Pub/Sub SPI: Adding IAM methods and auto-generated unit tests

[garrettjonesgoogle]

Note: grpc-google-pubsub-v1 depends on a new package grpc-google-iam-v1, which contains IAMPolicy and Policy.

[mziccard]

A couple of questions on dependencies:

• grpc-google-iam-v1 contains iam classes that will be used by all services, right?
• Why does grpc-google-common-protos depend on protobuf-java 3.0.0-beta-3 instead of the latest protobuf-java 3.0.0?

[garrettjonesgoogle]

• yes, grpc-google-iam-v1 will be shared by multiple services.
• I haven't updated the protobuf-java dependency in order to isolate changes. I will do a batch upgrade of grpc to 1.0.0 and protobuf to 3.0.0 in one go, without other changes at the same time.

[coveralls]

Coverage increased (+0.2%) to 85.434% when pulling 96f22d3 on garrettjonesgoogle:master into 7cd60df on GoogleCloudPlatform:master.

[mziccard]

@garrettjonesgoogle another general question: is there a real need for having a copy of setIamPolicy, setIamPolicy and testIamPermissions in both PublisherApi and SubscriberApi? It seems like a lot of duplicated code to me. Will other services use/require this approach as well? Did you consider putting the three methods in a dedicated PubSubIamApi class?

[garrettjonesgoogle]

It's the general pattern for IAM. In services that follow the suggested pattern (see https://github.com/googleapis/googleapis/blob/master/google/genomics/v1/datasets.proto ), the methods are redeclared in each service declaration. This allows their documentation and http paths to be customized. Pub/Sub doesn't follow the suggested pattern in the proto declarations, although it shows up the correct way in the http view (see https://developers.google.com/apis-explorer/#p/pubsub/v1/ ).

[mziccard]

Will setting the policy for a topic resource via the SubscriberApi (or for a subscription via the PublisherApi) fail? Or is this just for having specific docs and http endpoints?

[garrettjonesgoogle]

The way it's set up currently, it will fail validation when the flattened methods are used, but the request-based methods would not fail. The http endpoints actually have zero impact on the gRPC call; we just use the http endpoint information to inform our codegen (we derive the validation from it).

[mziccard]

I am closing this, rebased on top of master in #1229