fix: add missing handler for deserializing json value

[tomwojcik]

Fixes#1500

[google-cla]

Thanks for your pull request! It looks like this may be your first contribution to a Google open source project. Before we can look at your pull request, you'll need to sign a Contributor License Agreement (CLA).

View this failed invocation of the CLA check for more information.

For the most up to date status, view the checks section at the bottom of the pull request.

[tomwojcik]

@tswast bump, please review

[tswast]

Thanks!

One concern I have is the other direction. If someone supplies a parsed JSON object to an API like insert_rows, do we correctly serialize that?

[tswast]

Looks like everything is working well except for the type annotations:

nox > Running session mypy
nox > Creating virtual environment (virtualenv) using python3.8 in .nox/mypy
nox > python -m pip install -e '.[all]'
nox > python -m pip install mypy==1.6.1
nox > python -m pip install types-protobuf types-python-dateutil types-requests types-setuptools
nox > mypy -p google --show-traceback
google/cloud/bigquery/query.py:472: error: Cannot call function of unknown type [operator]
google/cloud/bigquery/query.py:629: error: Cannot call function of unknown type [operator]
google/cloud/bigquery/query.py:778: error: Cannot call function of unknown type [operator]
Found 3 errors in 1 file (checked 54 source files)
nox > Command mypy -p google --show-traceback failed with exit code 1
nox > Session mypy failed.

[MartijnvanElferen]

This change is causing the following error for insert_rows:

TypeError: _json_from_json() missing 1 required positional argument: 'field'

Is there a recommended fix for this?

[tswast]

Thanks @MartijnvanElferen for the report. I've filed #1756 to investigate further.

[tswast]

As a workaround insert_rows_json should avoid this type conversion code, but you're responsible for putting things in a format the BigQuery REST API understands.

[tswast]

@MartijnvanElferen Fix pending: #1757

Note: this PR will change the behavior of JSON data in insert_rows to call json.dumps, which is consistent with the new behavior of calling json.loads when reading row values. Previously, JSON support was happening via a generic fallback that didn't do any conversion of values passed in.

[MartijnvanElferen]

Thanks, @tswast. I'll refactor a couple of things to avoid the type conversion. Looking forward to the fix at the same time!

[kcooney]

@tswast This change should probably be listed as a potentially braking change in the release notes, since before this change reading the entire column would return a value of type str:

CREATETABLEmydataset.table(
id INT64,
proto_as_json JSON
);

fromgoogle.cloudimportbigqueryfromgoogle.protobufimportjson_formatfromcom.example.my_proto_pb2importMyProtoclient=bigquery.Client()
proto=MyProto(first_name="Jane", last_name="Doe") proto_as_json=json_format.MessageToJson(
proto, use_integers_for_enums=True, indent=None)
rows_to_insert= [
{"id": 1, "proto_as_json": f'JSON """{proto_as_json}"""'},
]
errors=client.insert_rows("mydataset.table", rows_to_insert)
assertnoterrorsrow=client.query("SELECT proto_as_json FROM mydataset.table").result().__next__()
json_str=row["proto_as_json"]
# Prior to this change, json_str was of type str, so we could do this:read_proto=MyProto()
json_format.Parse(json_str, read_proto)
assertproto==read_proto

Luckily, in our project we created a common helper method for the json_format.Parse() call, so we can update it to use json_format.ParseDict()

[tswast]

Yeah, it's a gray area with this since we never said we support the JSON type until this change.

We still don't support JSON in some code paths like to_dataframe()

[kcooney]

@tswast We saw the breakage when we tried to update a number of python libraries to newer versions. The release notes were very unhelpful in determining which library and version could have caused the problem (because the PR and bug listed in the release notes suggested that it only affected selecting subfields). It wasn't until we found a work around that we were able to track the issue to v3.15.0 and this change. Even if there were no prior promises of support of the JSON type, a short blurb in the release notes would be very much appreciated.