[Bug]: Thrift SDK crashes selecting an array-of-embedding column (AttributeError: 'NoneType' has no attribute 'dimension')

Is there an existing issue for the same bug?

• I have checked the existing issues.

Version or Commit ID

main @ 4954148 (current main as of 2026-09-08)

Other environment information

Actual behavior and How to reproduce it

Selecting from a table with an array-of-embedding column (e.g. "array,vector,4,float") over the thrift SDK crashes the client result parser:

AttributeError: 'NoneType' object has no attribute 'dimension'

In parse_single_array_bytes (python/infinity_sdk/infinity/remote_thrift/types.py), the Embedding arm reads the embedding metadata from the wrong DataType:

embedding_dimension = column_data_type.physical_type.embedding_type.dimension
match column_data_type.physical_type.embedding_type.element_type:

column_data_type is the ARRAY type, so its physical_type union carries array_type and embedding_type is None. The embedding info lives on element_data_type (the array's element type), which the function already extracts two lines above.

The server side supports this fine: InfinityThriftService::HandleArrayTypeRecursively serializes an array-of-embedding as i32 element count followed by count * dimension * element-size raw bytes, and create-table accepts array,vector,N,float (the element DataType is converted recursively). Only the client parse path is broken.

Repro (serverless, feeds the exact wire bytes the server produces):

from infinity.remote_thrift.types import parse_single_array_bytes
# col = DataType(logic_type=Array, array_type.element_data_type =
#                DataType(logic_type=Embedding, embedding_type=(float32, dim 4)))
buf = struct.pack('<i', 2) + struct.pack('<8f', 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0)
parse_single_array_bytes(col, buf, 0)  # AttributeError

Expected behavior

The parser reads the embedding dimension/element type from element_data_type (where the union actually holds embedding_type) and returns [[1.0, 2.0, 3.0, 4.0], [5.0, 6.0, 7.0, 8.0]] for the bytes above.

Additional information

Fix: two-line change to read embedding info from element_data_type instead of column_data_type, plus a regression test round-tripping an array,vector,4,float column (insert + select). PR coming.

Start in python/infinity_sdk/infinity/remote_thrift/types.py at parse_single_array_bytes and inspect how element_data_type is extracted for array values. Reproduce the failure with the serverless byte buffer from the issue, then add a regression test round-tripping an array,vector,4,float column through insert and select. Done means the parser returns two four-element arrays without raising AttributeError.