Problem

In pymongo_search_utils/index.py, the docstrings for the dimensions argument say to pass None when using auto-embeddings:

dimensions (int): Number of dimensions in embedding,
    {{None}} if using auto-embeddings

But dimensions is a required positional argument typed int, and the actual sentinel checked by _check_param_config is -1:

if auto_embedding_model is not None and (dimensions != -1 or similarity is not None):
    raise ValueError(
        "if auto_embedding_model is set, then neither dimensions nor similarity may be set."
    )

So a user following the docstring and passing dimensions=None with auto_embedding_model set gets a ValueError telling them not to set dimensions at all — which is also misleading, since it must be explicitly set to -1.

Note that similarity is genuinely Optional[str] and None is correct there; only dimensions is wrong.

Affected functions

All three take the same argument and repeat the same wrong text:

• vector_search_index_definition
• create_vector_search_index
• update_vector_search_index

Suggested fix

Correct the docstrings to say -1 when using auto-embeddings, e.g.:

dimensions (int): Number of dimensions in the embedding.
    Pass -1 when using auto-embeddings.

Optionally also reword the ValueError message so it states the required value rather than implying the argument should be omitted.