Previous algorithms assume that they can get an invalid Voronoi diagram. Because of that, during the multi-material segmentation, a copy of the Voronoi diagram was created, and there were several attempts to fix missing vertices and edges. But as it shows, this wasn't a good enough approach and sometimes led to several issues like bleeding layers.
After generalization, our approach for detection and repairs of invalid Voronoi diagrams from Arachne, we could assume that multi-material segmentation gets non-invalid Voronoi diagrams.
With this assumption, we reimplement multi-materials segmentation to work directly on the Voronoi diagram. That should make multi-material segmentation more stable.
So, this should fix several issues like bleeding layers. Also, memory consumption should decrease by a lot. Also, there should be some speedup of multi-materials segmentation.
into a graph with links to the layer above / below.
In addition:
Members of LayerRegion were made private, public interface const only.
this->m_xxx replaced with just m_xxx
SurfacesPtr was made a vector of const pointers.
This was a regression of 2.4.0-alpha1 wrt. 2.3.3, where handling of
multiple volumes was refactored.
As a result, handling multi-part objects with clip_multipart_objects
disabled was broken.
With clip_multipart_objects disabled, the multi-part objects no more
clip one another and it is sole responsibility of the user to handle
overlaps.
Fixes "add part" error - PrusaSlicer 2.4.0 b2 #7366
Previously, colored polygons were stored so that each polygon had a color assigned to it, which made it difficult to perform operations like union or so on all polygons of the same color. Now polygons are stored grouped by their assigned color/extruder.
for offset2() with clear meaning.
New ClipperUtils functions: expand(), shrink() as an alternative
for offset() with clear meaning.
All offset values for the new functions are positive.
Various offsetting ClipperUtils (offset, offset2, offset2_ex) working
over Polygons were marked as unsafe, sometimes producing invalid output
if called for more than one polygon. These functions were reworked
to offset polygons one by one. The new functions working over Polygons
shall work the same way as the old safe ones working over ExPolygons,
but working with Polygons shall be computationally more efficient.
Improvements in FDM support generator:
1) For both grid and snug supports: Don't filter out supports for which
the contacts are completely reduced by support / object XY separation.
2) Rounding / merging of supports using the closing radius parameter is
now smoother, it does not produce sharp corners.
3) Snug supports: When calculating support interfaces, expand the projected
support contact areas to produce wider, printable and more stable interfaces.
4) Don't reduce support interfaces for snug supports for steep overhangs,
that would normally not need them. Snug supports often produce very
narrow support interface regions and turning them off makes the support
interfaces disappear.
stl_stats are newly only accessed by TriangleMesh::stats(),
most of the direct access to TriangleMesh::stl is gone with the exception
of parsing input files (3MF, AMF, obj).
of a printer from MMU to non-MMU (when number of extruders changes).
Also slightly optimized in case a MMU-painted object is sliced in
single extruder mode.
Renamed its_create_neighbors_index() / its_create_neighbors_index_par() to its_face_neighbors() / its_face_neighbors_par().
New variant of its_face_edge_ids() to create edge IDs from face neighbors.
Fixed some incorrect use of _NDEBUG, it should be NDEBUG.
PrintObject::slice_support_volumes() returns newly Polygons, which are cheaper than ExPolygons.
Updated SeamPlacer and SupportMaterial to use regions defined as Polygons, not ExPolygons.
TriangleSelector::get_facets_strict() returning a patch with T-joints retriangulated.
New slice_mesh_slabs() - slicing projections of a triangle patch into top / bottom layers of slices, for MMU top / bottom segmentation.
TriangleMeshSlicer - use 64 mutexes instead of one when scattering sliced triangles into layers. This makes a big difference on modern many core desktop computers.
When applying MM segmented regions to input regions, the split regions are now re-merged with 10x higher positive offset epsilon to avoid creating gaps.
When testing for existence of paint-on supports or seam, use a more efficient has_facets() test, which does not deserialize into the expensive TriangleSelector tree structure.
GLIndexedVertexArray newly uses Eigen::AlignedBox<float, 3> for efficiency instead of our double based BoundingBoxf3.
Improved MMU painting refresh speed by optimizing generation of the vertex buffers.
Refactored MMU segmentation - projection of painted surfaces from top / bottom.
1) Parallelized.
2) Using the new slice_mesh_slabs() instead of projecting one triangle by the other and merging them with Clipper.
S3D's strategy for merging self intersecting models is "Even / Odd"
which PrusaSlicer now supports as an alternative to "Positive" rule.
Also added a "Close Holes" option to fill in all internal structures.
3D-Labprint Models aren't sliceable (till years) #3062#3708
