Fusion API field notes

Status: outline

This chapter is scaffolded, not written. Section 1 and the failure modes below are real; sections 2 to 5 are placeholders. PRD 15.3 fixes the seven-part shape used here and states the standard: a dedicated chapter that takes a reader from novice to able to implement — not a literature dump and not a proof, but a constructive path. Do not delete a section heading to avoid filling it in.

1. The problem in one paragraph

The shipped type stubs are a discovery aid, not a contract. Four documented assumptions were overturned by running code against the live application, and the corrections are not obvious from the stubs or from the published documentation. This chapter is the record of what was measured, on which build, with the probe that measured it — so that a future Fusion upgrade that changes one of these is caught by a smoke test rather than by a user with a wrongly cut tube.

2. The minimum background

To be written. Develop the notation from scratch, defining every symbol at first use.

3. The derivation

To be written. In steps small enough that a reader can check each one individually.

4. The algorithm

To be written. Pseudocode using the real function names in this codebase, so a reader can move from the page to the source without a translation step.

5. A worked numerical example

To be written. With actual numbers a reader can reproduce, and the expected output printed in full. This section is what separates a chapter that can be implemented from one that can only be admired.

6. Failure modes

  • Trusting BRepFace.isParamReversed. It genuinely varies between faces in one design. Measure normal . radial and branch on the observed sign (PRD 7.2).

  • Passing a millimeter tolerance to CurveEvaluator3D.getStrokes, which takes centimeters. 0.01 mm is 0.001 there.

  • Assuming getIsoCurve returns one curve. It returns an ObjectCollection that may hold more than one on a trimmed face; always check count.

  • Benchmarking on system Python. Fusion’s interpreter is about five times slower on identical pure-Python code, so an external figure overstates the available headroom fivefold.

7. References