bug: EmbeddingList search doesn't support non-float vectors in struct array

[zhuwenxing]

Description

PR #3268 fixed the insert path for non-float vectors (Float16, BFloat16, Int8, Binary) in struct arrays. However, the search path still has issues — EmbeddingList cannot be constructed with non-float vector data, and the binary EmbeddingList placeholder type is never used.

How to reproduce

from pymilvus import EmbeddingList
import numpy as np

# Float16 vector stored as bytes (as returned by entity helper)
vec = np.random.rand(128).astype(np.float16)
vec_bytes = vec.tobytes()

emb_list = EmbeddingList(dtype=np.float16)
emb_list.add(vec_bytes)  # ValueError: Embedding must be 1D, got shape ()

Root Cause

Issue 1: EmbeddingList.add() doesn't handle bytes input

In embedding_list.py:

def add(self, embedding):
    embedding = np.asarray(embedding)  # bytes → 0-D ndarray (shape=())
    if embedding.ndim != 1:            # 0 != 1 → ValueError
        raise ValueError(f"Embedding must be 1D, got shape {embedding.shape}")

For non-float vectors, the entity helper stores vectors as bytes. np.asarray(bytes) produces a 0-D ndarray, causing the dimension check to fail.

Suggested fix: Handle bytes input by converting to numpy array with the appropriate dtype:

if isinstance(embedding, bytes):
    if self._dtype is not None:
        embedding = np.frombuffer(embedding, dtype=self._dtype)
    else:
        raise ValueError("Cannot add bytes embedding without dtype specified")

Issue 2: _prepare_placeholder_str missing binary EmbeddingList branch

In prepare.py:

elif dtype == "byte":
    pl_type = PlaceholderType.BinaryVector  # Missing is_embedding_list check!
    pl_values = data

And:

elif isinstance(data[0], bytes):
    pl_type = PlaceholderType.BinaryVector  # Missing is_embedding_list check!
    pl_values = data

Both branches don't check is_embedding_list, so PlaceholderType.EmbListBinaryVector (already defined as 300 in types.py) is never used.

Server-side verification

Milvus server correctly handles non-float vector search when data is properly serialized:

Scenario	Result
element_filter + float16 ndarray + L2 metric	✅
MAX_SIM + float16 EmbeddingList (manually constructed) + MAX_SIM_L2	✅

No server-side changes needed — this is a pymilvus-only fix.

Related

• PR fix: support non-float vectors in struct array #3268 (insert path fix, merged)
• 8 struct array element search test cases are xfail due to this issue

[zhuwenxing]

/assign @SpadeA-Tang