fixed potential deadlock

Feature/mesh improvement

.github/FUNDING.yml

@@ -0,0 +1,4 @@
+# These are supported funding model platforms
+
+open_collective: libomathorg
+github: orange-cpp

CODEOWNERS

@@ -0,0 +1,37 @@
+##   List of maintainers for the omath library
+
+## This file purpose is to give newcomers to the project the responsible
+## developers when submitting a pull request on GitHub, or opening a bug
+## report in issues.
+
+## This file will notably establish who is responsible for a specific
+## area of omath. Being a maintainer means the following:
+##  - that person has good knownledge in the area
+##  - that person is able to enforce consistency in the area
+##  - that person may be available for giving help in the area
+##  - that person has push access on the repository
+## Being a maintainer does not mean the following:
+##  - that person is dedicated to the area
+##  - that person is working full-time on the area/on omath
+##  - that person is paid
+##  - that person is always available
+
+
+# omath core source code
+/source @orange-cpp
+/include @orange-cpp
+
+# Tests and becnchmarks
+/benchmark @orange-cpp
+/tests @orange-cpp @luadebug
+
+# Examples and documentation
+/examples @luadebug @orange-cpp
+/docs @orange-cpp
+
+# Misc like formating
+/scripts @luadebug
+/pixi @luadebug
+
+# CI/CD
+/.github/workflows @luadbg @orange-cpp

INSTALL.md

@@ -1,6 +1,6 @@
 # 📥Installation Guide
 
-## <img width="28px" src="https://vcpkg.io/assets/mark/mark.svg" /> Using vcpkg
+## <img width="28px" src="https://vcpkg.io/assets/mark/mark.svg" /> Using vcpkg (recomended)
 **Note**: Support vcpkg for package management
 1. Install [vcpkg](https://github.com/microsoft/vcpkg)
 2. Run the following command to install the orange-math package:
@@ -28,6 +28,46 @@ target("...")
     add_packages("omath")
 ```
 
+## <img width="28px" src="https://github.githubassets.com/favicons/favicon.svg" /> Using prebuilt binaries (GitHub Releases)
+
+**Note**: This is the fastest option if you don’t want to build from source.
+
+1. **Go to the Releases page**
+   - Open the project’s GitHub **Releases** page and choose the latest version.
+
+2. **Download the correct asset for your platform**
+   - Pick the archive that matches your OS and architecture (for example: Windows x64 / Linux x64 / macOS arm64).
+
+3. **Extract the archive**
+   - You should end up with something like:
+     - `include/` (headers)
+     - `lib/` or `bin/` (library files / DLLs)
+     - sometimes `cmake/` (CMake package config)
+
+4. **Use it in your project**
+
+   ### Option A: CMake package (recommended if the release includes CMake config files)
+   If the extracted folder contains something like `lib/cmake/omath` or `cmake/omath`, you can point CMake to it:
+
+   ```cmake
+   # Example: set this to the extracted prebuilt folder
+   list(APPEND CMAKE_PREFIX_PATH "path/to/omath-prebuilt")
+
+   find_package(omath CONFIG REQUIRED)
+   target_link_libraries(main PRIVATE omath::omath)
+   ```
+   ### Option B: Manual include + link (works with any layout)
+   If there’s no CMake package config, link it manually:
+   ```cmake
+   target_include_directories(main PRIVATE "path/to/omath-prebuilt/include")
+
+    # Choose ONE depending on what you downloaded:
+    # - Static library: .lib / .a
+    # - Shared library: .dll + .lib import (Windows), .so (Linux), .dylib (macOS)
+    
+    target_link_directories(main PRIVATE "path/to/omath-prebuilt/lib")
+    target_link_libraries(main PRIVATE omath)  # or the actual library filename
+    ```
 ## <img width="28px" src="https://upload.wikimedia.org/wikipedia/commons/e/ef/CMake_logo.svg?" /> Build from source using CMake 
 1. **Preparation**
    

include/omath/collision/gjk_algorithm.hpp

@@ -14,11 +14,15 @@ namespace omath::collision
         Simplex<VertexType> simplex; // valid only if hit == true and size==4
     };
 
+    struct GjkSettings final
+    {
+        float epsilon = 1e-6f;
+        std::size_t max_iterations = 64;
+    };
     template<class ColliderInterfaceType>
     class GjkAlgorithm final
     {
         using VectorType = ColliderInterfaceType::VectorType;
-
     public:
         [[nodiscard]]
         static VectorType find_support_vertex(const ColliderInterfaceType& collider_a,
@@ -36,7 +40,8 @@ namespace omath::collision
 
         [[nodiscard]]
         static GjkHitInfo<VectorType> is_collide_with_simplex_info(const ColliderInterfaceType& collider_a,
-                                                                   const ColliderInterfaceType& collider_b)
+                                                                   const ColliderInterfaceType& collider_b,
+                                                                   const GjkSettings& settings = {})
         {
             auto support = find_support_vertex(collider_a, collider_b, VectorType{1, 0, 0});
 
@@ -45,11 +50,11 @@ namespace omath::collision
 
             auto direction = -support;
 
-            while (true)
+            for (std::size_t iteration = 0; iteration < settings.max_iterations; ++iteration)
             {
                 support = find_support_vertex(collider_a, collider_b, direction);
 
-                if (support.dot(direction) <= 0.f)
+                if (support.dot(direction) <= settings.epsilon)
                     return {false, simplex};
 
                 simplex.push_front(support);
@@ -57,6 +62,7 @@ namespace omath::collision
                 if (simplex.handle(direction))
                     return {true, simplex};
             }
+            return {false, simplex};
         }
     };
 } // namespace omath::collision

include/omath/collision/mesh_collider.hpp

@@ -46,9 +46,26 @@ namespace omath::collision
         [[nodiscard]]
         const VertexType& find_furthest_vertex(const VectorType& direction) const
         {
-            return *std::ranges::max_element(
+            // The support query arrives in world space, but vertex positions are stored
-                    m_mesh.m_vertex_buffer, [&direction](const auto& first, const auto& second)
+            // in local space.  We need argmax_v { world(v) · d }.
-                    { return first.position.dot(direction) < second.position.dot(direction); });
+            //
+            // world(v) = M·v  (ignoring translation, which is constant across vertices)
+            // world(v) · d = v · Mᵀ·d
+            //
+            // So we transform the direction to local space once — O(1) — then compare
+            // raw local positions, which is far cheaper than calling
+            // vertex_position_to_world_space (full 4×4 multiply) for every vertex.
+            //
+            // d_local = upper-left 3×3 of M, transposed, times d_world:
+            //   d_local[j] = sum_i  M.at(i,j) * d[i]   (i.e. column j of M dotted with d)
+            const auto& m = m_mesh.get_to_world_matrix();
+            const VectorType d_local = {
+                    m[0, 0] * direction.x + m[1, 0] * direction.y + m[2, 0] * direction.z,
+                    m[0, 1] * direction.x + m[1, 1] * direction.y + m[2, 1] * direction.z,
+                    m[0, 2] * direction.x + m[1, 2] * direction.y + m[2, 2] * direction.z,
+            };
+            return *std::ranges::max_element(m_mesh.m_vertex_buffer, [&d_local](const auto& first, const auto& second)
+                                             { return first.position.dot(d_local) < second.position.dot(d_local); });
         }
         MeshType m_mesh;
     };

include/omath/containers/encrypted_variable.hpp

@@ -62,20 +62,13 @@ namespace omath::detail
         return splitmix64(base_seed() + 0xD1B54A32D192ED03ull * (Stream + 1));
     }
 
-    [[nodiscard]]
-    consteval std::uint64_t bounded_u64(const std::uint64_t x, const std::uint64_t bound)
-    {
-        return (x * bound) >> 64;
-    }
-
     template<std::int64_t Lo, std::int64_t Hi, std::uint64_t Stream>
     [[nodiscard]]
     consteval std::int64_t rand_uint8_t()
     {
         static_assert(Lo <= Hi);
-        const std::uint64_t span = static_cast<std::uint64_t>(Hi - Lo) + 1ull;
         const std::uint64_t r = rand_u64<Stream>();
-        return static_cast<std::int64_t>(bounded_u64(r, span)) + Lo;
+        return static_cast<std::int64_t>(r) + Lo;
     }
     [[nodiscard]]
     consteval std::uint64_t rand_u64(const std::uint64_t seed, const std::uint64_t i)