0.3.0
Performance Optimizations
Adding
__slots__for all classesRemoving all usage of
@propertycalls from internal callers (to avoid function call overhead)Avoiding un-necessary data copying, e.g.
nodes[[0], :]creates a copy butnodes[0, :].reshape((1, 2))does not (more details)Adding
_verifyswitches to selectively disable overly defensive value checking. Added toCurvedPolygonconstructor,Surface.evaluate_barycentric()(doc),Surface.evaluate_cartesian()(doc),Surface.evaluate_multi()(doc) andSurface.intersect()(doc). Internal callers with already verified data now skip verification stepsBailing out early if surface bounding boxes are disjoint in
Surface.intersect()method
Breaking Changes
Requiring
degreeinCurve(doc) andSurface(doc) constructors, but addingCurve.from_nodes()(doc) andSurface.from_nodes()(doc) factories to accept nodes only (computing the degree in the constructor every time is a waste of flops, especially if the caller knows the degree)Removing custom equality checks for
Curve(doc) andSurface(doc) objects. The previous implementation did not factor in all relevant valuesReturning
1xDarrays instead of flattenedD-dimensional 1D arrays fromCurve.evaluate()(doc),Surface.evaluate_barycentric()(doc),Surface.evaluate_cartesian()(doc), and related helpersRenaming
Intersection.left/rightproperties asfirst/secondproperties (They were poorly named originally, since “left” and “right” were in reference to where they were used in code, not geometry. This class is not part of the public interface, but it is documented.)
Bug Fixes
Handling cases where one corner of a surface touches another but their interiors don’t intersect (in the
Surface.intersect()method). Addingignored_cornerclassification to handle these curve-curve intersecions that don’t contribute to a surface-surface intersectionThrowing exception in the
Curve.locate()method when the subdivided intervals are very far apart (#13)Improving
Surface.is_validproperty by considering the signs of the Jacobian determinant at corner nodes (#12)
Miscellany
Adding possible strategy to avoid linear convergence in
newton_refine()helperAdding AppVeyor configuration to make sure there are no Windows issues, testing exclusively with
condainstallUpdating generated images with
matplotlib2.0