Mesh.hpp

#pragma once
 
#include <Core/Asset/GeometryData.hpp>
#include <Core/Containers/VectorArray.hpp>
#include <Core/Geometry/MeshPrimitives.hpp>
#include <Core/Geometry/StandardAttribNames.hpp>
#include <Core/Geometry/TriangleMesh.hpp>
#include <Core/Utils/BijectiveAssociation.hpp>
#include <Core/Utils/Color.hpp>
#include <Core/Utils/Log.hpp>
#include <Core/Utils/ObjectWithSemantic.hpp>
#include <Engine/Data/DisplayableObject.hpp>
#include <Engine/Data/ShaderProgram.hpp>
#include <Engine/RaEngine.hpp>
 
#include <globjects/Buffer.h>
#include <globjects/Program.h>
#include <globjects/VertexArray.h>
#include <globjects/VertexAttributeBinding.h>
 
#include <array>
#include <iterator>
#include <map>
#include <vector>
 
namespace Ra {
namespace Engine {
namespace Data {
class ShaderProgram;
 
using namespace Ra::Core::Utils;
 
class RA_ENGINE_API Vao
{
    std::unique_ptr<globjects::VertexArray> m_vao;
    std::vector<std::unique_ptr<globjects::Buffer>> m_vbos;
    std::vector<bool> m_dataDirty;
    std::map<std::string, int> m_handleToBuffer;
};
 
class RA_ENGINE_API AttribArrayDisplayable : public Displayable
{
  public:
    enum MeshRenderMode : uint {
        RM_POINTS                   = 0x0000,
        RM_LINES                    = 0x0001, // decimal value: 1
        RM_LINE_LOOP                = 0x0002, // decimal value: 2
        RM_LINE_STRIP               = 0x0003, // decimal value: 3
        RM_TRIANGLES                = 0x0004, // decimal value: 4
        RM_TRIANGLE_STRIP           = 0x0005, // decimal value: 5
        RM_TRIANGLE_FAN             = 0x0006, // decimal value: 6
        RM_QUADS                    = 0x0007, // decimal value: 7
        RM_QUAD_STRIP               = 0x0008, // decimal value: 8
        RM_POLYGON                  = 0x0009, // decimal value: 9
        RM_LINES_ADJACENCY          = 0x000A, // decimal value: 10
        RM_LINE_STRIP_ADJACENCY     = 0x000B, // decimal value: 11
        RM_TRIANGLES_ADJACENCY      = 0x000C, // decimal value: 12
        RM_TRIANGLE_STRIP_ADJACENCY = 0x000D, // decimal value: 13
        RM_PATCHES                  = 0x000E, // decimal value: 14
    };
 
  public:
    explicit AttribArrayDisplayable( const std::string& name,
                                     MeshRenderMode renderMode = RM_TRIANGLES );
    AttribArrayDisplayable( const AttribArrayDisplayable& rhs ) = delete;
    void operator=( const AttribArrayDisplayable& rhs )         = delete;
 
    ~AttribArrayDisplayable() {}
 
    using Displayable::getName;
 
    inline void setRenderMode( MeshRenderMode mode );
    inline MeshRenderMode getRenderMode() const;
 
 
    void setDirty( const Core::Geometry::MeshAttrib& type );
 
    void setDirty( const std::string& name );
 
    void setDirty( unsigned int index );
 
    void updateGL() override = 0;
 
    virtual const Core::Geometry::AttribArrayGeometry& getAttribArrayGeometry() const = 0;
    virtual Core::Geometry::AttribArrayGeometry& getAttribArrayGeometry()             = 0;
 
    Ra::Core::Utils::optional<gl::GLuint> getVboHandle( const std::string& name );
 
    Ra::Core::Utils::optional<gl::GLuint> getVaoHandle();
 
  protected:
    void updatePickingRenderMode();
 
    class AttribObserver
    {
      public:
        explicit AttribObserver( AttribArrayDisplayable* displayable, int idx ) :
            m_displayable( displayable ), m_idx( idx ) {}
        void operator()() {
            if ( m_idx < int( m_displayable->m_dataDirty.size() ) ) {
                m_displayable->m_dataDirty[m_idx] = true;
                m_displayable->m_isDirty          = true;
            }
            else {
                LOG( logDEBUG ) << "Invalid dirty bit notified on " << m_displayable->getName();
            }
        }
 
      private:
        AttribArrayDisplayable* m_displayable;
        int m_idx;
    };
 
  protected:
    std::unique_ptr<globjects::VertexArray> m_vao;
 
    MeshRenderMode m_renderMode { MeshRenderMode::RM_TRIANGLES };
 
    // m_vbos and m_dataDirty have the same size and are indexed thru m_handleToBuffer[attribName]
    std::vector<std::unique_ptr<globjects::Buffer>> m_vbos;
    std::vector<bool> m_dataDirty;
 
    // Geometry attrib name (std::string) to buffer id (int)
    // buffer id are indices in m_vbos and m_dataDirty
    std::map<std::string, unsigned int> m_handleToBuffer;
 
    bool m_isDirty { false };
};
 
class RA_ENGINE_API VaoIndices
{
  public:
    inline void setIndicesDirty();
 
    class IndicesObserver
    {
      public:
        explicit IndicesObserver( VaoIndices* displayable ) : m_displayable { displayable } {}
        void operator()() { m_displayable->m_indicesDirty = true; }
 
      private:
        VaoIndices* m_displayable;
    };
 
  protected:
    std::unique_ptr<globjects::Buffer> m_indices { nullptr };
    bool m_indicesDirty { true };
    size_t m_numElements { 0 };
};
 
template <typename I>
class IndexedAttribArrayDisplayable : public AttribArrayDisplayable, public VaoIndices
{
    using IndexType          = I;
    using IndexContainerType = Ra::Core::AlignedStdVector<IndexType>;
 
    template <typename T>
    inline void addAttrib( const std::string& name,
                           const typename Ra::Core::Utils::Attrib<T>::Container& data );
    template <typename T>
    inline void addAttrib( const std::string& name,
                           const typename Ra::Core ::Utils::Attrib<T>::Container&& data );
    inline void updateGL() override;
 
    inline void render( const ShaderProgram* prog ) override;
 
  protected:
    inline void autoVertexAttribPointer( const ShaderProgram* prog );
    IndexContainerType m_cpu_indices;
    AttribManager m_attribManager;
};
 
template <typename T>
class CoreGeometryDisplayable : public AttribArrayDisplayable
{
  public:
    using base         = AttribArrayDisplayable;
    using CoreGeometry = T;
 
    explicit CoreGeometryDisplayable( const std::string& name,
                                      MeshRenderMode renderMode = RM_TRIANGLES );
 
    // no need to detach observer in dtor since CoreGeometry is owned by this, and CoreGeometry dtor
    // will detachAll observers.
 
    inline const Core::Geometry::AbstractGeometry& getAbstractGeometry() const override;
    inline Core::Geometry::AbstractGeometry& getAbstractGeometry() override;
 
    inline const Core::Geometry::AttribArrayGeometry& getAttribArrayGeometry() const override;
    inline Core::Geometry::AttribArrayGeometry& getAttribArrayGeometry() override;
 
    inline const CoreGeometry& getCoreGeometry() const;
    inline CoreGeometry& getCoreGeometry();
 
    template <typename A>
    inline Ra::Core::Utils::AttribHandle<A> addAttrib( const std::string& name,
                                                       const typename Core::VectorArray<A>& data );
    inline size_t getNumVertices() const override;
 
    virtual void loadGeometry( CoreGeometry&& mesh );
 
    void updateGL() override;
 
    void setAttribNameCorrespondance( const std::string& meshAttribName,
                                      const std::string& shaderAttribName );
 
  protected:
    virtual void updateGL_specific_impl() {}
 
    void loadGeometry_common( CoreGeometry&& mesh );
    void setupCoreMeshObservers();
 
    void autoVertexAttribPointer( const ShaderProgram* prog );
 
    void addAttribObserver( const std::string& name );
 
    void addToTranslationTable( const std::string& name );
 
    BijectiveAssociation<std::string, std::string> m_translationTable {};
 
    CoreGeometry m_mesh;
};
 
class RA_ENGINE_API PointCloud : public CoreGeometryDisplayable<Core::Geometry::PointCloud>
{
    using base = CoreGeometryDisplayable<Core::Geometry::PointCloud>;
 
  public:
    using base::CoreGeometryDisplayable;
    inline explicit PointCloud(
        const std::string& name,
        typename base::CoreGeometry&& geom,
        typename base::MeshRenderMode renderMode = base::MeshRenderMode::RM_POINTS );
 
    inline explicit PointCloud( const std::string& name, MeshRenderMode renderMode = RM_POINTS );
 
    void render( const ShaderProgram* prog ) override;
 
    void loadGeometry( Core::Geometry::PointCloud&& mesh ) override;
 
  protected:
    void updateGL_specific_impl() override;
};
 
template <typename T>
class IndexedGeometry : public CoreGeometryDisplayable<T>, public VaoIndices
{
  public:
    using base = CoreGeometryDisplayable<T>;
    using CoreGeometryDisplayable<T>::CoreGeometryDisplayable;
    explicit IndexedGeometry(
        const std::string& name,
        typename base::CoreGeometry&& geom,
        typename base::MeshRenderMode renderMode = base::MeshRenderMode::RM_TRIANGLES );
 
    void render( const ShaderProgram* prog ) override;
 
    void loadGeometry( T&& mesh ) override;
 
  protected:
    void updateGL_specific_impl() override;
};
 
template <typename T>
class MultiIndexedGeometry : public CoreGeometryDisplayable<T>
{
  public:
    using base = CoreGeometryDisplayable<T>;
    using CoreGeometryDisplayable<T>::CoreGeometryDisplayable;
    explicit MultiIndexedGeometry(
        const std::string& name,
        typename base::CoreGeometry&& geom,
        typename base::MeshRenderMode renderMode = base::MeshRenderMode::RM_TRIANGLES );
    void render( const ShaderProgram* prog ) override;
 
    void loadGeometry( T&& mesh ) override;
 
  protected:
    void updateGL_specific_impl() override;
 
    using LayerSemanticCollection = Core::Utils::ObjectWithSemantic::SemanticNameCollection;
    using LayerSemantic           = Core::Utils::ObjectWithSemantic::SemanticName;
    using LayerKeyType            = std::pair<LayerSemanticCollection, std::string>;
 
    using EntryType = std::pair<bool, VaoIndices*>;
    struct RA_CORE_API KeyHash { std::size_t operator()( const LayerKeyType& k ) const {
        // Mix semantic collection into a single identifier string
        std::ostringstream stream;
    std::copy( k.first.begin(), k.first.end(), std::ostream_iterator<std::string>( stream, "" ) );
    std::string result = stream.str();
    std::sort( result.begin(), result.end() );
 
    // Combine with layer name hash
    return std::hash<std::string> {}( result ) ^ ( std::hash<std::string> {}( k.second ) << 1 );
}
}; // namespace Data
std::unordered_map<LayerKeyType, EntryType, KeyHash> m_indices;
}; // namespace Engine
 
class RA_ENGINE_API LineMesh : public IndexedGeometry<Core::Geometry::LineMesh>
{
    using base = IndexedGeometry<Core::Geometry::LineMesh>;
 
  public:
    using base::IndexedGeometry;
    inline explicit LineMesh(
        const std::string& name,
        typename base::CoreGeometry&& geom,
        typename base::MeshRenderMode renderMode = base::MeshRenderMode::RM_LINES );
    inline explicit LineMesh(
        const std::string& name,
        typename base::MeshRenderMode renderMode = base::MeshRenderMode::RM_LINES );
 
  protected:
  private:
};
 
class RA_ENGINE_API Mesh : public IndexedGeometry<Core::Geometry::TriangleMesh>
{
    using base = IndexedGeometry<Core::Geometry::TriangleMesh>;
 
  public:
    using base::IndexedGeometry;
    size_t getNumFaces() const override;
 
    using base::loadGeometry;
    [[deprecated]] void loadGeometry( const Core::Vector3Array& vertices,
                                      const std::vector<uint>& indices );
 
  protected:
  private:
};
 
template <typename T>
class RA_ENGINE_API GeneralMesh : public IndexedGeometry<T>
{
    using base      = IndexedGeometry<T>;
    using IndexType = Core::Vector3ui;
 
  public:
    using base::IndexedGeometry;
    inline size_t getNumFaces() const override;
 
  protected:
    inline void updateGL_specific_impl() override;
 
  private:
    inline void triangulate();
    Core::AlignedStdVector<IndexType> m_triangleIndices;
};
 
using PolyMesh = GeneralMesh<Core::Geometry::PolyMesh>;
using QuadMesh = GeneralMesh<Core::Geometry::QuadMesh>;
 
template <typename CoreMeshType>
CoreMeshType createCoreMeshFromGeometryData( const Ra::Core::Asset::GeometryData* data ) {
    CoreMeshType mesh;
    typename CoreMeshType::IndexContainerType indices;
 
    if ( !data->isLineMesh() ) {
        auto& geo = data->getGeometry();
        const auto& [layerKeyType, layerBase] =
            geo.getFirstLayerOccurrence( mesh.getLayerKey().first );
        const auto& layer = static_cast<
            const Core::Geometry::GeometryIndexLayer<typename CoreMeshType::IndexType>&>(
            layerBase );
        const auto& faces = layer.collection();
        indices.reserve( faces.size() );
        std::copy( faces.begin(), faces.end(), std::back_inserter( indices ) );
    }
#if 0
    // TODO manage line meshes in a "usual" way, i.e. as an indexed geometry with specific
    //  rendering properties (i.e. shader, as it is the case for point clouds)
    // Create a degenerated triangle to handle edges case.
    else {
        const auto& edges = ... access the LineIndexLayer
        indices.reserve( edges.size() );
        std::transform(
            edges.begin(), edges.end(), std::back_inserter( indices ), []( Ra::Core::Vector2ui v ) {
                return ( Ra::Core::Vector3ui { v( 0 ), v( 1 ), v( 1 ) } );
            } );
    }
#endif
 
    mesh.setIndices( std::move( indices ) );
 
    // This copy only "usual" attributes. See Core::Geometry::AttribManager::copyAllAttributes
    mesh.vertexAttribs().copyAllAttributes( data->getGeometry().vertexAttribs() );
 
    return mesh;
}
 
namespace RenderMeshType {
template <class CoreMeshT>
struct getType {};
 
template <>
struct getType<Ra::Core::Geometry::LineMesh> {
    using Type = Ra::Engine::Data::LineMesh;
};
 
template <>
struct getType<Ra::Core::Geometry::TriangleMesh> {
    using Type = Ra::Engine::Data::Mesh;
};
 
template <>
struct getType<Ra::Core::Geometry::QuadMesh> {
    using Type = Ra::Engine::Data::QuadMesh;
};
 
template <>
struct getType<Ra::Core::Geometry::PolyMesh> {
    using Type = Ra::Engine::Data::PolyMesh;
};
} // namespace RenderMeshType
 
template <typename CoreMeshType>
typename RenderMeshType::getType<CoreMeshType>::Type*
createMeshFromGeometryData( const std::string& name, const Ra::Core::Asset::GeometryData* data ) {
    using MeshType = typename RenderMeshType::getType<CoreMeshType>::Type;
 
    auto mesh = createCoreMeshFromGeometryData<CoreMeshType>( data );
 
    MeshType* ret = new MeshType { name };
    ret->loadGeometry( std::move( mesh ) );
 
    return ret;
}
 
 
void AttribArrayDisplayable::setRenderMode( MeshRenderMode mode ) {
    m_renderMode = mode;
    updatePickingRenderMode();
}
 
AttribArrayDisplayable::MeshRenderMode AttribArrayDisplayable::getRenderMode() const {
    return m_renderMode;
}
 
 
void VaoIndices::setIndicesDirty() {
    m_indicesDirty = true;
}
 
 
template <typename I>
template <typename T>
void IndexedAttribArrayDisplayable<I>::addAttrib(
    const std::string& name,
    const typename Ra::Core::Utils::Attrib<T>::Container& data ) {
    auto handle = m_attribManager.addAttrib<T>( name );
    m_attribManager.getAttrib( handle ).setData( data );
    m_handleToBuffer[name] = m_dataDirty.size();
    m_dataDirty.push_back( true );
    m_vbos.emplace_back( nullptr );
    m_isDirty = true;
}
 
template <typename I>
template <typename T>
void IndexedAttribArrayDisplayable<I>::addAttrib(
    const std::string& name,
    const typename Ra::Core ::Utils::Attrib<T>::Container&& data ) {
    auto handle = m_attribManager.addAttrib<T>( name );
    m_attribManager.getAttrib( handle ).setData( std::move( data ) );
    m_handleToBuffer[name] = m_dataDirty.size();
    m_dataDirty.push_back( true );
    m_vbos.emplace_back( nullptr );
    m_isDirty = true;
}
 
template <typename I>
void IndexedAttribArrayDisplayable<I>::updateGL() {
    if ( m_isDirty ) {
        // Check that our dirty bits are consistent.
        ON_ASSERT( bool dirtyTest = false;
                   for ( const auto& d : m_dataDirty ) { dirtyTest = dirtyTest || d; } );
        CORE_ASSERT( dirtyTest == m_isDirty, "Dirty flags inconsistency" );
 
        if ( !m_indices ) {
            m_indices      = globjects::Buffer::create();
            m_indicesDirty = true;
        }
        if ( m_indicesDirty ) {
            m_indices->setData(
                static_cast<gl::GLsizeiptr>( m_cpu_indices.size() * sizeof( IndexType ) ),
                m_cpu_indices.data(),
                GL_STATIC_DRAW );
            m_indicesDirty = false;
        }
 
        m_numElements = m_cpu_indices.size();
 
        if ( !m_vao ) { m_vao = globjects::VertexArray::create(); }
        m_vao->bind();
        m_vao->bindElementBuffer( m_indices.get() );
        m_vao->unbind();
 
        auto func = [this]( Ra::Core::Utils::AttribBase* b ) {
            auto idx = m_handleToBuffer[b->getName()];
 
            if ( m_dataDirty[idx] ) {
                if ( !m_vbos[idx] ) { m_vbos[idx] = globjects::Buffer::create(); }
                m_vbos[idx]->setData( b->getBufferSize(), b->dataPtr(), GL_DYNAMIC_DRAW );
                m_dataDirty[idx] = false;
            }
        };
        m_attribManager.for_each_attrib( func );
        GL_CHECK_ERROR;
        m_isDirty = false;
    }
}
 
template <typename I>
void IndexedAttribArrayDisplayable<I>::autoVertexAttribPointer( const ShaderProgram* prog ) {
 
    auto glprog           = prog->getProgramObject();
    gl::GLint attribCount = glprog->get( GL_ACTIVE_ATTRIBUTES );
 
    for ( GLint idx = 0; idx < attribCount; ++idx ) {
        const gl::GLsizei bufSize = 256;
        gl::GLchar name[bufSize];
        gl::GLsizei length;
        gl::GLint size;
        gl::GLenum type;
        glprog->getActiveAttrib( idx, bufSize, &length, &size, &type, name );
        auto loc = glprog->getAttributeLocation( name );
 
        auto attribName = name; // m_translationTableShaderToMesh[name];
        auto attrib     = m_attribManager.getAttribBase( attribName );
 
        if ( attrib && attrib->getSize() > 0 ) {
            m_vao->enable( loc );
            auto binding = m_vao->binding( idx );
            binding->setAttribute( loc );
            CORE_ASSERT( m_vbos[m_handleToBuffer[attribName]].get(), "vbo is nullptr" );
#ifdef CORE_USE_DOUBLE
            binding->setBuffer( m_vbos[m_handleToBuffer[attribName]].get(),
                                0,
                                attrib->getNumberOfComponents() * sizeof( float ) );
#else
 
            binding->setBuffer(
                m_vbos[m_handleToBuffer[attribName]].get(), 0, attrib->getStride() );
#endif
            binding->setFormat( attrib->getNumberOfComponents(), GL_SCALAR );
        }
        else { m_vao->disable( loc ); }
    }
}
 
template <typename I>
void IndexedAttribArrayDisplayable<I>::render( const ShaderProgram* prog ) {
    if ( m_vao ) {
        autoVertexAttribPointer( prog );
        m_vao->bind();
        m_vao->drawElements( static_cast<GLenum>( m_renderMode ),
                             GLsizei( m_numElements ),
                             GL_UNSIGNED_INT,
                             nullptr );
        m_vao->unbind();
    }
}
 
 
template <typename CoreGeometry>
CoreGeometryDisplayable<CoreGeometry>::CoreGeometryDisplayable( const std::string& name,
                                                                MeshRenderMode renderMode ) :
    base( name, renderMode ) {
    setupCoreMeshObservers();
}
 
template <typename CoreGeometry>
const Ra::Core::Geometry::AbstractGeometry&
CoreGeometryDisplayable<CoreGeometry>::getAbstractGeometry() const {
    return m_mesh;
}
 
template <typename CoreGeometry>
Ra::Core::Geometry::AbstractGeometry& CoreGeometryDisplayable<CoreGeometry>::getAbstractGeometry() {
    return m_mesh;
}
 
template <typename CoreGeometry>
const Ra::Core::Geometry::AttribArrayGeometry&
CoreGeometryDisplayable<CoreGeometry>::getAttribArrayGeometry() const {
    return m_mesh;
}
 
template <typename CoreGeometry>
Ra::Core::Geometry::AttribArrayGeometry&
CoreGeometryDisplayable<CoreGeometry>::getAttribArrayGeometry() {
    return m_mesh;
}
 
template <typename CoreGeometry>
const CoreGeometry& CoreGeometryDisplayable<CoreGeometry>::getCoreGeometry() const {
    return m_mesh;
}
 
template <typename CoreGeometry>
CoreGeometry& CoreGeometryDisplayable<CoreGeometry>::getCoreGeometry() {
    return m_mesh;
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::addToTranslationTable( const std::string& name ) {
    m_translationTable.insert( name, name );
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::addAttribObserver( const std::string& name ) {
    // this observer is called each time an attrib is added or removed from m_mesh
    auto attrib = m_mesh.getAttribBase( name );
    // if attrib not nullptr, then it's an attrib add, so attach an observer to it
 
    if ( attrib ) {
        auto itr = m_handleToBuffer.find( name );
        if ( itr == m_handleToBuffer.end() ) {
            m_handleToBuffer[name] = m_dataDirty.size();
 
            addToTranslationTable( name );
 
            m_dataDirty.push_back( true );
            m_vbos.emplace_back( nullptr );
        }
        auto idx = m_handleToBuffer[name];
        attrib->attach( AttribObserver( this, idx ) );
    }
    // else it's an attrib remove, do nothing, cleanup will be done in updateGL()
    else {}
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::autoVertexAttribPointer( const ShaderProgram* prog ) {
 
    auto glprog           = prog->getProgramObject();
    gl::GLint attribCount = glprog->get( GL_ACTIVE_ATTRIBUTES );
 
    for ( GLint idx = 0; idx < attribCount; ++idx ) {
        const gl::GLsizei bufSize = 256;
        gl::GLchar name[bufSize];
        gl::GLsizei length;
        gl::GLint size;
        gl::GLenum type;
        glprog->getActiveAttrib( idx, bufSize, &length, &size, &type, name );
        auto loc = glprog->getAttributeLocation( name );
 
        auto attribNameOpt = m_translationTable.keyIfExists( name );
        if ( attribNameOpt ) {
            auto attribName = *attribNameOpt;
            auto attrib     = m_mesh.getAttribBase( attribName );
            if ( attrib && attrib->getSize() > 0 ) {
                m_vao->enable( loc );
                auto binding = m_vao->binding( idx );
                binding->setAttribute( loc );
                CORE_ASSERT( m_vbos[m_handleToBuffer[attribName]].get(), "vbo is nullptr" );
#ifdef CORE_USE_DOUBLE
                binding->setBuffer( m_vbos[m_handleToBuffer[attribName]].get(),
                                    0,
                                    attrib->getNumberOfComponents() * sizeof( float ) );
#else
 
                binding->setBuffer(
                    m_vbos[m_handleToBuffer[attribName]].get(), 0, attrib->getStride() );
#endif
                binding->setFormat( attrib->getNumberOfComponents(), GL_SCALAR );
            }
            else { m_vao->disable( loc ); }
        }
        else { m_vao->disable( loc ); }
    }
}
 
template <typename T>
void CoreGeometryDisplayable<T>::loadGeometry_common( T&& mesh ) {
    m_mesh = std::move( mesh );
    setupCoreMeshObservers();
}
 
template <typename T>
 
void CoreGeometryDisplayable<T>::setupCoreMeshObservers() {
    int idx = 0;
    m_dataDirty.resize( m_mesh.vertexAttribs().getNumAttribs() );
    m_vbos.resize( m_mesh.vertexAttribs().getNumAttribs() );
    // here capture ref to idx to propagate idx incrementation
    m_mesh.vertexAttribs().for_each_attrib( [&idx, this]( Ra::Core::Utils::AttribBase* b ) {
        auto name              = b->getName();
        m_handleToBuffer[name] = idx;
        m_dataDirty[idx]       = true;
 
        // create a identity translation if name is not already translated.
        addToTranslationTable( name );
 
        b->attach( AttribObserver( this, idx ) );
        ++idx;
    } );
 
    // add an observer on attrib manipulation.
    m_mesh.vertexAttribs().attachMember(
        this, &CoreGeometryDisplayable<CoreGeometry>::addAttribObserver );
    m_isDirty = true;
}
 
template <typename CoreGeometry>
template <typename A>
Ra::Core::Utils::AttribHandle<A>
CoreGeometryDisplayable<CoreGeometry>::addAttrib( const std::string& name,
                                                  const typename Core::VectorArray<A>& data ) {
    return m_mesh.addAttrib( name, data );
}
 
template <typename CoreGeometry>
size_t CoreGeometryDisplayable<CoreGeometry>::getNumVertices() const {
    return m_mesh.vertices().size();
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::loadGeometry( CoreGeometry&& /*mesh*/ ) {
    CORE_ASSERT( false, "must be specialized" );
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::updateGL() {
    if ( m_isDirty ) {
        // Check that our dirty bits are consistent.
        ON_ASSERT( bool dirtyTest = false;
                   for ( auto d : m_dataDirty ) { dirtyTest = dirtyTest || d; } );
        CORE_ASSERT( dirtyTest == m_isDirty, "Dirty flags inconsistency" );
        CORE_ASSERT( !( m_mesh.vertices().empty() ), "No vertex." );
 
        updateGL_specific_impl();
#ifdef CORE_USE_DOUBLE
        // need convserion
        auto func = [this]( Ra::Core::Utils::AttribBase* b ) {
            auto idx = m_handleToBuffer[b->getName()];
 
            if ( m_dataDirty[idx] ) {
                if ( !m_vbos[idx] ) { m_vbos[idx] = globjects::Buffer::create(); }
 
                auto stride      = b->getStride();
                auto eltSize     = b->getNumberOfComponents();
                auto size        = b->getSize();
                auto data        = std::make_unique<float[]>( size * eltSize );
                const void* ptr  = b->dataPtr();
                const char* cptr = reinterpret_cast<const char*>( ptr );
 
                for ( size_t i = 0; i < size; i++ ) {
                    auto tptr = reinterpret_cast<const Scalar*>( cptr + i * stride );
                    for ( size_t j = 0; j < eltSize; ++j ) {
                        data[i * eltSize + j] = tptr[j];
                    }
                }
 
                m_vbos[idx]->setData(
                    size * eltSize * sizeof( float ), data.get(), GL_DYNAMIC_DRAW );
 
                m_dataDirty[idx] = false;
            }
        };
 
#else
        auto func = [this]( Ra::Core::Utils::AttribBase* b ) {
            auto idx = m_handleToBuffer[b->getName()];
 
            if ( m_dataDirty[idx] ) {
                if ( !m_vbos[idx] ) { m_vbos[idx] = globjects::Buffer::create(); }
                m_vbos[idx]->setData( b->getBufferSize(), b->dataPtr(), GL_DYNAMIC_DRAW );
                m_dataDirty[idx] = false;
            }
        };
#endif
        m_mesh.vertexAttribs().for_each_attrib( func );
 
        // cleanup removed attrib
        for ( auto buffer : m_handleToBuffer ) {
            // do not remove name from handleToBuffer to keep index ...
            // we could also update handleToBuffer, m_vbos, m_dataDirty
            if ( !m_mesh.hasAttrib( buffer.first ) && m_vbos[buffer.second] ) {
                m_vbos[buffer.second].reset( nullptr );
                m_dataDirty[buffer.second] = false;
            }
        }
 
        GL_CHECK_ERROR;
        m_isDirty = false;
    }
}
 
template <typename CoreGeometry>
void CoreGeometryDisplayable<CoreGeometry>::setAttribNameCorrespondance(
    const std::string& meshAttribName,
    const std::string& shaderAttribName ) {
 
    m_translationTable.replace( meshAttribName, shaderAttribName );
}
 
 
template <typename T>
IndexedGeometry<T>::IndexedGeometry( const std::string& name,
                                     typename base::CoreGeometry&& geom,
                                     typename base::MeshRenderMode renderMode ) :
    base( name, renderMode ) {
    loadGeometry( std::move( geom ) );
}
 
template <typename T>
void IndexedGeometry<T>::loadGeometry( T&& mesh ) {
    setIndicesDirty();
    base::loadGeometry_common( std::move( mesh ) );
 
    // indices
    base::m_mesh.attach( IndicesObserver( this ) );
}
 
template <typename T>
void IndexedGeometry<T>::updateGL_specific_impl() {
    if ( !m_indices ) {
        m_indices      = globjects::Buffer::create();
        m_indicesDirty = true;
    }
    if ( m_indicesDirty ) {
        // m_indices->setData( m_mesh.m_indices, GL_DYNAMIC_DRAW );
        m_numElements =
            base::m_mesh.getIndices().size() * base::CoreGeometry::IndexType::RowsAtCompileTime;
 
        m_indices->setData(
            static_cast<gl::GLsizeiptr>( base::m_mesh.getIndices().size() *
                                         sizeof( typename base::CoreGeometry::IndexType ) ),
            base::m_mesh.getIndices().data(),
            GL_STATIC_DRAW );
        m_indicesDirty = false;
    }
    if ( !base::m_vao ) { base::m_vao = globjects::VertexArray::create(); }
    base::m_vao->bind();
    base::m_vao->bindElementBuffer( m_indices.get() );
    base::m_vao->unbind();
}
 
template <typename T>
void IndexedGeometry<T>::render( const ShaderProgram* prog ) {
    if ( base::m_vao ) {
        GL_CHECK_ERROR;
        base::m_vao->bind();
        base::autoVertexAttribPointer( prog );
        GL_CHECK_ERROR;
        base::m_vao->drawElements( static_cast<GLenum>( base::m_renderMode ),
                                   GLsizei( m_numElements ),
                                   GL_UNSIGNED_INT,
                                   nullptr );
        GL_CHECK_ERROR;
        base::m_vao->unbind();
        GL_CHECK_ERROR;
    }
}
 
 
template <typename T>
MultiIndexedGeometry<T>::MultiIndexedGeometry( const std::string& name,
                                               typename base::CoreGeometry&& geom,
                                               typename base::MeshRenderMode renderMode ) :
    base( name, renderMode ) {
    loadGeometry( std::move( geom ) );
}
 
template <typename T>
void MultiIndexedGeometry<T>::loadGeometry( T&& mesh ) {
    m_indices.clear();
 
    base::loadGeometry_common( std::move( mesh ) );
    CORE_ASSERT( false, "not implemented yet" );
 
    // indices
    //  base::m_mesh.attach( IndicesObserver( this ) );
}
 
template <typename T>
void MultiIndexedGeometry<T>::updateGL_specific_impl() {
    CORE_ASSERT( false, "not implemented yet" );
    //    if ( !m_indices )
    //    {
    //        m_indices      = globjects::Buffer::create();
    //        m_indicesDirty = true;
    //    }
    //    if ( m_indicesDirty )
    //    {
    //        /// this one do not work since m_indices is not a std::vector
    //        // m_indices->setData( m_mesh.m_indices, GL_DYNAMIC_DRAW );
    //        m_numElements =
    //            base::m_mesh.getIndices().size() *
    //            base::CoreGeometry::IndexType::RowsAtCompileTime;
    //
    //        m_indices->setData(
    //            static_cast<gl::GLsizeiptr>( base::m_mesh.getIndices().size() *
    //                                         sizeof( typename base::CoreGeometry::IndexType ) ),
    //            base::m_mesh.getIndices().data(),
    //            GL_STATIC_DRAW );
    //        m_indicesDirty = false;
    //    }
    //    if ( !base::m_vao ) { base::m_vao = globjects::VertexArray::create(); }
    //    base::m_vao->bind();
    //    base::m_vao->bindElementBuffer( m_indices.get() );
    //    base::m_vao->unbind();
}
 
template <typename T>
void MultiIndexedGeometry<T>::render( const ShaderProgram* ) {
    CORE_ASSERT( false, "not implemented yet" );
    //    if ( base::m_vao )
    //    {
    //        GL_CHECK_ERROR;
    //        base::m_vao->bind();
    //        base::autoVertexAttribPointer( prog );
    //        GL_CHECK_ERROR;
    //        base::m_vao->drawElements( static_cast<GLenum>( base::m_renderMode ),
    //                                   GLsizei( m_numElements ),
    //                                   GL_UNSIGNED_INT,
    //                                   nullptr );
    //        GL_CHECK_ERROR;
    //        base::m_vao->unbind();
    //        GL_CHECK_ERROR;
    //    }
}
 
 
PointCloud::PointCloud( const std::string& name,
                        typename base::CoreGeometry&& geom,
                        typename base::MeshRenderMode renderMode ) :
    base( name, renderMode ) {
    loadGeometry( std::move( geom ) );
}
 
PointCloud::PointCloud( const std::string& name, typename base::MeshRenderMode renderMode ) :
    base( name, renderMode ) {}
 
 
LineMesh::LineMesh( const std::string& name,
                    typename base::CoreGeometry&& geom,
                    typename base::MeshRenderMode renderMode ) :
    base( name, std::move( geom ), renderMode ) {}
 
LineMesh::LineMesh( const std::string& name, typename base::MeshRenderMode renderMode ) :
    base( name, renderMode ) {}
 
 
template <typename T>
size_t GeneralMesh<T>::getNumFaces() const {
    return this->getCoreGeometry().getIndices().size();
}
 
template <typename T>
void GeneralMesh<T>::updateGL_specific_impl() {
    if ( !this->m_indices ) {
        this->m_indices      = globjects::Buffer::create();
        this->m_indicesDirty = true;
    }
    if ( this->m_indicesDirty ) {
        triangulate();
        // m_indices->setData( m_mesh.m_indices, GL_DYNAMIC_DRAW );
        this->m_numElements = m_triangleIndices.size() * GeneralMesh::IndexType::RowsAtCompileTime;
 
        this->m_indices->setData( static_cast<gl::GLsizeiptr>( m_triangleIndices.size() *
                                                               sizeof( GeneralMesh::IndexType ) ),
                                  m_triangleIndices.data(),
                                  GL_STATIC_DRAW );
        this->m_indicesDirty = false;
    }
    if ( !base::m_vao ) { base::m_vao = globjects::VertexArray::create(); }
    base::m_vao->bind();
    base::m_vao->bindElementBuffer( this->m_indices.get() );
    base::m_vao->unbind();
}
 
template <typename T>
void GeneralMesh<T>::triangulate() {
    m_triangleIndices.clear();
    m_triangleIndices.reserve( this->m_mesh.getIndices().size() );
    for ( const auto& face : this->m_mesh.getIndices() ) {
        if ( face.size() == 3 ) { m_triangleIndices.push_back( face ); }
        else {
            int minus { int( face.size() ) - 1 };
            int plus { 0 };
            while ( plus + 1 < minus ) {
                if ( ( plus - minus ) % 2 ) {
                    m_triangleIndices.emplace_back( face[plus], face[plus + 1], face[minus] );
                    ++plus;
                }
                else {
                    m_triangleIndices.emplace_back( face[minus], face[plus], face[minus - 1] );
                    --minus;
                }
            }
        }
    }
}
 
template <>
inline void GeneralMesh<Core::Geometry::QuadMesh>::triangulate() {
    m_triangleIndices.clear();
    m_triangleIndices.reserve( 2 * this->m_mesh.getIndices().size() );
    // assume quads are convex
    for ( const auto& face : this->m_mesh.getIndices() ) {
        m_triangleIndices.emplace_back( face[0], face[1], face[2] );
        m_triangleIndices.emplace_back( face[0], face[2], face[3] );
    }
}
 
} // namespace Ra
} // namespace Engine
} // namespace Ra