Open CASCADE Technology Version 8.0.0 Release Candidate 4

Open Cascade is delighted to announce the release of Open CASCADE Technology version 8.0.0 Release Candidate 4 to the public.

• Release Candidate 3
• Release Candidate 2
• Release Candidate 1

Overview

Version 8.0.0-rc4 is a candidate release incorporating 111 improvements and bug fixes compared to version 8.0.0-rc3, bringing the total improvements since version 7.9.0 to over 400 changes.

This release focuses on:

• Redesigned geometry evaluation architecture: New EvalD* API with POD result structs replaces old virtual D0/D1/D2/D3 methods, elementary geometry evaluation devirtualized via std::variant dispatch, new EvalRep descriptor system decouples geometry identity from evaluation strategy, all 29 leaf Geom/Geom2d classes marked final
• Elimination of heap indirection in core geometry classes: BSpline/Bezier classes use direct value-member arrays instead of handle-wrapped heap storage, always-populated weights via static unit-weights buffer eliminates pervasive null-check patterns across ~100 call sites
• Topological data structure overhaul: TopoDS_TShape hierarchy replaces linked-list child storage with contiguous arrays, bit-packs shape state into uint16_t, devirtualizes ShapeType(), and introduces index-based iteration
• Modern C++ foundation layer: Standard_Failure inherits from std::exception, error handlers use thread_local storage eliminating global mutex contention, reference counting uses optimized memory ordering, mesh plugin system replaced with registry-based factory
• New high-performance collections: Robin Hood hash maps (NCollection_FlatDataMap/FlatMap), insertion-order-preserving maps (NCollection_OrderedMap/OrderedDataMap), header-only KD-Tree for spatial queries, C++17 structured binding support via Items() views, unified map API with Contained/TryEmplace/TryBind
• Numerical solver improvements: Laguerre polynomial root-finder, coordinate-wise Brent polishing for PSO/DE improving precision by 4+ orders of magnitude, batch 2D curve evaluation, BSplCLib interpolation hot-path optimizations
• Comprehensive automated migration toolkit: 12-phase Python script suite in adm/scripts/migration_800/ for migrating external projects to 8.0.0 APIs

What is a Release Candidate

A Release Candidate is a tag on the master branch that has completed all test rounds and is stable to use. Release candidates progress in parallel with maintenance releases, with the difference that maintenance releases remain binary compatible with minor releases and cannot include most improvements. The cycle for a release candidate is planned to be 5-10 weeks, while maintenance releases occur once per quarter.

What's New in OCCT 8.0.0-rc4

Foundation Classes

High-Performance Collections

• New NCollection_FlatDataMap and NCollection_FlatMap #1015: Cache-friendly open-addressing hash containers with Robin Hood hashing:

  • All key-value pairs stored inline in contiguous array (eliminates per-element heap allocations)
  • Robin Hood hashing reduces probe sequence variance for more predictable performance
  • Power-of-2 sizing for fast modulo operations via bitwise AND
  • Cached hash codes for faster collision handling and rehashing
  • Also in this PR: optimized Standard_Transient reference counting with explicit memory ordering (following std::shared_ptr pattern), deprecated Standard_Mutex in favor of std::mutex

• New NCollection_KDTree #1073: Header-only static balanced KD-Tree for efficient spatial point queries:

  • Nearest-neighbor search in O(log N), k-nearest, range (sphere), box (AABB), and sphere containment queries
  • Works with gp_Pnt, gp_Pnt2d, gp_XYZ, gp_XY out-of-the-box
  • Optional per-point radii via compile-time template parameter
  • Weighted nearest queries support

• New NCollection_OrderedMap and NCollection_OrderedDataMap #1072: Insertion-order-preserving hash containers using intrusive doubly-linked list:

  • O(1) hash lookup, O(1) append/remove
  • Deterministic iteration in insertion order
  • O(1) removal (unlike NCollection_IndexedMap which requires O(n) swap-and-shrink)

• Try and Emplace methods for NCollection maps* #1022: Non-throwing lookup operations and in-place construction:

  if (auto* pValue = aMap.TryBind(key, defaultValue)) { /* use pValue */ }
  aMap.Emplace(key, constructorArgs...);  // No copy/move!

• Emplace methods for NCollection containers #1035: In-place construction support for NCollection_List (EmplaceAppend, EmplacePrepend, EmplaceBefore, EmplaceAfter), NCollection_Sequence, NCollection_DynamicArray, NCollection_Array1, and NCollection_Array2

• Items() views with C++17 structured bindings #1038: Key-value pair iteration for NCollection map classes:

  for (auto [aKey, aValue] : aMap.Items()) { ... }

  Added Items() for DataMap, FlatDataMap, IndexedDataMap and IndexedItems() for IndexedMap and IndexedDataMap

• NCollection_List optimization #1040: std::initializer_list constructor, improved const-correctness, optimized move constructor, Exchange() method

• Unified map API and collection performance optimizations #1065: API unification and performance improvements across NCollection:

  • Unified map API: Contained() added to all map types returning std::optional<std::reference_wrapper<T>>, TryEmplace/TryBind parity across all map types
  • NCollection_UBTree/EBTree: iterative stack-based traversal replacing recursion (prevents stack overflow on deep trees), move semantics
  • NCollection_LocalArray: move semantics, Reallocate() for use as growable stack
  • NCollection_CellFilter: proper move semantics replacing destructive-copy hack
  • Removed dead Sun WorkShop/Borland compiler workarounds

• TColStd_PackedMapOfInteger refactoring #1023: Improved implementation of specialized integer set

• Keep deprecated NCollection aliases #1026: Deprecated package type aliases (TColStd_*, TopTools_*) kept for backward compatibility with deprecation warnings

Exception Handling Revolution

• Standard_Failure inherits from std::exception #984: Bridges OCCT's exception system with standard C++:

  • OCCT exceptions now caught by standard catch (const std::exception&) blocks
  • Internal storage switched from occ::handle to std::shared_ptr
  • New what() method implements std::exception interface
  • New ExceptionType() virtual method for exception class identification
  • Exception classes simplified to pure data containers
  • Removed Raise(), Instance(), Throw() static methods - use throw instead

• Thread-local error handler stack #980: Replaced global mutex-protected stack with thread_local storage:

  • Zero lock overhead - no mutex acquisition for error handler operations
  • Perfect scalability - threads never contend on error handler state
  • Especially beneficial in TBB/OpenMP parallelized algorithms
  • Removed: Catches(), LastCaughtError() methods
  • Updated OCC_CATCH_SIGNALS macro with new Raise() re-throw method

• Use throw instead of legacy Standard_Failure::Raise #983: Migrated codebase to modern C++ exception throwing

Math and Solver Enhancements

• Cache-friendly matrix multiplication #1015: Changed math_Matrix::Multiply() from i-j-k to i-k-j loop order for row-major storage with significant speedup for large matrices

• SIMD-friendly vector norm #1015: 4-way loop unrolling for math_VectorBase::Norm()/Norm2() with pairwise partial sum combination

• Optimized atomic reference counting #1015: Standard_Transient uses explicit memory ordering (memory_order_relaxed for increment, memory_order_release for decrement with acquire fence only at zero)

• Laguerre polynomial solver and Newton API refactoring #1086: New MathPoly_Laguerre for general polynomial root finding with Laguerre + deflation, including Quintic/Sextic/Octic helpers. Refactored specialized Newton solvers (2D/3D/4D) to unified fixed-size API

• Coordinate-wise polishing for PSO and DE solvers #1088: Brent-based coordinate-wise polishing phase improving component-level precision from ~1e-4 to 1e-8+ for separable functions. New BrentAlongCoordinate in MathUtils_LineSearch, new MathUtils_Random utility

• PLib polynomial evaluation optimization #953

• MathRoot and MathSys enhancements #954: New mathematical utilities

• math_DirectPolynomialRoots refactoring #937

• TKMath modernization with new packages #944

• math_Vector Resize functionality #957

Bnd Package Improvements

• Bnd package improvements #1051: Multiple bug fixes (Bnd_Box::Add, IsOut, Distance; Bnd_Range::Common; Bnd_Sphere::SquareDistances; Bnd_OBB degenerate cases), performance optimizations (early return fast paths for IsOut), and API improvements (Contains()/Intersects() wrappers, Center()/Min()/Max()/Get() returning std::optional, IntersectStatus enum). Added [[nodiscard]] and noexcept annotations

• BVH Box and Rays improvements #882

String Enhancements

• std::u16string_view support for TCollection_ExtendedString #1009: Modern Unicode string handling
• TCollection_AsciiString UTF-8 fix #1070: Fixed multibyte UTF-8 handling in UsefullLength() which was treating individual UTF-8 continuation bytes as non-graphic, causing premature truncation of strings ending with non-ASCII characters

Convert Package Refactoring

• Replace handle-based APIs with direct array access #1057: Replaced heap-allocated handle-based storage with direct NCollection_Array members throughout the Convert package. Deprecated single-element accessors in favor of batch const-reference accessors

Other Foundation Improvements

• TopLoc_Location::HashCode optimization #1006: Faster hash computation for location objects
• gp_Pln refactoring #1003: Improved plane geometry class
• Extend precompiled headers #1029: Faster compilation times

Modeling Data

BSpline/Bezier Memory Optimization

• BSpline/Bezier classes refactored to direct array members #1056: Replaced handle-based NCollection_HArray1/HArray2 members with direct NCollection_Array1/Array2 value members in all Geom BSpline and Bezier classes (curves and surfaces). Eliminates heap indirection and reference counting overhead. Bug fixes for Geom_BSplineCurve::IsEqual skipping knot comparison, Geom_BSplineSurface::SetUNotPeriodic/SetVNotPeriodic wrong constructor, Geom_BezierSurface::Increase self-referencing Init

• Always-populated weights and direct array access #1058: BSpline/Bezier weights arrays are now always populated (non-rational geometry uses non-owning view over static unit-weights buffer, zero allocation). New WeightsArray() accessor always returns valid reference. Bug fix in Hermit.cxx: fixed long-standing typo Pole0 < 3 that should be Pole0 < Pole3

• Optimize BSplCLib interpolation and blend evaluation #1082: Four categories of hot-path optimization: (1) static initialization for GeomFill convertor matrices that were recomputed on every call, (2) stack allocation for small matrices/arrays to avoid heap allocation, (3) raw pointer access in hot loops replacing multi-layer accessor chains with bounds checks, (4) eliminated redundant recomputation with cached solver instances and NbPoles() results. Also fixes undefined behavior in BSplCLib::NbPoles

Geometry Evaluation Overhaul

• EvalRep descriptors and dispatch for Geom/Geom2d #1089: New extensible evaluation dispatch architecture that decouples geometry identity from evaluation strategy. Per-object Set/Get/Clear EvalRep API with support for full, derivative-bounded, and parameter-mapped descriptors. Enables alternate evaluation paths -- e.g., an offset surface can carry its equivalent non-offset surface as an EvalRep, bypassing the expensive offset evaluation path. Migrated Geom_OffsetSurface equivalent-surface path as proof-of-concept

• Redesigned evaluation hierarchy with EvalD0/D1/D2/D3 API #1064, #1094: Fundamental redesign of the geometry evaluation dispatch hierarchy across all 32 Geom/Geom2d curve and surface classes. New EvalD0/EvalD1/EvalD2/EvalD3/EvalDN virtual methods serve as the primary dispatch points, returning new POD result structs (Geom_CurveD1/D2/D3, Geom_SurfD1/D2/D3, Geom2d_CurveD1/D2/D3). Old D0/D1/D2/D3/DN methods retained as non-virtual inline backward-compatible wrappers. The final API uses direct struct returns with exception-based error handling (chosen over std::optional wrapping for evaluation hot-path performance)

• Devirtualize adaptor dispatch, mark leaf Geom classes final #1063: Eliminates virtual method dispatch on the geometry evaluation hot path by storing gp_* primitives directly in std::variant inside adaptor classes and dispatching via switch/enum to ElCLib/ElSLib static methods. Marks all 29 concrete leaf classes as final in Geom_* and Geom2d_* hierarchies, enabling compiler devirtualization. Bug fix: ShallowCopy in GeomAdaptor_Curve/Surface now correctly copies elementary types in the variant

• Geom2dGridEval package for batch 2D curve evaluation #1079: New package mirroring GeomGridEval for 3D, providing batch evaluation of 2D curves with specialized evaluators for conics (analytical) and BSpline/Bezier (cache-based)

• Optimize adaptor Bezier cache and grid eval threshold #1084: Removed redundant IsCacheValid() checks for Bezier curves/surfaces (single span, always valid), lowered cache threshold for more aggressive cache-based evaluation

TShape Hierarchy Optimization

• TShape hierarchy redesign for performance and memory efficiency #1027: Fundamental redesign of OCCT's most critical topological data structure:
  • Child storage changed from NCollection_List (linked list) to NCollection_DynamicArray (contiguous memory) with type-specific default bucket sizes (e.g., TEdge=2 for vertices, TWire=8 for edges)
  • ShapeType() now non-virtual - embedded in compact uint16_t myState bit-packed field (4 bits for type, 8 bits for flags)
  • New TopoDS_TShapeDispatch for std::visit-style devirtualized type dispatch
  • TopAbs::Compose(), Reverse(), Complement() moved inline to header
  • TopoDS_Iterator refactored from list-based to index-based iteration
  • Result: Smaller TShape objects, cache-friendly child traversal, faster shape exploration

Other Modeling Data Improvements

• Simplify EmplaceValue in Array1 and Array2 #1087

Modeling Algorithms

Geometry Evaluation Optimization

• GeomGridEval optimization and simplification #908, #951, #952, #1031: Improved surface grid evaluation with sequential processing
• Bnd_BoundSortBox optimization #958
• Optimized point-to-plane projection helper #959: Batch processing support
• Optimize properties computation for complex compounds #1091: Reduced TopLoc_Location composition overhead in edge pcurve lookup, added fast-path exits in TopLoc_Location::Predivided() for identity and equal-location cases, cached face surface/location in BRepGProp_Face

Algorithm Refactoring

• IntCurveSurface and HLRBRep intersection refactoring #912, #936: Complete code sharing for Polyhedron classes
• Offset curve and surface evaluators refactoring #930
• Extrusion and revolution Utils refactoring #948: Accept pre-computed curve values
• Evaluator classes refactoring #935: Inline Utils and variant-based Adaptors
• HLRAlgo_PolyData::Box replacement with Bnd_Box #923
• Extrema_ExtPS refactoring #978
• BOPDS refactoring #1007
• IntTools Box calculation optimization #990
• Update tolerance settings in BRepBlend_AppFuncRoot #1083: Regression fix after BSplCLib interpolation optimization

Bug Fixes

• Fixed solid-level caching bugs in BRepGProp volume properties #1092: Fixed SkipShared semantics broken for same-placement duplicates causing double-counting, and free faces/shells dropped when shared solids exist
• Fixed crash in ComputePolesIndexes() #1049: Fixed bounds checking where theOutMinIdx could exceed upper bound and theOutMaxIdx could be less than lower bound
• Fixed partial torus creation with inverted V range #928
• Fixed 0-based index in BRep_Tool::CurveOnSurface call #949
• Fixed curve concatenation to use actual endpoints #926
• Fixed unnecessary loop iteration in BRepLib::BuildCurve3d #921

Shape Healing

• GlueEdgesWithPCurves validation fix #981
• Unstable PCurve Processing fix #967
• Remove edges from map during face unification in ShapeUpgrade_UnifySameDomain #941

Mesh

• Registry-based factory pattern replacing plugin system #1033: Major architecture change:

  • Replaced legacy DISCRETPLUGIN/DISCRETALGO symbol-based plugin system
  • New BRepMesh_DiscretAlgoFactory abstract base with static registry
  • BRepMesh_IncrementalMeshFactory for "FastDiscret" algorithm
  • XBRepMesh_Factory for "XBRepMesh" extended meshing
  • Eliminated symbol collisions when TKMesh and TKXMesh both loaded
  • Cleaner C++ design without dynamic symbol lookup
  • Removed: BRepMesh_PluginMacro.hxx, BRepMesh_PluginEntryType.hxx, BRepMesh_FactoryError.hxx
  • Removed Draw commands: mpsetfunctionname, mpgetfunctionname, mperror

• Fixed point-in-polygon check for CCW polygons in BRepMesh_Delaun #920

Visualization

• Remove obsolete UNLIT shading optimization #1069: Removed implicit optimization in OpenGl_Aspects that forced UNLIT shading when material had no reflection properties. This was breaking PBR materials, interior color handling, and texture modulation. Legacy code now explicitly sets SetShadingModel(Unlit)

• Avoid redundant shape copies in AIS_ColoredShape::dispatchColors #1068: Deferred EmptyCopied() and BRep_Builder::Add() to avoid redundant shape construction; compound built only when required

Data Exchange

• STEP: Refactor pnindex handling in CreatePolyTriangulation #1067: Separated direct node indexing and pnindex remapping into distinct code paths

Build and Configuration

• Accept empty FILES content #1017
• Clean up the FILES #1002
• Disable usage of VTK by default #939
• Fix CMake static linking warnings #1075: Fixed missing spaces in warning messages

Testing

• Update CI workflow to build and test on Ubuntu with GCC #1028
• Use current run ID to download test results for platforms #1025
• Update font installation process for Windows to use Noto Sans CJK #999
• Remove long-running DRAW test cases with low diagnostic value #1093: Removed 5 test cases consuming ~590s total CI time
• Cleanup GTests layout and move TKXCAF test #1080: Removed empty GTests stubs, moved XCAFDoc_Test.cxx to correct location

Coding Quality

Global Refactoring

• Global Refactoring OCCT as part of 8.0.0 #955: Comprehensive codebase modernization
• Clang-Tidy application with refactoring #965, #977
• Fix GCC warnings #975
• Fixed MSVC warnings #1004, #1060
• Suppress macOS system header warnings #997
• Fix critical CodeQL static analysis warnings #1074: Fixed use-after-free in Interface_ParamSet::Append(), upcast pointer arithmetic bug in delabella.cpp, signed integer overflow in AdvApp2Var_MathBase. Also fixed pure virtual call during destruction in NCollection_SparseArrayBase by replacing virtual dispatch with function pointers, eliminating the vtable entirely
• Fix clang warning suppression for function pointer casts #1059
• Fix compilation warnings #1034

Code Cleanup

• HArray and HSequence Definitions refactoring #962
• HLRBRep algorithms: Replace Standard_Address with typed pointers #947, #961
• Remove unused typedefs and includes #971
• Remove unused code and comments #968
• Remove redundant null checks before deallocation #1077: Cleaned up 39 files
• BSplineCurve and BSplineSurface parameter preparation refactoring #972
• Add constexpr compatibility to more gp classes #933
• Translate French comments and modernize constants #932
• Move semantics and default constructor for CSLib_Class2d #919
• Optimize memory management in BOPAlgo classes #915
• Prevent copy and move operations in BRepAlgoAPI_BuilderAlgo #913
• Remove unused Bnd_Box and related code from BRepClass3d_SolidExplorer #931
• Revert type definitions for Standard_CString replacements #1021
• Include Standard_ErrorHandler header in OSD_signal #996
• Remove redundant pragma lib comment in OSD_Host.cxx #902

Temporary Changes

• Temporarily remove samples from the repository #960

Documentation

• Improve code examples in modeling algorithms #1016
• Refactor documentation with new coding rules #1013
• Update AI Assistant guidelines #1005
• Fix issue with documentation build #992

Migration Guide

OCCT 8.0.0 includes a comprehensive set of automated migration scripts in adm/scripts/migration_800/ to help external projects adapt to the new API. These scripts require Python 3.6+ with no external dependencies.

Automated Migration Toolkit

The migration scripts can be run all at once or individually. Each script supports --dry-run to preview changes before applying.

Full Automated Migration (Recommended)

# Linux/macOS - full migration with preview
./adm/scripts/migration_800/run_migration.sh /path/to/your/src --dry-run

# Linux/macOS - apply changes
./adm/scripts/migration_800/run_migration.sh /path/to/your/src

# Windows
adm\scripts\migration_800\run_migration.bat /path/to/your/src --dry-run

The full migration runs 12 phases in order:

After all phases, verify_migration.py runs automatically to report any remaining legacy patterns.

Individual Scripts for Targeted Migration

Each script can be run independently for granular control:

Handle Migration

python3 adm/scripts/migration_800/migrate_handles.py --dry-run /path/to/your/src

// Before                                    // After
Handle(Geom_Circle) aCircle;                 occ::handle<Geom_Circle> aCircle;
Handle(Geom_Circle)::DownCast(aCurve);       occ::down_cast<Geom_Circle>(aCurve);

Standard_* Type Migration

python3 adm/scripts/migration_800/migrate_standard_types.py --dry-run /path/to/your/src

// Before                                    // After
Standard_Boolean isOk = Standard_True;       bool isOk = true;
Standard_Integer aCount = 0;                 int aCount = 0;
Standard_Real aTol = 0.001;                  double aTol = 0.001;
Standard_CString aName = "test";             const char* aName = "test";

Full type mapping:

Standard_* Macro Migration

python3 adm/scripts/migration_800/migrate_macros.py --dry-run /path/to/your/src

Note: Standard_UNUSED requires manual migration to [[maybe_unused]] due to stricter placement rules.

NCollection Typedef Migration

# Step 1: Use pre-generated JSON from OCCT (or collect from source)
python3 adm/scripts/migration_800/replace_typedefs.py --dry-run \
    --input adm/scripts/migration_800/collected_typedefs.json /path/to/your/src

// Before                                    // After
TColStd_ListOfInteger aList;                 NCollection_List<int> aList;
TopTools_MapOfShape aMap;                    NCollection_Map<TopoDS_Shape> aMap;
TColStd_Array1OfReal anArr;                  NCollection_Array1<double> anArr;
TColgp_SequenceOfPnt aSeq;                   NCollection_Sequence<gp_Pnt> aSeq;

Pre-generated JSON files are included in adm/scripts/migration_800/ so external projects do not need to re-scan the OCCT source:

• collected_typedefs.json - NCollection typedef mappings
• collected_deprecated_typedefs.json - Deprecated typedef patterns and replacements

Exception Raise Migration

python3 adm/scripts/migration_800/migrate_raise_to_throw.py --dry-run

// Before (removed)                          // After (required)
Standard_Failure::Raise("error");            throw Standard_Failure("error");
Standard_OutOfRange::Raise("index");         throw Standard_OutOfRange("index");

H-Collection Macro Migration

python3 adm/scripts/migration_800/migrate_hcollections.py --dry-run

Converts DEFINE_HARRAY1, DEFINE_HARRAY2, and DEFINE_HSEQUENCE macros to the new NCollection_HArray1/HArray2/HSequence template classes.

Verification

After migration, verify completeness:

python3 adm/scripts/migration_800/verify_migration.py --verbose /path/to/your/src

Manual Migration Notes

The following API changes require manual attention beyond what the automated scripts handle.

Exception Handling

OCCT exceptions now inherit from std::exception and can be caught using standard C++ exception handling:

// Now works! (OCCT exceptions inherit from std::exception)
try {
    // OCCT operations
} catch (const std::exception& e) {
    std::cerr << e.what() << std::endl;
}

// Or catch specific OCCT exception types
try {
    // OCCT operations
} catch (const Standard_Failure& e) {
    std::cerr << e.ExceptionType() << ": " << e.what() << std::endl;
}

The GetMessageString() method is deprecated; use what() instead.

NCollection_Map API Changes

Seek()/ChangeSeek() have been removed from NCollection_Map. Use Contained() instead:

// Before (removed)
const KeyType* pKey = aMap.Seek(aKey);

// After
auto anOpt = aMap.Contained(aKey);
if (anOpt.has_value()) { const KeyType& aFoundKey = anOpt->get(); }

BSpline/Bezier Weights Migration

The nullable Weights() pattern has been replaced with always-valid WeightsArray():

// Before (nullable)
const NCollection_Array1<double>* pWeights = aCurve->Weights();
if (pWeights != nullptr) { /* use weights */ }

// After (always valid)
const NCollection_Array1<double>& aWeights = aCurve->WeightsArray();
// Non-rational curves return a view over static unit-weights buffer (no allocation)

Copy-out accessor overloads are deprecated in favor of const-reference returning versions:

// Before (deprecated, copies data)
NCollection_Array1<gp_Pnt> aPoles;
aCurve->Poles(aPoles);

// After (zero-copy)
const NCollection_Array1<gp_Pnt>& aPoles = aCurve->Poles();

Convert Package Migration

Single-element accessors are deprecated in favor of batch const-reference accessors:

// Before (deprecated)
gp_Pnt aPole = aConverter.Pole(i);

// After
const NCollection_Array1<gp_Pnt>& aPoles = aConverter.Poles();

Visualization: UNLIT Shading

The implicit optimization that forced UNLIT shading when material had no reflection properties has been removed. If relying on zero-material properties for UNLIT shading, explicitly set the shading model:

// Before (implicit, now removed)
// Setting material with no reflection would auto-switch to UNLIT

// After (explicit)
anAspect->SetShadingModel(Graphic3d_TypeOfShadingModel_Unlit);

Mesh Plugin System Migration

The legacy DISCRETPLUGIN/DISCRETALGO symbol-based plugin system has been replaced with a registry-based factory:

// Before (removed)
BRepMesh_PluginEntryType aFunc = BRepMesh::PluginEntry("DISCRETPLUGIN");

// After
occ::handle<BRepMesh_DiscretAlgoFactory> aFactory = BRepMesh_DiscretAlgoFactory::FindFactory("FastDiscret");
if (!aFactory.IsNull()) {
    occ::handle<BRepMesh_DiscretAlgo> anAlgo = aFactory->Create();
}

Bnd Package API Changes

Bnd_Range::IsIntersected magic int returns replaced with IntersectStatus enum:

// Before
int aResult = aRange.IsIntersected(...);

// After
Bnd_Range::IntersectStatus aResult = aRange.IsIntersected(...);

Geom Classes Marked Final

All 29 concrete leaf classes in Geom_* and Geom2d_* hierarchies are now marked final. If your code inherits from these concrete classes (e.g., Geom_BSplineCurve, Geom_Plane), this will cause a compilation error. Inherit from the abstract base classes instead (e.g., Geom_BoundedCurve, Geom_ElementarySurface).

Removed Functionality

Deprecated Functionality

Performance Improvements Summary

• Devirtualized adaptor dispatch #1063: Eliminates virtual calls for elementary geometry evaluation via std::variant
• BSpline/Bezier direct array members #1056: Eliminates heap indirection and reference counting
• Always-populated weights with zero-alloc views #1058: Eliminates null checks and copies for non-rational geometry
• BSplCLib interpolation optimization #1082: Static init, stack allocation, raw pointer hot loops
• Properties computation for complex compounds #1091: TopLoc_Location fast paths, cached face surface/location
• Thread-local error handling #980: Zero lock overhead for exception handling in parallel code
• TShape hierarchy optimization #1027: Non-virtual ShapeType(), compact state storage, better cache locality
• NCollection_FlatDataMap/FlatMap #1015: Robin Hood hashing with open addressing for cache-friendly hash maps
• Matrix multiplication optimization #1015: Cache-friendly i-k-j loop order
• Vector norm with SIMD unrolling #1015: 4-way loop unrolling
• Optimized atomic reference counting #1015: Explicit memory ordering
• NCollection_List optimization #1040: Optimized move constructor, improved const-correctness
• Tree and collection optimizations #1065: Iterative traversal, move semantics across containers
• AIS_ColoredShape::dispatchColors #1068: Avoids redundant shape copies
• TopLoc_Location::HashCode optimization #1006
• Adaptor Bezier cache optimization #1084: Removed redundant validity checks
• Bnd package optimizations #1051: Early return fast paths
• PLib polynomial evaluation optimization #953
• GeomGridEval optimization #908, #951, #952, #1031
• Bnd_BoundSortBox optimization #958
• IntTools Box calculation optimization #990
• Coordinate-wise polishing for PSO/DE #1088: Precision improvement to 1e-8+
• Extended precompiled headers #1029: Faster compilation

Acknowledgments

We thank all contributors who helped make this release possible through their code contributions, bug reports, and testing.

New Contributors

• @Andrej730 made their first contribution in #902

Full Changelog: V8_0_0_rc3...V8_0_0_rc4