Skip to content

Draftwright 0.4.30

Latest

Choose a tag to compare

@pzfreo pzfreo released this 17 Sep 22:57
37128ff

Draftwright 0.4.30 improves imported AP242 PMI fidelity and makes automatic layout failure more honest. It also shortens ordinary pull-request feedback by moving full coverage to explicit and release-grade runs.

AP242 PMI fidelity

  • Conversion-based length units, including inch-authored dimensions, are resolved through complete conversion chains with finite-value and metadata validation. Imported labels retain their authored presentation while normalized values remain available for geometry and comparison.
  • Angular dimensions retain their support members, source association, included-angle geometry, tolerance presentation and editable Sheet representation. Imported angular support patterns now render without losing provenance.
  • Diameter limit presentation, standalone datums, tolerance-zone qualifiers, surface-group labels and common-label relationships survive extraction, lowering and generated-script round trips. Per-hole sources remain attached to the physical occurrence that owns them.
  • CTC04 datum systems preserve corroborating occurrence paths and composite datum support geometry through lowering.

Automatic planning and honest failure

  • Every required placement drop enters the bounded automatic scale/page recovery path, including drawings with no optional isometric view to remove.
  • A recovery candidate cannot win by replacing a required drop with overlapping or otherwise unreadable ink. Benign informational lint remains admissible.
  • When CTC04 exhausts recovery, the returned drawing records one explicit plan_incomplete result whose blocked measurement, hole-requirement and source provenance matches the actual required drops. The engine continues to own placement; no annotation is hand-positioned.

Development workflow

  • Pull requests run changed-area compatibility tests and changed-line coverage. Full coverage remains available through the full-coverage label and full release-grade runs, reducing ordinary PR latency without lowering the project coverage floor.
  • The test corpus reuses expensive recognition, drawing and export evidence more aggressively while retaining scheduled stress, mutation and cross-platform gates.

Known limits

CTC04 is still geometrically infeasible under the bounded current planner; 0.4.30 reports that outcome explicitly rather than presenting overlapping ink or silent required drops as success. Independent B-rep coverage for geometry outside the recogniser vocabulary (#1597) and explicit omissions for undimensioned envelope axes (#1560) remain open trust work for a subsequent release.

Full changelog: v0.4.29...v0.4.30