Thegoogle.cloud.bigquery_v2 modules are frequently out-of-date. This primarily affect the usability of BigQuery ML.

Since BigQuery is a REST API, protobuf changes must be synced manually in internal code repo before they can be published tohttps://github.com/googleapis/googleapis

In the years since we first introducedgoogle.cloud.bigquery_v2, we've seen that BigQuery ML adds model stats and model types much more frequently than the protobuf changes are actually published.

We've worked around this by:

• I and@shollyman (mostly@shollyman lately) republish the protos when a customer has a problem, such as inlist_models() of dataset containing a BigQuery AutoML models errors with Invalid enum value AUTOML_REGRESSOR for enum type #293
• Avoiding the worst exceptions when an enum or property isn't present, such as inMake Model constructor robust against new model types #334

Solution

• Removegoogle.cloud.bigquery_v2 modules
• Update owlbot config to stop generating "client"
• In the BigQuery ML classes, whereever we were returning protobuf object, return the JSON-parsed API response, instead.
  • Complex types become dictionaries.
  • Enums become strings.
• In the routines logic, create manual wrappers and enums where needed.

Supersedes#319