How do I relocate / rename / replace a registered base path on an existing dataset?

Summary

When a dataset is created with initial_bases=[DatasetBasePath(...)], there doesn't appear to be a supported way to later change that base path — for example, to point the same logical base at a new S3 location after data has been moved. LanceDataset.add_bases() only adds new bases (and rejects duplicate names or paths), and initial_bases is documented as CREATE-only. I'd like to know what the intended workflow is.

Environment

• pylance 4.0.1
• Python 3.12
• Linux

Reproducer

# /// script
# requires-python = ">=3.10"
# dependencies = [
#   "pylance>=0.38",
#   "pyarrow",
# ]
# ///
import lance
import pyarrow as pa
from lance import DatasetBasePath

S3_BASE = "s3://my-bucket/my-prefix"
VIDEO_URI = f"{S3_BASE}/abc/video.mp4"

table = pa.table({
    "id": pa.array([1], type=pa.int64()),
    "video": lance.blob_array([VIDEO_URI]),
})

lance.write_dataset(
    table,
    "./videos.lance",
    data_storage_version="2.2",
    initial_bases=[DatasetBasePath(path=S3_BASE, name="my_base")],
    allow_external_blob_outside_bases=False,
)

After the dataset exists, the registered bases look like:

{1: DatasetBasePath(id=1, name=Some("my_base"), path=s3://my-bucket/my-prefix, is_dataset_root=false)}

What I want to do

Relocate the external blobs to a new bucket/prefix (e.g., s3://my-new-bucket/my-new-prefix) and update the dataset so that the base previously named my_base now resolves to the new path —

1. Re-add with the same name or same path

ds._ds.add_bases(
    [DatasetBasePath(path="s3://my-new-bucket/my-new-prefix", name="my_base")],
    None,
)

Fails with:

ValueError: Invalid user input: Conflict detected: Base path with name 'Some("my_base")'
or path 's3://my-new-bucket/my-new-prefix' already exists.

Same error when passing id=1 explicitly to match the existing base.

2. Add a new base under a different name

ds._ds.add_bases(
    [DatasetBasePath(path="s3://my-new-bucket/my-new-prefix", name="my_base_v2")],
    None,
)

This succeeds, but now the manifest carries both bases, and the original one is still present with no way to remove it. I couldn't find a remove_bases / update_bases / rename_base API on LanceDataset or on the inner _Dataset.

3. Re-run write_dataset with new initial_bases

Per the docstring:

initial_bases: Only used in CREATE mode. Cannot be specified in APPEND or OVERWRITE modes.

So this only works after deleting the dataset directory and recreating from scratch.

Questions

1. Is there a supported way to rename, replace, or remove an existing DatasetBasePath on a dataset, or to repoint one at a new path?
2. If not, is the expected workflow to add_bases(...) under a new name and then rewrite the blob columns so their URIs fall under the new base, leaving the old base as a dangling manifest entry forever?
3. Is there a recommended pattern for the "I physically moved the files in S3, now update the dataset" scenario?

Thanks!

Start by tracing the DatasetBasePath and add_bases entry points, then compare them with write_dataset's CREATE-only initial_bases behavior. Determine the supported workflow for renaming, replacing, or removing a registered base after external blobs move; done should be a documented, tested answer or an agreed API design.