Fix "AttributeError: 'str' object has no attribute 'get'"

[weineran]

• This is a bug fix.
• The current code assumes that response_data will always be a dict or a mapping of some kind, but it is possible for response_data to be a str.
• When response_data is a str, then the function fails on response_data.get("error") but the error response is swallowed so the user doesn't know why it failed. The user just sees: AttributeError: 'str' object has no attribute 'get' and a stacktrace
• This is a quick and dirty fix that guarantees the .get call will succeed, and the result is that the error response bubbles up to the user.
• Feel free to make this fix more robust/clean.

Steps to reproduce the bug

• Set credentials like this:

  credentials = google.oauth2.credentials.Credentials(
      token=None,
      refresh_token=os.getenv('GCP_REFRESH_TOKEN'),
      token_uri=os.getenv('GCP_TOKEN_URI'),
      client_id=os.getenv('GCP_CLIENT_ID'),
      client_secret=os.getenv('GCP_CLIENT_SECRET'))

  But ensure that there is a typo in GCP_TOKEN_URI. Example: "https://oauth2.googleapis.com/token1"

• Use the credentials to authenticate. Example:

  df.to_gbq(destination_table=tablename, project_id=project_id, if_exists=if_exists, table_schema=schema, credentials=credentials)

  Expected: Authentication should fail with a useful error response Actual: Authentication fails with an AttributeError: "AttributeError: 'str' object has no attribute 'get'"

[arithmetic1728]

fixed in #1316 @weineran thank you for pointing out the issue and for your fix!

[weineran]

Awesome!