SourceConfiguration Union type deserializes to wrong type (e.g., SourceAirtable for S3 sources)

[devin-ai-integration]

Summary

When calling get_source() or similar methods that return a SourceResponse, the configuration field is incorrectly deserialized to SourceAirtable regardless of the actual source type. This prevents users from accessing source-specific configuration fields like bucket, streams, or globs for S3 sources.

Root Cause

The SourceConfiguration type is a Python Union of 500+ source configuration types. When deserializing JSON responses:

1. dataclasses_json tries each type in the Union in order until one succeeds
2. SourceAirtable appears early in the Union (position 8) and has all optional fields
3. The @dataclass_json(undefined=Undefined.EXCLUDE) decorator ignores unknown fields
4. Since SourceAirtable has no required fields, it successfully deserializes ANY JSON payload

The OpenAPI spec uses oneOf without a discriminator field, so there is no way for the SDK to determine the correct type to deserialize to.

Reproduction

from airbyte_api import AirbyteAPI
from airbyte_api.models import Security

client = AirbyteAPI(security=Security(bearer_auth="..."))

# Get an S3 source
response = client.sources.get_source(source_id="1f5ca207-5c40-48d6-b9d1-6667de9fe427")
source = response.source_response

print(f"Source type: {source.source_type}")  # "s3"
print(f"Config type: {type(source.configuration).__name__}")  # "SourceAirtable" (WRONG!)
print(f"Has bucket? {hasattr(source.configuration, 'bucket')}")  # False (WRONG!)

Expected Behavior

The configuration field should be deserialized to SourceS3 when source_type is "s3".

Workaround

Users can access the raw configuration dict or manually deserialize to the correct type:

import json
from airbyte_api import utils
from airbyte_api.models import SourceS3

# Option 1: Access raw config dict
raw_config = response.raw_response.json()["configuration"]
print(raw_config["bucket"])  # Works
print(raw_config["streams"][0]["globs"])  # Works

# Option 2: Manually deserialize to correct type
config_json = json.dumps(response.raw_response.json()["configuration"])
s3_config = utils.unmarshal_json(config_json, SourceS3)
print(s3_config.bucket)  # Works
print(s3_config.streams[0].globs)  # Works

Potential Fix

A proper fix would require either:

1. Adding a discriminator field to the OpenAPI spec for SourceConfiguration using the sourceType property
2. Modifying Speakeasy's generation to handle discriminated unions based on a sibling field

Since this is a generated SDK, any direct code changes would be overwritten on regeneration.

Context

This issue was reported by a customer trying to modify streams.globs at runtime from Airflow. Investigation requested by @iherdt-airbyte.

Related: Commit 87f7e7ba removed some discriminators from the OpenAPI spec in September 2024.

[iherdt-airbyte]

One more detail related to this: the start_date parameter (and potentially other date fields) returned by get_source in the SourceS3 object (and others) is formatted as a Python-style datetime string (e.g., 2026-01-04 00:00:00+00:00). However, the API's update endpoint requires strict ISO 8601 formatting (e.g., 2026-01-04T00:00:00.000000Z). Users currently have to manually reformat these date strings before they can successfully use the object returned by the SDK in an update_source call.