Python: Bug: search ignores the advertised vector_field_name and searches another vector field #14740

Describe the bug

The Python VectorSearch.search() overloads advertise vector_field_name, but the implementation accepts vector_property_name and leaves the advertised keyword in **kwargs. With two vector fields, the public in-memory search silently selects the default field instead of the requested field. The same query therefore returns a different record.

I found this through source inspection and reproduced it locally through real InMemoryCollection.upsert() and search() calls. This is not a production-incident report.

To Reproduce

import asyncio

from semantic_kernel.connectors.in_memory import InMemoryCollection
from semantic_kernel.data.vector import VectorStoreCollectionDefinition, VectorStoreField


async def main():
    definition = VectorStoreCollectionDefinition(fields=[
        VectorStoreField('key', name='id', type='str'),
        VectorStoreField('vector', name='title', type='float', dimensions=2),
        VectorStoreField('vector', name='body', type='float', dimensions=2),
    ])
    collection = InMemoryCollection(record_type=dict, definition=definition, collection_name='field-selection-repro')
    await collection.upsert([
        {'id': 'a', 'title': [1.0, 0.0], 'body': [0.0, 1.0]},
        {'id': 'b', 'title': [0.0, 1.0], 'body': [1.0, 0.0]},
    ])
    observations = {}
    for key, kwargs in {'overload_parameter': {'vector_field_name': 'body'},
                        'runtime_parameter': {'vector_property_name': 'body'},
                        'default': {}}.items():
        found = await collection.search(vector=[1.0, 0.0], top=1, **kwargs)
        observations[key] = [hit.record['id'] async for hit in found.results]
    print(observations)

asyncio.run(main())

Observed output:

{'overload_parameter': ['a'], 'runtime_parameter': ['b'], 'default': ['a']}

Expected behavior

The advertised vector_field_name="body" selects the body vector, returning b, just as vector_property_name="body" does. Callers following the overload should not silently search title instead.

Platform

• Language: Python 3.12.13, macOS ARM64.
• Source: current main, df1cb33d0d32699c8d6a2e5b3ea7e5f8990d86d5 (full Python source snapshot).
• AI model: none; vectors are supplied directly. No cloud, database, model call, or external-service substitute is used.

Additional context / proposed contribution

The mismatch is in python/semantic_kernel/data/vector.py. The overloads use vector_field_name, while the implementation resolves vector_property_name.

I would like to take this fix. My proposed compatibility approach is to accept the advertised keyword as an alias, retain vector_property_name for existing callers, and reject conflicting values rather than silently choosing one. I would add public in-memory search regressions for default selection, both spellings, and the conflict case.

Could a maintainer confirm the intended direction and assign this to Elon-Jiang if contributions are welcome? If the intended fix is to correct the overloads instead, please let me know. I am waiting for agreement before preparing a product patch, per CONTRIBUTING.

AI assistance was used for investigation, reproduction, and this report. Current all-state keyword searches and the 24 open vector-related PRs did not reveal an overlapping fix; those searches are scoped rather than an assertion of exhaustive absence.