Weft v0.4.49

---

ddx:
id: weft.release-0.4.49
type: release-notes
activity: deploy
status: draft
authoring:
home: repo
links:
- id: weft.implementation
kind: informed_by

Weft 0.4.49

Release scope and audience

2026-10-21; Python wheel release for Truss embedders, distributed through GitHub
release assets. The release owner is the repository maintainer. This release
optimizes backend0.4.41 complete-entity metadata lowering.

Highlights and changes

Attribution, timestamps and load metadata each materialize once per source
occurrence and selected row. All eight metadata outputs, model/binding pins,
decoders and native authorization obligations remain checked. Native Truss
tests observe three attribution calls (one direct and two nested proofs), one
timestamp call and one load call per row, rather than eight, two and one.

An explicit optional WASM compile_truss_runtime_json export uses the same Rust
lowering as Python. Five compilation/refusal cases pass byte-exact Chromium
parity. This grants no browser database execution capability.

Breaking changes and required actions

No language or ABI migration is required. Install the matching platform wheel
from the release assets and verify SHA256SUMS. Truss0.0.45 selects this compiler.
The new WASM export requires the truss-postgresql-pg18 build feature.

Migration or rollback guidance

Keep compiler and host pins aligned. Hold adoption if native result or authority
checks fail; restore the previous matched compiler/host release pair. This
release makes no database schema changes.

Known issues and support

Legacy query ABI1.1 retains its full-source population bound before filtering,
aggregation or LIMIT. Wildcard relationship outputs require distinct aliases;
EXCLUDE/REPLACE syntax is not delivered. The sampled Hohfeld host timings do
not establish all proposed latency targets; source, scope and noisy machine
load are recorded in Truss's evidence.

Functional evidence is local macOS ARM64/PostgreSQL18.6: 372 affected Rust tests,
five actual Chromium cases and the final Truss compiled-query/native budget
campaign. Other release platforms receive build and installed-package checks;
native functional equivalence is not claimed for them. This is not whole-workspace
or live Databricks qualification.

References

See qualification,
reproduction and evidence,
and the issue tracker.