BigQuery: [Primitive-ize API] SchemaField

[max-sixty]

Consistent with the recent work to allow the API to accept some primitives as well as specific objects, the schema argument to LoadJobConfig* could accept a list of dicts. Currently it requires constructing a bunch of SchemaField objects:

    schema = [
        # could be just column_dicts
        bigquery.schema.SchemaField.from_api_repr(column_dict) for column_dict in column_dicts
    ]

     with open(empty_path, "rb") as source_file: 
        job_config = bigquery.job.LoadJobConfig(
            schema=schema, write_disposition="WRITE_TRUNCATE"
        )

        job = gbq_client.load_table_from_file(
            source_file,
            table_ref,
            location="US",
            job_config=job_config,
        )

We're trying to move to using the BQ Python API rather than subprocessing out to a shell with bq; it's got much better but it's still a bit Java-esque, and these are examples of times it's a tougher sell than constructing a string a sending it to bash. As ever, lmk if the API isn't designed for these cases and you'd encourage users to use bash.

*this could also be a dict, though less of an imperative.

[max-sixty]

A good example of an API I think is nice and pythonic is the Kubernetes python API - for each object you can either supply:

• An object (of which there are literally hundreds...)
• A primitive, such as a dict / string
  ...and the library will do the appropriate coercions

[tswast]

We did recently add a slightly higher-level method for converting from JSON schema to list of SchemaField. I think it's reasonable to automatically convert in some of the classes/methods that currently accept a list of SchemaField.

Might be a little bit tricky, though, since we can't do type inference since they'd both be lists.

[max-sixty]

We did recently add a slightly higher-level method for converting from JSON schema to list of SchemaField

Nice, thanks. What's the method?

we can't do type inference since they'd both be lists.

Yes, good point. I think the K8s API rebuilds the whole graph of objects, serializing / de-serializing where needed. That's not realistic to start, though. A hand-written imperative check doesn't seem satisfying either...

[tswast]

Client.schema_from_json(file_or_path) is the related method.

Not quite what you're asking for though, since it's for files.

Edit: updated link to googleapis.dev

[max-sixty]

Perfect, thanks!

[tswast]

There's even precedent for accepting a JSON input in even the gRPC-based google-cloud-python libraries, so I'm hopeful there's a way we can do this in general. Maybe wherever we use to-api-repr in the client, try to do that and catch attribute error. If attribute error, treat it as the resource itself. Could make a good function in _helpers.py.