TypedShape::Custom(u32)
is nowTypedShape::Custom(&dyn Shape)
.
- Fix occasional crash in mesh/mesh intersection if some of the vertex coordinates are very small.
- Fix edge case where some of the principal angular inertia are clamped to zero for decimeter-sized objects.
- Have ball-ball shape casting take into account the
stop_on_penetration
flags. - Don’t panic in EPA for a corner case that needs some additional debugging. Show a debug log instead.
- Implement concave polygons intersections:
polygons_intersection_points
,polygon_intersection
.
- Update
bitflags
to version ^2.3 - Update
nalgebra
to 0.33. - Update
indexmap
to 2.
- Fix a regression in ball vs. convex shape contact manifold calculation.
- Add
ShapeCastOptions
that includes two new options for (linear) shape-casting.ShapeCastOptions::target_distance
which will return a hit as soon as the moving shapes are closer than this distance; andcompute_impact_geometry_on_penetration
which forces the calculation of proper witness points and normals even if the shapes are initially intersecting (time_of_impact == 0.0
).
This version modifies many names related to shape-casting:
- Renamed
TOI
toShapeCastHit
. - Renamed
TOIStatus
toShapeCastStatus
. - Rename
RayIntersection::toi
toRayIntersection::time_of_impact
. - More generally, all occurrences of the word
toi
have been replaced bytime_of_impact
for better clarity. - Rename
query::time_of_impact
toquery::cast_shapes
. More generally, all the functions prefixed withtime_of_impact_
(e.g.time_of_impact_ball_ball
) are now prefixed withcast_shapes_
(e.g.cast_shapes_ball_ball
). - Rename
QueryDispatcher::time_of_impact
toQueryDispatcher::cast_shapes
. - The (linear) shape-casting functions like
query::cast_shapes
(previously namedquery::time_of_impact) now take a
ShapeCastOptionsinstead of the
max_toiand
stop_at_penetration` arguments. - Rename
query::nonlinear_time_of_impact
toquery::cast_shapes_nonlinear
. - Rename
QueryDispatcher::nonlinear_time_of_impact
toQueryDispatcher::cast_shapes_nonlinear
. - Rename
NonlinearTOIMode
toNonlinearShapeCastMode
, andNonlinearTOIMode::DirectionalTOI
toNonlinearShapeCastMode::Directional
. - Rename
TimeOfImpactStatus::Penetrating
toShapeCastStatus::PenetratingOrWithinTargetDist
.
- Remove CUDA support to break free from the toolchain restriction required by cust.
- Rework internal edges resolution using normal cones. This implies the modification of the
SimdCompositeShape::map_part_at
,TypedSimdCompositeShape::map_typed_part
, andTypedSimdCompositeShape::map_untyped_part
trait functions so that the closure argument takes an extra argument for the (optional) normal constraints. This argument can be safely ignored by user code unless applying the normal collection is relevant to your use-case. - Contact manifolds will now retain all contacts (including the ones further than the specified
prediction
distance) whenever any contact is actually closer than thisprediction
distance. - Typo fix: renamed
TopologyError::BadAdjascentTrianglesOrientation
toBadAdjacentTrianglesOrientation
.
- Fix contacts between convex shapes being occasionally ignored due to some rounding errors.
- Remove crash when entering unreachable code in non-linear TOI calculation.
- Fix accuracy issue in triangle-mesh center-of-mass calculation when the mesh isn’t manifold.
- Add
SdpMatrix2::inverse_and_get_determinant_unchecked
. This is useful for computing the inverse in a AoSoA SIMD setting. - Add
Aabb::intersects_moving_aabb
to perform a swept test between two moving aabbs.
- Add
Qbvh::traverse_depth_first_with_context
,Qbvh::traverse_depth_first_node_with_stack_and_context
, and the relatedSimdVisitorWithContext
trait to allow parent nodes to pass a custom context to its children during recursion.
- The
point_in_poly2d
now handles arbitrary (convex and non-convex) polygons. The previous implementation that only supported convex polygons has been renamedpoint_in_convex_poly2d
.
- Fix a crash in
Qbvh::refit
that results from the QBVH tree becoming increasingly imbalanced.
- Add
Aabb::scaled_wrt_center
to scale an AABB while keeping its center unchanged.
- Fix ball-convex manifolds missing contacts in some corner cases.
- Fix panic in
TriMesh::intersection_with_plane
.
- When using
rkyv
, fixCheckBytes
implementation for types archived as themselves. - Fix occasional crash in the
QBVH
incremental update.
- Fix
Polyline::flat_indices
that returned an incorrectly sized slice. - Fix serialization of
SimdAabb
into map-like formats like JSON, YAML, RON.
- Add validation when using
rkyv
safe API whenever applicable toparry
types.
- Improved performance of intersection checks involving composite shapes (compound shapes, trimeshes, polylines, etc.)
This version was yanked. See the release notes for 0.13.3 instead.
- Add workaround to address jitter issue due to incorrectly empty contact manifolds generated sometimes for convex
polyhedron.
- About
rkyv
support: most POD structs (Aabb
,Ball
,Cuboid
, etc.) are now archived as themselves instead of being archived as different types (for exampleAabb
is archived asAabb
itself istead ofArchivedAabb
).
- In 3D, add
transformation::try_convex_hull
for a convex hull calculation that will return an error instead of panicking on unsupported inputs.
- Fixed duplicate faces in the connected components returned by
TriMesh::connected_components
.
- Add
TriMesh::canonical_intersection_with_plane
for intersecting with planes aligned with one of the coordinate axes. - Add
TriMesh::intersection_with_plane
for intersecting with arbitrary planes. - Add
TriMesh::intersection_with_local_plane
for intersecting with arbitrary planes in the same space as the mesh - Add
IntersectResult
as the output type for the above functions. - Add
Polyline::extract_connected_components
which splits a compound polyline into its connected components. - Add implementations of
bytemuck::Pod
andbytemuck::Zeroable
for all the simple shapes that allow it (Cuboid
,Ball
,Cone
, etc.), and for bounding volumes (BoundingSphere
andAabb
).
Qbvh::leaf_data
now requires&self
instead of&mut self
.- Replace the
Qbvh::leaf
boolean by a bitflags.
- Add
Qbvh::remove
,Qbvh::pre_update_or_insert
,Qbvh::refit
,Qbvh::rebalance
to allow modifying aQbvh
without having to rebuild it completely. - Add
QbvhNode::is_leaf
to get if a node is a leaf or not. - Add
SharedShape::trimesh_with_flags
for building a trimesh with specific pre-processing flags.
- Fix
Triangle::contains_point
.
- Add
SharedShape::trimesh_with_flags
for constructing a triangle mesh with flags specified by the user.
- Rename
AABB
toAabb
to comply with Rust’s style guide. - Rename
QBVH
toQbvh
to comply with Rust’s style guide.
- Add
ConvexPolygon::offsetted
to dilate a polygon. - Add
CudaTriMesh
andCudaTriMeshPtr
for triangle-meshes usable with CUDA. - Add a no-std implementation of point-projection on a triangle mesh.
- Fix ghost collisions on internal edges on flat 3D meshed and flat 3D heightfields.
- Fix pseudo-normals calculation that could generate invalid normals for triangles with some small vertex angles.
- Fix
Aabb::bounding_sphere
which returned a bounding sphere that was too big.
- Add to
query::time_of_impact
a boolean argumentstop_at_penetration
. If set tofalse
the linear shape-cast won’t immediately stop if the shape is penetrating another shape at its starting point and its trajectory is such that it’s existing that penetration configuration.
- Add 2D
Heightfield::to_polyline
to get the explicit vertices/indices of a 2D heightfield seen as a polyline. - Add the support for linear shape-cast (
query::time_of_impact
) for heightfields. - Make the convex polyhedron scaling more forgiving regarding normals to avoid frequent unjustified panics.
- Fix panic happening when building a convex polyhedron with empty inputs.
- Add the support of Heightfields on CUDA kernels written in Rust using the
cust
crate. - Add the
rkyv-serialize
feature that enables the implementation ofrkyv
serialization/deserialization for most shapes. - Add the
parallel
feature that enables methods for the parallel traversal of Qbvh trees:Qbvh::traverse_bvtt_parallel
,Qbvh::traverse_bvtt_node_parallel
,Qbvh::traverse_depth_first_parallel
,Qbvh::traverse_depth_first_node_parallel
.
- Fix the application of non-uniform scaling to balls.
- Remove
&self
argument fromCompound::decompose_trimesh
. - Switch to
cust
0.3 (for partial CUDA support). - Rename
RoundShape::base_shape
toRoundShape::inner_shape
.
- Allow custom balancing strategies for the Qbvh construction. Some strategies are allowed to generate new leaves during the splitting process.
- Allow using point projection on heightfields from a CUDA kernel.
- Add the simultaneous traversal of two Qbvhs.
- Add computation of
MassProperties
for aTriMesh
. - Add
.to_outline
methods to compute the outline of a 3D shape (useful for debug-rendering). - Add method to apply a scaling factor to some shapes. Shapes not supporting non-uniform scaling (like balls) will return a convex approximation of the scaled result.
- Add methods to split (into up to two parts) a Cuboid, Segment, or TriMesh with an axis-aligned plane.
- Add the computation of the intersection mesh between two TriMesh, or between a Cuboid and a TriMesh.
- Until now, the orientation of the polygon computed by 2D convex hull
computation
parry2d::transformation::convex_hull
andparry2d::transformation::convex_hull_idx
wasn't specified (and was generally in clockwise order). Now, this orientation is explicitly specified in the documentation and is set to counter-clockwise order (which is coherent with orientation expected by, e.g., theConvexPolygon
type).
- Add
parry::utils::obb
which computes a (possibly sub-optimal) OBB for a set of points. - Add
Polyline::project_local_point_assuming_solid_interior_ccw
which projects a point on the polyline contour, and is able to detect if that points is located inside of the polyline, assuming that the polyline is closed and oriented counter-clock-wise. - Add (3D only)
TriMesh::compute_pseudo_normals
. If this is called, and if the trimesh respects some constraints ( oriented with outward normals, manifold almost everywhere, etc.) any subsequent point queries with thesolid
argument set totrue
will properly set thePointProjection::is_inside
to true when the point lies in the interior of the trimesh. - Added the implementation of the ear-clipping and Hertel-Mehlhorn algorithm that can be used for 2D triangulation and convex decomposition.
- Add the ability to use a subset of Parry’s features in a
no-std
context. If the newcuda
cargo feature of Parry is enabled, all features compatible withno-std
can be used inside of a CUDA kernel written in Rust thanks to the rust-cuda ecosystem.
- Fix the orientation of the polygons generated by the 2D convex polygon decomposition (they are now always oriented
counter-clockwise as expected by the
ConvexPolygon
type). - Fix the intersection test between a 2D ball and a 2D compound shape.
- Add the method
Aabb::volume
to compute the volume of an Aabb.
- Update the codebase to use
nalgebra v0.29
.
- Fix a bug where the normal returned by ray-casting on polylines would not be normalized.
- Implement
Debug, Clone, PartialEq
forVHACDParameters
. - Add a method to reverse the order of a polyline.
- Add a method to remove duplicate vertices form a
TriMesh
(and adjusting the index buffer accordingly). - Add a method to iterate through all the lean data stored by a Qbvh.
- Implement the Interval Newton Method for computing all the roots of a non-linear scalar function.
- Implement the intersection test between a spiral and an Aabb.
- Rename all occurrences of
quadtree
toqbvh
. Using the termquadtree
was not representative of the actual acceleration structure being used (which is a BVH).
- Fix a bug where
query::contact
would returnNone
for a all intersecting a cuboid in such a way that one of the cuboid's vertices coincides exactly with the ball's center.
- Updated all dependencies to their latest version.
- Fix ray-casting against solid triangles.
- Fix NaN when adding mass-properties with a zero mass.
ShapeType
now implementsPartialEq, Eq, Hash
.
- The order of vertices output by
Cuboid::to_polyline
has been modified to actually represent the cuboid's boundary (instead of passing through its diagonal).
SharedShape
now implementsAsRef<dyn Shape>
.- Add the optional method
Shape::compute_swept_aabb
to theShape
trait.
- Renamed
SimdQuadTree
toQbvh
(Quaternary Bounding Volume Hierarchy). The incorrect nameSimdQuadTree
is now deprecated. Qbvh::clear_and_rebuild
is now slightly more general.
- Switch to
nalgebra
0.26.
- Add a special case for the triangle/cuboid edge-edge case in the SAT implementation.
- Add contact manifold computation between a convex shape and a HalfSpace.
- Add a special case (instead of GJK) for computing the distance between a ball and a convex shape.
- Add a
Shape::clone_box
method to clone a shape trait-object. - Add a
Shape::make_mut
method to take a mutable reference to the shape trait-object. - Add methods like
Shape::as_shape_mut
to downcast a&mut Shape
to a concrete mutable shape. - Add
MassProperties::set_mass
as a simple way to modify the mass while automatically adjusting the angular inertia. - Make the fields of
BoundingSphere
public. - Add
Aabb::EDGES_VERTEX_IDS
andAabb::FACES_VERTEX_IDS
which are index tables describing the vertices of a given edge, or of a given face, of the Aabb.
- Remove the
target_dist
argument fromquery::time_of_impact
. - The
RigidBodyMotion
and all its implementors have been removed. Use theNonlinearRigidMotion
struct instead. - The
query::nonlinear_time_of_impact
function arguments changed completely to accommodate for the newNonlinearRigidMotion
struct. See the doc ofquery::nonlinear_time_of_impact
for details.
- Fix the separation computation between a cuboid and a support-map shape in the SAT implementation.
- Fix a bug causing invalid results with
query::time_of_impact
whenever the first shape was rotated.
- Add the
Shape::as_typed_shape
method to convert the shape trait-object to an enum. - Add the
TypedShape
enum with variants containing references to concrete shapes (except for user-defined custom shapes). - Add dedicated algorithms for the projection of a point on a cone or a cylinder.
- Improve the overall shape serialization performance.
- Add
Aabb::vertices
to get an array containing its vertices. - Add
Aabb::split_at_center
to split an Aabb into 4 (in 2D) or 8 (in 3D) parts. - Add
Aabb::to_trimesh
to compute the mesh representation of an Aabb.
- Remove the
Shape::as_serialize
method.
- Fix a bug causing making some ball/convex shape contact manifold computation fail when they are penetrating deeply.