[哈基米 (BlueX888)]

Fixes #14713

Problem

SQL Server vector searches are ordered using the shared DISTANCE_FUNCTION_DIRECTION_HELPER, which models dot product as a similarity. VECTOR_DISTANCE is a distance for every SQL Server metric — the dot metric is documented as the negative dot product, with smaller values more similar — so this has two consequences:

• A vector field left at the default distance function (VectorStoreField("vector", dimensions=...) → DistanceFunction.DEFAULT) is accepted by DISTANCE_FUNCTION_MAP (mapped to cosine) but has no entry in the direction helper. Every search fails at runtime:

  VectorSearchExecutionException: An error occurred during the search:
  <DistanceFunction.DEFAULT: 'DEFAULT'> -> KeyError: <DistanceFunction.DEFAULT: 'DEFAULT'>
  

• A field using DistanceFunction.DOT_PROD builds ORDER BY ... DESC, returning the least similar rows first.

InMemoryCollection already resolves DEFAULT to cosine; SQL Server's supported metrics are all distances, so the direction is always ascending.

Change

Order SQL Server vector results ascending for every supported metric and drop the now-unused direction-helper import. The metric name passed to VECTOR_DISTANCE is unchanged.

Verification

• test_build_search_query now runs with COSINE_DISTANCE, EUCLIDEAN_DISTANCE, DOT_PROD, and DEFAULT, asserting the matching VECTOR_DISTANCE metric and ORDER BY ... ASC.
• Added a public-path SqlServerCollection.search test whose model leaves the vector field at DistanceFunction.DEFAULT.
• On the base commit the DOT_PROD and DEFAULT cases fail (DESC and KeyError respectively); with the change, all 82 tests in tests/unit/connectors/memory/test_sql_server.py pass.
• ruff check and package mypy -p semantic_kernel are clean.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.