Skip to content

Latest commit

 

History

History
174 lines (143 loc) · 16.4 KB

NEWS.md

File metadata and controls

174 lines (143 loc) · 16.4 KB

Changelog

1.2.1

  • Weighted triangulations have now been implemented. This is not a minor version bump since the API was already in place for this, but it was not implemented. See #180.

v1.2.0

  • Warnings are now thrown when you try and triangulate point sets not in the plane. The is_planar function has been introduced for this. See #178.

v1.1.4

  • Fixed a bug with curve-bounded refinement with custom edge(s) structs. See #175.

v1.1.3

  • sort_triangle is now public. See #174.

v1.1.2

  • Clarified type stability of triangulate in docstring, and notes about field access and public API. See #171.

v1.1.1

  • Fixed issue on nightly with symbols being marked as both public and exported. See #168.

v.1.1.0

There are a lot of changes in this release, most of them irrelevant for the user. The most important change is the following:

  • We now support a choice between fast, exact, and adaptive predicates via FastKernel(), ExactKernel(), and AdaptiveKernel(), respectively. The default is now AdaptiveKernel(). Moreover, triangle areas are now computed using the adaptive orient predicate to be more robust. See #165.

Previously, ExactPredicates.jl was used everywhere, which can be slow and not necessary for certain point sets. The FastKernel() option has no exact or adaptive arithmetic and so should be used with caution. The documentation discusses these choices in more detail.

To actually configure the choice of predicate, you can e.g. in triangulate use the predicates keyword argument and pass one of DelaunayTriangulation.FastKernel(), DelaunayTriangulation.ExactKernel(), or DelaunayTriangulation.AdaptiveKernel(). If you are computing a predicate manually, then the predicate is instead passed as the first argument.

Some other changes:

  • Added DelauanyTriangulation.validate_triangulation for validating triangulations. See #131.
  • Fixed a bug with the currently unused orient(p, q, r, s) predicate. See #131.
  • Added private functions getz, _getz, getxyz, and _getxyz. See #131.
  • jump_and_march has now been renamed to find_triangle. For compatibility, jump_and_march still works and is simply an alias of find_triangle. See #133.
  • Mutable structs now use const on fields that aren't changed. For compatibility with older versions, this is implemented using a macro that is a no-op where this is not supported. See #140.
  • We now use the public word to define public functions. This is only included on Julia versions v1.11 and above. See #140.
  • We now test on the pre-release. See #140.
  • The module DelaunayTriangulation now has a docstring. See #140.
  • The .md files for tutorials and applications in the docs have been properly updated to match their literate counterparts. See #140.
  • We now use a workflow to enforce changes to NEWS.md for any PRs. See #140.
  • Improved the error message for an incorrect orientation. See #144.
  • Added a CONTRIBUTING.md file and issue templates. See #160.
  • Added is_point2 and is_point3 to detect if a given input is a point. This allows vector coordinates to be passed to convert_boundary_points_to_indices. See #161.
  • Removed an allocation from add_vertex!. See #163.
  • Fixed an issue with the user-supplied rng not being passed to lock_convex_hull!.

v1.0.5

  • Disabled deepcopy on PolygonTrees and made it a no-op. See #129.

v1.0.4

Nothing breaking. Main changes:

  • Fixes some issue with type instabilities
  • Adds construct_polygon_hierarchy to the public API list (it was already intended to be there)
  • All computations are now in the provided precision except for the triangle area and circumcenter which are done in Float64. A warning is given when a non-Float64 precision is given.
  • Closes #118
  • Redesigns the Polygon struct to have an is_circular field, avoiding the need for views
  • Completely refactors validate_triangulation. Still only lives in the test files though. Maybe one day it can live inside the package itself incase users somehow have a use for it..
  • Clean up the runtests.jl file, and make sure that the README/docs are fully tested
  • Use Aqua and fix ambiguities
  • Remove accidental piracy of minimum(::Nothing)
  • Remove CI spam from method redefinitions
  • Fix some issues with doc images and some typos
  • Fix #109

v1.0.3

  • Removed some old function definitions that were no longer needed anymore following the new Makie release.

v1.0.0

In addition to the changes below, note that many bugs have been fixed. Feel free to make any issues or PRs if you encounter any problems.

Added

  • Triangulation and refinement of curve-bounded domains has now been added.
  • The Triangulation struct has some more fields. One of these is weights, although this is not to be used just yet; it is possible to computed weighted Delaunay triangulations of convex polygons, but not for general domains currently.
  • Where reasonable, default methods for the geometric primitive interfaces have now been added. For example, getx now has the default definition getx(p) = p[1].
  • You can now use each_boundary_edge to iterate over the boundary edges (in no specific order), rather than having to use keys(get_boundary_edge_map(tri)) as you had to before 1.0.
  • The documentation has had a complete overhaul.
  • The retriangulate function has now been added, allowing for the retriangulation of a given Triangulation.
  • The function dist can now be used for computing the distance between a point and the triangulation, rather than having to use distance_to_polygon.
  • You can now use find_polygon to find what polygon in a provided boundary of a Triangulation contains a given point.
  • Mesh refinement now, by default, defaults to the diametral lens definition of encroachment rather than the diametral circle definition. You can toggle this behaviour using the use_lens keyword in refine!.
  • Now whenever encroached edges are split during mesh refinement, all free vertices inside that edge's diametral circle will be deleted prior to splitting.

Breaking

Please note that a lot of these changes that follow are only technically breaking, because I failed to properly specify what the public API was for this package (this has now been corrected in v1.0 and onwards).

  • All references to constrained edges are now referred to as segments. For example, the field constrained_edges is now interior_segments, and all_constrained_edges is now all_segments. The same goes for the associated getters.
  • The keyword edges in triangulate is now segments.
  • lock_convex_hull! will now delete from interior_segments field any edges it encounters along the convex hull during locking.
  • add_edge! is now add_segment!.
  • triangulate_convex no longer accepts the recompute_centers keyword.
  • RefinementTargets is now RefinementConstraints.
  • maxiters is no longer a valid keyword argument of refine!. Instead, you should just pass max_points appropriately. With this change, the default for max_points is no longer typemax(Int) but is instead max(1000, num_solid_vertices(tri))^2.
  • refine! no longer returns statistics(tri). It now only returns tri. If you want the statistics, just use statistics(tri) afterwards.
  • refine! no longer accepts max_radius_edge_ratio. Instead, you must provide min_angle. (Internally, it is still max_radius_edge_ratio that gets primarily used, but min_angle is more interpretable for a user.)
  • lock_convex_hull is no longer a keyword of argument of refine! - if it has to be locked, it will be automatically.
  • edge_indices has been removed and is now edge_vertices.
  • indices has been removed and is now triangle_vertices.
  • initialise_edges and initialise_triangles have been removed. Just use the types instead, e.g. initialise_edges(::Type{S}) is now just S().
  • is_empty has been removed. Just use isempty.
  • compare_unoriented_edge is now compare_unoriented_edges.
  • initial and terminal are no longer exported. edge_vertices is more appropriate for this.
  • geti, getj, getk are no longer exported. triangle_vertices is more appropriate for this.
  • each_point and each_point_index are no longer exported. You are encougared to rely on each_solid_vertex instead. Just note that each_solid_vertex is not an ordered iterator.
  • num_points is no longer exported. You are encouraged to instead use num_solid_vertices.
  • remove_duplicate_triangles has been removed. Just use sort_triangles and unique!.
  • I used to refer to distinct parts of a boundary as segments. This is misleading, since it's the first name that I use for constrained edges. So, I now use the term sections. For example, has_multiple_segments is now has_multiple_sections.
  • has_multiple_sections can no longer be used on the ghost vertex map.
  • getboundarynodes has been removed. Just use and extend get_boundary_nodes instead.
  • BoundaryIndex is now GhostVertex. Similar references to boundary_index have been changed to ghost_vertex.
  • boundary_map is now ghost_vertex_map.
  • boundary_index_ranges is now ghost_vertex_ranges.
  • integer_type is removed (exceptfor Triangulations). number_type should be used for this.
  • Previously, get_boundary_index (now called get_ghost_vertex) errored if neither of the three arguments was a ghost vertex. Now, if none of the arguments is a ghost vertex, the function returns k. Similarly for the two-argument version.
  • rotate_ghost_triangle_to_standard_form and rotate_triangle_to_standard_form are removed. Use sort_triangle instead.
  • num_outer_boundary_segments has been removed.
  • Iteration over Adjacent and Adjacent2Vertex has been removed.
  • Adjacent2Vertex is now defined by Adjacent2Vertex{IntType,EdgesType} rather than Adjacent2Vertex{IntType,EdgesType,EdgeType}.
  • clear_empty_points! is now clear_empty_vertices!.
  • The fields of ConvexHull are now points and vertices rather than points and indices. Additionally, the type is now ConvexHull{PointsType, IntegerType} instead of ConvexHull{PointsType, Vector{IntegerType}}.
  • num_points is no longer defined on ConvexHulls.
  • The boundary_map (now ghost_vertex_map) and boundary_index_ranges (now ghost_vertex_ranges) are now Dicts instead of OrderedDicts.
  • The constructor Triangulation(points, triangles, boundary_nodes; kwargs...) now uses the keyword argument delete_ghosts instead of add_ghost_triangles, with default delete_ghosts=false which is opposite to the previous default add_ghost_triagnles=false.
  • The function get_empty_representative_points has been deleted.
  • get_convex_hull_indices is now get_convex_hull_vertices.
  • min_max is removed - just use Base.minmax.
  • nearest_power_of_two has been removed.
  • intersection_of_ray_with_boundary has been removed.
  • identify_side has been removed.
  • intersection_of_ray_with_edge has been removed.
  • The field names of TriangulationStatistics have been renamed to match the new segment terminology, and the field num_convex_hull_points is now num_convex_hull_vertices.
  • get_total_area is now get_area. The field name total_area in TriangulationStatistics is now area.
  • point_position_relative_to_box has been removed.
  • is_boundary_edge has had its definition changed: is_boundary_edge(tri, i, j) is now is_boundary_edge(tri, j, i), so that we have consistency with is_boundary_triangle.
  • is_outer_boundary_index is now is_exterior_ghost_vertex.
  • is_outer_ghost_triangle is now is_exterior_ghost_triangle.
  • is_outer_ghost_edge is now is_exterior_ghost_edge.
  • is_outer_boundary_node is now is_exterior_ghost_vertex.
  • The keyword arguments in add_ghost_triangles! and delete_ghost_triangles! have been removed.
  • The keyword argument exterior_curve_index in triangulate has been removed. The exterior curves will be automatically computed.
  • peek_triangle_ρ has been removed, and triangle_dequeue! now does a dequeue_pair!.
  • The keyword recompute_representative_point is now recompute_representative_points in triangulate.
  • The keyword add_ghost_triangles in triangulate_rectangle (and triangulate_convex) is now delete_ghosts to be consistent with triangulate. The default is delete_ghosts=false, consistent with the previous default of add_ghost_triangles=true.
  • voronoi now has the signature voronoi(tri; clip=false, smooth=false, kwargs...) instead of voronoi(tri, clip=false) (note the difference in clip as a positional argument to a keyword argument). centroidal_smooth is still exported (the kwargs... are passed to it), but this can be a good alternative.
  • Gmsh support has been removed.
  • get! is no longer defined on adjacent2vertex.
  • Certificate is no longer exported.
  • The arguments to construct_ghost_vertex_map (previously construct_boundary_map), construct_boundary_edge_map, and construct_ghost_vertex_ranges (previously construct_boundary_index_ranges) are now positional.
  • get_vertices is no longer exported.
  • map_boundary_index is now map_ghost_vertex.
  • polylabel is no longer an alias for pole_of_inaccessibility.
  • convert_to_boundary_edge is now convert_to_edge_adjoining_ghost_vertex.
  • References to a point_order are now an insertion_order.
  • add_boundary_information! is no longer exported.
  • balanced_power_of_two_quarternary_split has been removed.
  • To provide a custom constraint for triangulation, you now need to provide a constraint of the form (tri::Triangulation, T::Triangle) -> Bool with the keyword argument custom_constraint, where the returned result is true if the triangle T should be refined and false otherwise.
  • is_triangle_seditious no longer has the unused argument ρ at the end of its signature.
  • It is no longer possible to use delete_point! on points that are on the boundary or adjoin segments. It is too finicky, and for constrained triangulations you shouldn't be doing those operations anyway. For unconstrained triangulations, deleting a point on the boundary might be reasonable but it is difficult and delete_point! will not update the convex hull field of the triangulation for you either. The previous implementation did work sometimes, but it required playing around with how ghost triangles are interpreted and was unfortunately not reliable. PRs to address this are welcome, but it would need some careful thought and highly extensive testing (e.g. thousands of random tests for random triangulations, especially triangulations with very skinny or very large triangles on the boundary; triangulations with many collinearities such as from triangulate_rectangle; triangulations with holes; and triangulations with disjoint domains).
  • is_positively_oriented now only accepts integers for the second argument, to avoid issues with misleading results when testing triangles.
  • Some methods like get_right_boundary_node have now been defined only for Triangulations, instead of lowering to a more complicated method that uses the adjacent field and other fields like we used to.