystore.

In addition, if the project uses TypeSpec.Rest.ResourceCreateModel<>, we either generate it under typespec.rest package, or use the guessed root namespace.

---

Hence for Java, there is not as much a problem of rootNamespace, but packageName. E.g. if we had a package-name tspconfig, the problem thus be solved without knowing a rootNamespace.

CC @srnagar