[prrao87]

Why

Release 0.9.0, which ships the 11 changes merged since v0.8.1: five planner performance improvements, Linux aarch64 wheels, and the removal of the deprecated refresh(). That removal breaks the Python and Rust APIs, so this is a minor bump rather than a patch.

What changed

This PR only moves the version. What the release carries:

Breaking

• CypherEngine.refresh() is removed from Python and Rust after six releases as a deprecated alias. checkout_latest() behaves identically. (refactor!: remove the deprecated refresh() #137)

Performance

• Grouped node properties from Lance tables. In lazy mode, grouping by a node after a hop that fans out groups by its id and reads the other properties once per node, where in-memory tables already did. Discarded lookups are also dropped from plans in both modes. Lazy IC5 is 8.8% faster and IC7 6.3% faster; the complex suite's lazy geomean is 0.988, and memory mode is unchanged. (perf(planner): look up grouped node columns from lance tables #140)
• Two-hop counts from node degrees in three more shapes: same-type hops that share an end (perf(planner): count same-type two-hop paths that share an end #131), undirected hops (perf(planner): count two-hop paths over undirected hops #132), and a second hop in an OPTIONAL MATCH (perf(planner): count two-hop paths whose second hop is optional #133). On graph-benchmark (2.4M FOLLOWS), (a)-[:FOLLOWS]->(b)<-[:FOLLOWS]-(c) RETURN count(*) drops from 7,958 ms to 43 ms, and the undirected chain from 8,512 ms to 93 ms.
• WHERE conditions that accept either direction of an edge now imply a condition per variable that can run below the expansion. The official LDBC IC14 text drops from about 2 minutes to 0.31 s on SF1. (perf(query): narrow each variable through an or of ands in where #130)

None of these change query results.

Platforms and CI

• Linux aarch64 wheels (manylinux_2_28), built natively and tested on Python 3.12 to 3.14. The Windows build fetches protoc from the protobuf release instead of Chocolatey, whose outage failed the first v0.8.1 run. (ci: aarch64 Linux wheels, protoc without Chocolatey, pinned actions and zizmor #139)
• Every GitHub Action is pinned to a commit SHA, kept current by Dependabot and audited by zizmor. (ci: aarch64 Linux wheels, protoc without Chocolatey, pinned actions and zizmor #139, build(deps): bump the github-actions group with 8 updates #141)

Docs and tests

• An early warning that string and UUID keys make every hop fall back to joins (docs: warn early that string and uuid keys slow traversal #138), runnable quoted-name examples in the Cypher and SQL guides (docs: make the quoted-name examples runnable #143), and a test regex the newer spell checker accepts (test: match the missing-parameter error case-insensitively #142).

The bump itself

• The version moves from 0.8.1 to 0.9.0 in the three crates, the Python package, the bump-my-version config and the three lockfiles. Only version lines change: 8 files, 11 lines each way. No dependency moved.

Validation

• bump-my-version, cargo update --workspace (both workspaces) and uv lock as RELEASE.md describes; no docs or READMEs pin 0.8.1.
• CI passes: tests, clippy, format, lint, spell check and the docs preview.
• The performance numbers come from each PR's own benchmark runs against main at the time; a release doesn't rerun the suites.

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying lance-csr with    Cloudflare Pages

Latest commit:	81bf438
Status:	✅  Deploy successful!
Preview URL:	https://217ad293.lance-csr.pages.dev
Branch Preview URL:	https://chore-bump-0-9-0.lance-csr.pages.dev

View logs