Class ObjLoader

Class ObjLoader#

Class Documentation#

class ObjLoader#

Public Functions

ObjLoader(const std::shared_ptr<VulkanDevice> &device, vk::CommandPool command_pool)#
ObjLoader() = default#

CPU-only construction, for parsing off the render thread (or in a test). loadModel() is unavailable on such an instance - it has no device to upload with.

std::shared_ptr<Model> loadModel(const std::string &modelFile)#
bool parseCpu(const std::string &modelFile)#

Reads and decodes the OBJ into CPU-side vertex/index/material arrays.

Touches NO Vulkan, which is the point: this is ~99% of model load time (measured 1906 ms parse + 867 ms vertex build against 15 ms of GPU upload on the bundled 27 MB model), so it is the part worth moving to a worker thread. Returns false on a malformed or missing file, leaving the instance empty.

std::shared_ptr<Model> uploadParsed()#

Builds the Vulkan-side Model from the last parseCpu result.

MUST run on the thread that owns the device. Returns nullptr on a device-free loader or before a successful parse - the pairing is a programming error, not a runtime condition, so it logs rather than failing silently.

inline void adoptParsed(ObjLoader &&other)#

Moves another loader’s parse results into this one, so a device-owning loader can upload what a device-free worker produced. Copying the arrays instead would duplicate tens of MB for no reason.

inline const std::vector<Vertex> &getVertices() const#

Results of the last parseCpu / loadModel. Exposed so a worker can hand the data to the thread that owns the device.

inline const std::vector<unsigned int> &getIndices() const#
inline const std::vector<ObjMaterial> &getMaterials() const#
inline const std::vector<unsigned int> &getMaterialIndices() const#
inline const std::vector<std::string> &getTextureNames() const#
inline const std::vector<unsigned char> &getTextureSrgbFlags() const#
inline const std::vector<MeshRange> &getMeshRanges() const#

One entry per OBJ shape (o/g group): its slice of the flat vertices/ indices/materialIndex arrays. uploadParsed builds one Mesh per range, so a multi-shape OBJ becomes a multi-mesh Model (backlog #10, the OBJ parallel of the glTF primitive split) while the flat getters stay exactly as they were - the ObjParseUnit tests assert only on those. Unlike glTF primitives, OBJ shapes share the attribute pool, so a per-shape range is only contiguous because loadVertices resets the vertex-dedup map per shape (a vertex shared across shapes is stored once per shape - pixel-identical geometry). A single-shape OBJ yields one range spanning everything - behaviour-identical to before. MeshRange itself is the shared kataglyphis.vulkan.mesh_range type.