/*
Crafter®.Asset
Copyright (C) 2026 Catcrafts®
catcrafts.net

This library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License version 3.0 as published by the Free Software Foundation;

This library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.

You should have received a copy of the GNU Lesser General Public
License along with this library; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301 USA
*/

export module Crafter.Asset:Mesh;
import :Compression;
import Crafter.Math;
import std;
namespace fs = std::filesystem;

export namespace Crafter {
    // Minimal Wavefront .mtl record. Only diffuse-map path is captured;
    // normal/specular/etc. would extend this in a backward-compatible way.
    // `mapKd` is verbatim from the file (relative to the .mtl's directory)
    // and empty if the material defined no diffuse texture.
    struct MtlMaterial {
        std::string mapKd;
    };

    // Parse a .mtl file into { materialName → MtlMaterial }. Strips trailing
    // \r/\n/spaces on names + paths so DOS-line-ending files match the
    // `usemtl` tokens in their sibling .obj. Materials without `map_Kd`
    // are still present in the map with an empty `mapKd`.
    inline std::unordered_map<std::string, MtlMaterial> LoadMTL(fs::path path) {
        std::ifstream file(path);
        std::unordered_map<std::string, MtlMaterial> result;
        std::string current;
        std::string line;
        auto trim = [](std::string& s) {
            while (!s.empty() && (s.back() == '\r' || s.back() == '\n' || s.back() == ' ' || s.back() == '\t')) s.pop_back();
            while (!s.empty() && (s.front() == ' ' || s.front() == '\t')) s.erase(s.begin());
        };
        while (std::getline(file, line)) {
            if (line.starts_with("newmtl ")) {
                current = line.substr(7);
                trim(current);
                result.try_emplace(current);
            } else if (line.starts_with("map_Kd ") && !current.empty()) {
                std::string val = line.substr(7);
                trim(val);
                result[current].mapKd = std::move(val);
            }
        }
        return result;
    }

    struct __attribute__((packed)) VertexNormalTangentUVPacked {
        Vector<float, 3, 4> normal;
        Vector<float, 3, 4> tangent;
        Vector<float, 2, 4> uv;
    };

    struct VertexNormalTangentUV {
        Vector<float, 3, 0> normal;
        Vector<float, 3, 0> tangent;
        Vector<float, 2, 0> uv;
    };

    // GDeflate-compressed counterpart of MeshAsset<T>::Save output. Three
    // regions: [vertex, index, data]. dataCount==0 leaves the data region
    // empty (zero compressedSize/decompressedSize). dataStride records sizeof(T)
    // at compress time so consumers can validate.
    struct CompressedMeshAsset {
        std::uint32_t vertexCount = 0;
        std::uint32_t indexCount = 0;
        std::uint32_t dataCount = 0;
        std::uint32_t dataStride = 0;
        Compression::CompressedBlob blob;
    };

    namespace MeshAssetFormat {
        inline constexpr char magic[4] = {'C', 'G', 'D', 'M'};
        inline constexpr std::uint32_t version = 1;
    }

    inline CompressedMeshAsset LoadCompressedMesh(fs::path path) {
        std::ifstream file(path, std::ios::binary);
        char magic[4];
        file.read(magic, 4);
        if (std::memcmp(magic, MeshAssetFormat::magic, 4) != 0) {
            Compression::Fatal("LoadCompressedMesh: bad magic on " + path.string());
        }
        std::uint32_t version = 0;
        file.read(reinterpret_cast<char*>(&version), sizeof(version));
        if (version != MeshAssetFormat::version) {
            Compression::Fatal("LoadCompressedMesh: unsupported version on " + path.string());
        }
        CompressedMeshAsset out;
        file.read(reinterpret_cast<char*>(&out.vertexCount), sizeof(out.vertexCount));
        file.read(reinterpret_cast<char*>(&out.indexCount), sizeof(out.indexCount));
        file.read(reinterpret_cast<char*>(&out.dataCount), sizeof(out.dataCount));
        file.read(reinterpret_cast<char*>(&out.dataStride), sizeof(out.dataStride));
        out.blob = Compression::ReadBlob(file);
        return out;
    }

    template <typename T>
    struct MeshAsset {
        std::vector<Vector<float, 3, 3>> vertexes;
        std::vector<std::uint32_t> indexes;
        std::vector<T> datas;

        void Save(fs::path path) {
            std::ofstream file(path, std::ios::binary);

            std::uint32_t vertexCount = vertexes.size();
            std::uint32_t indexCount = indexes.size();
            std::uint32_t dataCount = datas.size();

            file.write(reinterpret_cast<char*>(&vertexCount), sizeof(vertexCount));
            file.write(reinterpret_cast<char*>(&indexCount), sizeof(indexCount));
            file.write(reinterpret_cast<char*>(&dataCount), sizeof(dataCount));

            file.write(reinterpret_cast<char*>(vertexes.data()), vertexCount * sizeof(Vector<float, 3, 3>));
            file.write(reinterpret_cast<char*>(indexes.data()), indexCount * sizeof(std::uint32_t));
            file.write(reinterpret_cast<char*>(datas.data()), dataCount * sizeof(T));
        }

        void SaveCompressed(fs::path path) const {
            std::array<std::span<const std::byte>, 3> streams = {
                std::as_bytes(std::span(vertexes)),
                std::as_bytes(std::span(indexes)),
                std::as_bytes(std::span(datas)),
            };
            Compression::CompressedBlob blob = Compression::CompressStreams(streams);

            std::ofstream file(path, std::ios::binary);
            file.write(MeshAssetFormat::magic, 4);
            std::uint32_t version = MeshAssetFormat::version;
            std::uint32_t vc = static_cast<std::uint32_t>(vertexes.size());
            std::uint32_t ic = static_cast<std::uint32_t>(indexes.size());
            std::uint32_t dc = static_cast<std::uint32_t>(datas.size());
            std::uint32_t stride = static_cast<std::uint32_t>(sizeof(T));
            file.write(reinterpret_cast<const char*>(&version), sizeof(version));
            file.write(reinterpret_cast<const char*>(&vc), sizeof(vc));
            file.write(reinterpret_cast<const char*>(&ic), sizeof(ic));
            file.write(reinterpret_cast<const char*>(&dc), sizeof(dc));
            file.write(reinterpret_cast<const char*>(&stride), sizeof(stride));
            Compression::WriteBlob(file, blob);
        }

        static MeshAsset<T> Load(fs::path path) {
            MeshAsset<T> mesh;
            
            std::ifstream file(path, std::ios::binary);

            std::uint32_t vertexCount = 0;
            std::uint32_t indexCount  = 0;
            std::uint32_t dataCount   = 0;

            file.read(reinterpret_cast<char*>(&vertexCount), sizeof(vertexCount));
            file.read(reinterpret_cast<char*>(&indexCount),  sizeof(indexCount));
            file.read(reinterpret_cast<char*>(&dataCount),   sizeof(dataCount));

            mesh.vertexes.resize(vertexCount);
            mesh.indexes.resize(indexCount);
            mesh.datas.resize(dataCount);

            file.read(reinterpret_cast<char*>(mesh.vertexes.data()), vertexCount * sizeof(Vector<float, 3, 3>));
            file.read(reinterpret_cast<char*>(mesh.indexes.data()), indexCount * sizeof(std::uint32_t));
            file.read(reinterpret_cast<char*>(mesh.datas.data()), dataCount * sizeof(T));

            return mesh;
        }

        // Parse a .obj split by `usemtl` directives. Returns one
        // MeshAsset<T> per material group (in encounter order), each with
        // its own deduplicated vertex/index/data arrays — i.e. shared
        // vertices across materials are duplicated so each MeshAsset is
        // self-contained and can become a standalone BLAS. Dedup is
        // O(1)-per-vertex via a hash map keyed on (vIdx, vtIdx, vnIdx),
        // unlike LoadOBJ's linear-scan dedup (which is O(N²) and gets
        // pathological on Sponza-class inputs).
        //
        // Faces appearing before any `usemtl` accumulate under the empty
        // string material name; consumers can decide whether to drop them
        // or treat them as a default. Non-triangle faces are fan-triangulated
        // around their first vertex.
        static std::vector<std::pair<std::string, MeshAsset<T>>> LoadOBJSplit(fs::path path)
            requires (std::same_as<T, VertexNormalTangentUVPacked> || std::same_as<T, VertexNormalTangentUV>) {
            std::ifstream file(path);

            std::vector<Vector<float, 3, 0>> positions;
            std::vector<Vector<float, 3, 0>> normals;
            std::vector<Vector<float, 2, 0>> uvs;

            std::vector<std::pair<std::string, MeshAsset<T>>> result;
            std::unordered_map<std::string, std::size_t> matIndex;
            std::vector<std::unordered_map<std::uint64_t, std::uint32_t>> dedupPerMat;

            std::size_t curMat = static_cast<std::size_t>(-1);

            auto useMaterial = [&](std::string name) {
                while (!name.empty() && (name.back() == '\r' || name.back() == '\n' || name.back() == ' ' || name.back() == '\t')) name.pop_back();
                auto it = matIndex.find(name);
                if (it != matIndex.end()) { curMat = it->second; return; }
                matIndex.emplace(name, result.size());
                result.emplace_back(std::move(name), MeshAsset<T>{});
                dedupPerMat.emplace_back();
                curMat = result.size() - 1;
            };

            std::string line;
            while (std::getline(file, line)) {
                if (line.starts_with("usemtl ")) {
                    useMaterial(line.substr(7));
                } else if (line.starts_with("vt ")) {
                    std::istringstream iss(line.substr(3));
                    Vector<float, 2, 0> uv;
                    iss >> uv.x >> uv.y;
                    uvs.push_back(uv);
                } else if (line.starts_with("vn ")) {
                    std::istringstream iss(line.substr(3));
                    Vector<float, 3, 0> n;
                    iss >> n.x >> n.y >> n.z;
                    normals.push_back(n);
                } else if (line.starts_with("v ")) {
                    std::istringstream iss(line.substr(2));
                    Vector<float, 3, 3> p;
                    iss >> p.x >> p.y >> p.z;
                    positions.push_back(Vector<float, 3, 0>{p.x, p.y, p.z});
                } else if (line.starts_with("f ")) {
                    if (curMat == static_cast<std::size_t>(-1)) useMaterial(std::string{});
                    MeshAsset<T>& mesh = result[curMat].second;
                    auto& dedup = dedupPerMat[curMat];
                    std::istringstream iss(line.substr(2));
                    std::string vg;
                    std::vector<std::uint32_t> faceIndices;
                    while (iss >> vg) {
                        std::istringstream vss(vg);
                        std::string vs, vts, vns;
                        std::getline(vss, vs, '/');
                        std::getline(vss, vts, '/');
                        std::getline(vss, vns, '/');
                        std::uint32_t vi  = vs.empty()  ? 0u : static_cast<std::uint32_t>(std::stoi(vs) - 1);
                        std::uint32_t vti = vts.empty() ? 0u : static_cast<std::uint32_t>(std::stoi(vts) - 1);
                        std::uint32_t vni = vns.empty() ? 0u : static_cast<std::uint32_t>(std::stoi(vns) - 1);
                        std::uint64_t key = (static_cast<std::uint64_t>(vi)  << 42)
                                          | (static_cast<std::uint64_t>(vti) << 21)
                                          |  static_cast<std::uint64_t>(vni);
                        if (auto it = dedup.find(key); it != dedup.end()) {
                            faceIndices.push_back(it->second);
                        } else {
                            std::uint32_t newIdx = static_cast<std::uint32_t>(mesh.vertexes.size());
                            const Vector<float, 3, 0>& p  = positions[vi];
                            const Vector<float, 3, 0>& n  = normals[vni];
                            const Vector<float, 2, 0>& uv = uvs[vti];
                            mesh.vertexes.push_back(Vector<float, 3, 3>{p.x, p.y, p.z});
                            mesh.datas.push_back(T{n, {0,0,0}, uv});
                            dedup.emplace(key, newIdx);
                            faceIndices.push_back(newIdx);
                        }
                    }
                    for (std::size_t i = 1; i + 1 < faceIndices.size(); ++i) {
                        mesh.indexes.push_back(faceIndices[0]);
                        mesh.indexes.push_back(faceIndices[i]);
                        mesh.indexes.push_back(faceIndices[i + 1]);
                    }
                }
            }

            // Accumulate face tangents into each vertex's tangent slot,
            // then normalize. Mirrors LoadOBJ's tangent calculation but
            // guarded against degenerate UVs (the LoadOBJ form NaNs out
            // silently — fine for the cube but blows up some Sponza tris).
            for (auto& entry : result) {
                MeshAsset<T>& mesh = entry.second;
                for (std::uint32_t i = 0; i + 2 < mesh.indexes.size(); i += 3) {
                    std::uint32_t i0 = mesh.indexes[i];
                    std::uint32_t i1 = mesh.indexes[i + 1];
                    std::uint32_t i2 = mesh.indexes[i + 2];
                    Vector<float, 3, 0> edge1 = mesh.vertexes[i1] - mesh.vertexes[i0];
                    Vector<float, 3, 0> edge2 = mesh.vertexes[i2] - mesh.vertexes[i0];
                    Vector<float, 2, 0> dUV1 = mesh.datas[i1].uv - mesh.datas[i0].uv;
                    Vector<float, 2, 0> dUV2 = mesh.datas[i2].uv - mesh.datas[i0].uv;
                    float denom = dUV1.x * dUV2.y - dUV1.y * dUV2.x;
                    if (denom == 0.0f) continue;
                    float f = 1.0f / denom;
                    Vector<float, 3, 0> tangent;
                    tangent.x = f * (dUV2.y * edge1.x - dUV1.y * edge2.x);
                    tangent.y = f * (dUV2.y * edge1.y - dUV1.y * edge2.y);
                    tangent.z = f * (dUV2.y * edge1.z - dUV1.y * edge2.z);
                    tangent.Normalize();
                    mesh.datas[i0].tangent += tangent;
                    mesh.datas[i1].tangent += tangent;
                    mesh.datas[i2].tangent += tangent;
                }
                for (T& v : mesh.datas) {
                    v.tangent.Normalize();
                }
            }

            return result;
        }

        static MeshAsset<T> LoadOBJ(fs::path path) requires (std::same_as<T, VertexNormalTangentUVPacked> || std::same_as<T, VertexNormalTangentUV>) {
            std::ifstream file(path);

            MeshAsset<T> mesh;
                
            std::string line;
        
            std::vector<Vector<float, 3, 0>> positions;
            std::vector<Vector<float, 3, 0>> normals;
            std::vector<Vector<float, 2, 0>> uvs;
        
            while (std::getline(file, line)) {
                if (line.substr(0, 2) == "vt") {
                    std::istringstream iss(line.substr(3));
                    Vector<float,2, 0> uv;
                    iss >> uv.x >> uv.y;
                    uvs.push_back(uv);
                } else if (line.substr(0, 2) == "vn") {
                    std::istringstream iss(line.substr(3));
                    Vector<float, 3, 0> normal;
                    iss >> normal.x >> normal.y >> normal.z;
                    normals.push_back(normal);
                } else if (line.substr(0, 1) == "v") {
                    std::istringstream iss(line.substr(2));
                    Vector<float,3, 3> position;
                    iss >> position.x >> position.y >> position.z;
                    positions.push_back(position);
                } else if (line.substr(0, 1) == "f") {
                    std::istringstream iss(line.substr(2));
                    std::string vertexGroup;

                    while (iss >> vertexGroup) {
                        std::istringstream vss(vertexGroup);
                        std::string vIndex, vtIndex, vnIndex;

                        std::getline(vss, vIndex, '/');
                        std::getline(vss, vtIndex, '/');
                        std::getline(vss, vnIndex, '/');

                        Vector<float, 3, 0> v  = positions[std::stoi(vIndex) - 1];
                        Vector<float, 2, 0> vt = uvs[std::stoi(vtIndex) - 1];
                        Vector<float, 3, 0> vn = normals[std::stoi(vnIndex) - 1];

                        for (std::uint32_t i = 0; i < mesh.datas.size(); i++) {
                            if (mesh.datas[i].normal == vn && mesh.datas[i].uv == vt && mesh.vertexes[i] == v) {

                                mesh.indexes.push_back(i);
                                goto skip;
                            }
                        }

                        mesh.indexes.push_back(mesh.datas.size());
                        mesh.datas.push_back({vn, {0,0,0}, vt});
                        mesh.vertexes.push_back(v);
                        skip:;
                    }
                }
            }

            // Accumulate face normals and tangents
            for (std::uint32_t i = 0; i < mesh.indexes.size(); i += 3)
            {
                std::uint32_t i0 = mesh.indexes[i];
                std::uint32_t i1 = mesh.indexes[i + 1];
                std::uint32_t i2 = mesh.indexes[i + 2];

                // Edges of the triangle
                Vector<float, 3, 0> edge1 = mesh.vertexes[i1] - mesh.vertexes[i0];
                Vector<float, 3, 0> edge2 = mesh.vertexes[i2] - mesh.vertexes[i0];

                // Texture coordinate deltas
                Vector<float, 2, 0> deltaUV1 = mesh.datas[i1].uv - mesh.datas[i0].uv;
                Vector<float, 2, 0> deltaUV2 = mesh.datas[i2].uv - mesh.datas[i0].uv;

                // Tangent calculation
                float f = 1.0f / (deltaUV1.x * deltaUV2.y - deltaUV1.y * deltaUV2.x);
                Vector<float, 3, 0> tangent;
                tangent.x = f * (deltaUV2.y * edge1.x - deltaUV1.y * edge2.x);
                tangent.y = f * (deltaUV2.y * edge1.y - deltaUV1.y * edge2.y);
                tangent.z = f * (deltaUV2.y * edge1.z - deltaUV1.y * edge2.z);

                // Normalize tangent
                tangent.Normalize();

                // Accumulate normals and tangents for each vertex
                mesh.datas[i0].tangent += tangent;
                mesh.datas[i1].tangent += tangent;
                mesh.datas[i2].tangent += tangent;
            }

            // Normalize vertex normals and tangents
            for (T& v : mesh.datas) {
                v.tangent.Normalize();
            }

            return mesh;
        }
    };

    template <>
    struct MeshAsset<void> {
        std::vector<Vector<float, 3, 3>> vertexes;
        std::vector<std::uint32_t> indexes;
        void Save(fs::path path) {
            std::ofstream file(path, std::ios::binary);

            std::uint32_t vertexCount = vertexes.size();
            std::uint32_t indexCount = indexes.size();

            file.write(reinterpret_cast<char*>(&vertexCount), sizeof(vertexCount));
            file.write(reinterpret_cast<char*>(&indexCount), sizeof(indexCount));

            file.write(reinterpret_cast<char*>(vertexes.data()), vertexCount * sizeof(Vector<float, 3, 3>));
            file.write(reinterpret_cast<char*>(indexes.data()), indexCount * sizeof(std::uint32_t));
        }

        void SaveCompressed(fs::path path) const {
            // Three regions to keep file format identical to the templated
            // variant; the data region is empty (skipped on the GPU path).
            std::array<std::span<const std::byte>, 3> streams = {
                std::as_bytes(std::span(vertexes)),
                std::as_bytes(std::span(indexes)),
                std::span<const std::byte>{},
            };
            Compression::CompressedBlob blob = Compression::CompressStreams(streams);

            std::ofstream file(path, std::ios::binary);
            file.write(MeshAssetFormat::magic, 4);
            std::uint32_t version = MeshAssetFormat::version;
            std::uint32_t vc = static_cast<std::uint32_t>(vertexes.size());
            std::uint32_t ic = static_cast<std::uint32_t>(indexes.size());
            std::uint32_t dc = 0;
            std::uint32_t stride = 0;
            file.write(reinterpret_cast<const char*>(&version), sizeof(version));
            file.write(reinterpret_cast<const char*>(&vc), sizeof(vc));
            file.write(reinterpret_cast<const char*>(&ic), sizeof(ic));
            file.write(reinterpret_cast<const char*>(&dc), sizeof(dc));
            file.write(reinterpret_cast<const char*>(&stride), sizeof(stride));
            Compression::WriteBlob(file, blob);
        }

        static MeshAsset<void> Load(fs::path path) {
            MeshAsset<void> mesh;
            
            std::ifstream file(path, std::ios::binary);

            std::uint32_t vertexCount = 0;
            std::uint32_t indexCount  = 0;

            file.read(reinterpret_cast<char*>(&vertexCount), sizeof(vertexCount));
            file.read(reinterpret_cast<char*>(&indexCount),  sizeof(indexCount));

            mesh.vertexes.resize(vertexCount);
            mesh.indexes.resize(indexCount);

            file.read(reinterpret_cast<char*>(mesh.vertexes.data()), vertexCount * sizeof(Vector<float, 3, 3>));
            file.read(reinterpret_cast<char*>(mesh.indexes.data()), indexCount * sizeof(std::uint32_t));

            return mesh;
        }
        static MeshAsset<void> LoadOBJ(fs::path path) {
            std::ifstream file(path);

            MeshAsset<void> mesh;
                
            std::string line;
        
            while (std::getline(file, line)) {
                if (line.rfind("v ", 0) == 0) {
                    std::istringstream iss(line.substr(2));
                    Vector<float,3, 3> position;
                    iss >> position.x >> position.y >> position.z;
                    mesh.vertexes.push_back(position);
                } else if (line.rfind("f ", 0) == 0) {
                    std::istringstream iss(line.substr(2));
                    std::string vertexGroup;

                    while (iss >> vertexGroup) {
                        mesh.indexes.push_back(std::stoi(vertexGroup) - 1);
                    }
                }
            }

            return mesh;
        }
    };
}