Skip to content

Bundlebase 0.6.0

Two big additions: text search with the new search() table function, and the ability to attach and join other bundles.

New Features

Text search

The way text indexes work was completely revamped. You now create text indexes and query them with the search() table function:

bundle.create_index(["Company"], "text", name="company_search")
bundle.commit("Created text index")

result = bundle.query(
    "SELECT Company, _score FROM search('company_search', 'Group') ORDER BY _score DESC"
)

Multi-column indexes and field-specific queries work too:

bundle.create_index(["Company", "City"], "text", name="company_city")
bundle.commit("Created multi-column index")

result = bundle.query(
    "SELECT * FROM search('company_city', 'Company:group') ORDER BY _score DESC LIMIT 10"
)

If there's only one text index on the bundle, you can skip the name:

SELECT * FROM search('machine learning')

Results include a _score column with BM25 relevance scores. Multiple tokenizers are available including stemming for English, German, French, Spanish, and others — pass args={"tokenizer": "en_stem"} when creating the index.

See the text search guide for the full details.

Attaching and joining bundles

Bundles can now reference other bundles. Attach pulls in another bundle's data, and join links them on a key:

# Attach another bundle's data
bundle.attach("bundle:///path/to/other/bundle")

# Join bundles on a key
bundle.join("regions", 'base."Country" = regions."Country"',
       "bundle:///path/to/regions/bundle")

Remote bundles work too — use bundle+s3://bucket/path for S3-backed bundles.

The attached/joined data reflects the target bundle's current state, including any filters or transformations applied to it.

See the attaching guide and joins guide.

CLI renamed to bundlebase

The CLI binary is now just bundlebase instead of bundlebase-cli.

Other changes

  • Index caching — deserialized indexes are now cached in an LRU cache, avoiding repeated disk reads for hot queries
  • Better explain() and query() output — improved formatting in the Python API

Breaking Changes

text_search() replaced by search()

The old text_search() UDF is gone. Use the search() table function instead, which supports named indexes, multi-column queries, and relevance scoring.

CLI binary name

If you have scripts referencing bundlebase-cli, update them to bundlebase.

pip install bundlebase==0.6.0

Let me know if you run into issues, especially with the new search function or bundle-to-bundle joins.