Module dual 

Forward-mode automatic differentiation for constraint residuals.

Every residual is written once, generically over Scalar (the same trait the rest of the kernel runs on), and evaluated either with a plain Scalar value (the value only) or with Dual (the value plus its exact partial derivatives with respect to the constraint’s own variables). A constraint touches only a handful of variables, so a dual number carries a small fixed-size gradient rather than one entry per sketch variable.

Building Dual over an interval Scalar rather than a bare f64 means a residual’s value is a rigorous enclosure of the true real number it stands for, not a float that silently absorbs whatever rounding sin, sqrt or PI introduced along the way — the sketch solver’s own uncertainty about an irrational quantity shows up as interval width instead of vanishing.

Structs

Dual

A value together with its gradient with respect to up to MAX_LOCAL_VARS seeded variables.

Constants

MAX_LOCAL_VARS

The most variables a single constraint may depend on. The largest constraints (tangency or equality between two arcs) touch two arcs of five variables each.