#ifdef RIVE_WASM_MODULE
#if defined(RIVE_CANVAS) && defined(RIVE_ORE)
// Module-side ore objects over rive_gpu_v1 handles, the same shape
// render_proxy.cpp gives the 2D surface. The binding files in lua_gpu.cpp
// compile in untouched and talk to this context; factories realize against
// imports namespace slice by namespace slice, the rest report through
// lastError like any backend construction failure.
#include "rive/renderer/ore/ore_types.hpp"
#include "rive/renderer/ore/ore_buffer.hpp"
#include "rive/renderer/ore/ore_texture.hpp"
#include "rive/renderer/ore/ore_sampler.hpp"
#include "rive/renderer/ore/ore_shader_module.hpp"
#include "rive/renderer/ore/ore_bind_group_layout.hpp"
#include "rive/renderer/ore/ore_bind_group.hpp"
#include "rive/renderer/ore/ore_pipeline.hpp"
#include "rive/renderer/ore/ore_render_pass.hpp"
#include "rive/renderer/render_canvas.hpp"
#include "rive/renderer/render_target.hpp"
#include "rive/wasm/module_render.hpp"
#include "rive/renderer/ore/ore_context.hpp"
#include "rive/wasm/rive_bindings_v1.h"

#include <algorithm>
#include <cstring>
#include <string>
#include <vector>

namespace rive
{
namespace
{

class ModuleOreBuffer : public ore::Buffer
{
public:
  ModuleOreBuffer(uint32_t handle, uint32_t size, ore::BufferUsage usage)
      : ore::Buffer(size, usage)
      , m_handle(handle)
  {}
  ~ModuleOreBuffer() override { rive_gpu_buffer_release(m_handle); }
  uint32_t handle() const { return m_handle; }
  void update(const void* data, uint32_t size, uint32_t offset) override
  {
    rive_gpu_buffer_update(
        m_handle,
        offset,
        static_cast<const uint8_t*>(data),
        size);
  }

private:
  uint32_t m_handle;
};

class ModuleOreTexture : public ore::Texture
{
public:
  ModuleOreTexture(uint32_t handle, const ore::TextureDesc& desc)
      : ore::Texture(desc)
      , m_handle(handle)
  {}
  // Handle 0 is the canvas metadata wrapper, which owns nothing host side.
  ~ModuleOreTexture() override
  {
    if (m_handle != 0)
    {
      rive_gpu_texture_release(m_handle);
    }
  }
  uint32_t handle() const { return m_handle; }
  void uploadImpl(const ore::TextureDataDesc& data) override
  {
    rive_gpu_texture_upload_v1 region;
    region.bytesPerRow = data.bytesPerRow;
    region.rowsPerImage = data.rowsPerImage;
    region.mipLevel = data.mipLevel;
    region.layer = data.layer;
    region.x = data.x;
    region.y = data.y;
    region.z = data.z;
    region.width = data.width;
    region.height = data.height;
    region.depth = data.depth;
    uint32_t byteCount = data.bytesPerRow * data.rowsPerImage *
                         std::max(1u, data.depth);
    rive_gpu_texture_upload(
        m_handle,
        &region,
        sizeof(region),
        static_cast<const uint8_t*>(data.data),
        byteCount);
  }

private:
  uint32_t m_handle;
};

class ModuleOreSampler : public ore::Sampler
{
public:
  ModuleOreSampler(uint32_t handle) : m_handle(handle) {}
  ~ModuleOreSampler() override { rive_gpu_sampler_release(m_handle); }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
};

class ModuleOreTextureView : public ore::TextureView
{
public:
  ModuleOreTextureView(
      uint32_t handle,
      rcp<ore::Texture> texture,
      const ore::TextureViewDesc& desc)
      : ore::TextureView(std::move(texture), desc)
      , m_handle(handle)
  {}
  ~ModuleOreTextureView() override { rive_gpu_texture_view_release(m_handle); }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
};

class ModuleOreShaderModule : public ore::ShaderModule
{
public:
  ModuleOreShaderModule(uint32_t handle, const ore::ShaderModuleDesc& desc)
      : m_handle(handle)
  {
    // Parses the binding-map sidecar into m_bindingMap so module-side
    // layout derivation walks the same data every backend does.
    applyBindingMapFromDesc(desc);
  }
  ~ModuleOreShaderModule() override
  {
    rive_gpu_shader_module_release(m_handle);
  }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
};

class ModuleOreBindGroupLayout : public ore::BindGroupLayout
{
public:
  ModuleOreBindGroupLayout(
      uint32_t handle,
      const ore::BindGroupLayoutDesc& desc)
      : m_handle(handle)
  {
    m_groupIndex = desc.groupIndex;
    m_entries.assign(desc.entries, desc.entries + desc.entryCount);
  }
  ~ModuleOreBindGroupLayout() override
  {
    rive_gpu_bind_group_layout_release(m_handle);
  }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
};

class ModuleOreBindGroup : public ore::BindGroup
{
public:
  ModuleOreBindGroup(uint32_t handle, const ore::BindGroupDesc& desc)
      : m_handle(handle)
  {
    m_layoutRef = ref_rcp(desc.layout);
    recordDynamicRanges(desc);
    for (uint32_t i = 0; i < desc.uboCount; i++)
    {
      m_retainedBuffers.push_back(ref_rcp(desc.ubos[i].buffer));
      if (m_layoutRef->hasDynamicOffset(desc.ubos[i].slot))
      {
        m_dynamicOffsetCount++;
      }
    }
    for (uint32_t i = 0; i < desc.textureCount; i++)
    {
      m_retainedViews.push_back(ref_rcp(desc.textures[i].view));
    }
    for (uint32_t i = 0; i < desc.samplerCount; i++)
    {
      m_retainedSamplers.push_back(ref_rcp(desc.samplers[i].sampler));
    }
  }
  ~ModuleOreBindGroup() override { rive_gpu_bind_group_release(m_handle); }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
};

class ModuleOrePipeline : public ore::Pipeline
{
public:
  ModuleOrePipeline(uint32_t handle, const ore::PipelineDesc& desc)
      : ore::Pipeline(desc)
      , m_handle(handle)
      , m_vertexEntry(desc.vertexEntryPoint)
      , m_fragmentEntry(desc.fragmentEntryPoint)
  {
    // The base copies the caller's entry point pointers; repoint them at
    // owned storage so desc() stays valid past construction.
    m_desc.vertexEntryPoint = m_vertexEntry.c_str();
    m_desc.fragmentEntryPoint = m_fragmentEntry.c_str();
  }
  ~ModuleOrePipeline() override { rive_gpu_pipeline_release(m_handle); }
  uint32_t handle() const { return m_handle; }

private:
  uint32_t m_handle;
  std::string m_vertexEntry;
  std::string m_fragmentEntry;
};

// Mirrors RenderPassRecording: the base validators run module side so
// lastError parity holds, then every command forwards over the wire.
class ModuleOreRenderPass : public ore::RenderPass
{
public:
  ModuleOreRenderPass(
      ore::Context* context,
      uint32_t handle,
      const ore::RenderPassDesc& desc)
      : ore::RenderPass(context)
      , m_handle(handle)
  {
    populateAttachmentMetadata(desc);
  }
  ~ModuleOreRenderPass() override { rive_gpu_pass_release(m_handle); }
  uint32_t handle() const { return m_handle; }

  void setPipeline(ore::Pipeline* pipeline) override
  {
    if (!checkPipelineCompat(pipeline))
    {
      return;
    }
    rive_gpu_pass_set_pipeline(
        m_handle,
        static_cast<ModuleOrePipeline*>(pipeline)->handle());
  }
  void setVertexBuffer(uint32_t slot, ore::Buffer* buffer, uint32_t offset)
      override
  {
    rive_gpu_pass_set_vertex_buffer(
        m_handle,
        slot,
        static_cast<ModuleOreBuffer*>(buffer)->handle(),
        offset);
  }
  void setIndexBuffer(
      ore::Buffer* buffer,
      ore::IndexFormat format,
      uint32_t offset) override
  {
    rive_gpu_pass_set_index_buffer(
        m_handle,
        static_cast<ModuleOreBuffer*>(buffer)->handle(),
        (uint32_t)format,
        offset);
  }
  void setBindGroup(
      uint32_t groupIndex,
      ore::BindGroup* bg,
      const uint32_t* dynamicOffsets,
      uint32_t dynamicOffsetCount) override
  {
    if (groupIndex < ore::kMaxBindGroups)
    {
      m_boundGroups[groupIndex] = ref_rcp(bg);
    }
    rive_gpu_pass_set_bind_group(
        m_handle,
        groupIndex,
        static_cast<ModuleOreBindGroup*>(bg)->handle(),
        dynamicOffsets,
        dynamicOffsetCount * (uint32_t)sizeof(uint32_t));
  }
  void setViewport(
      float x,
      float y,
      float width,
      float height,
      float minDepth,
      float maxDepth) override
  {
    rive_gpu_pass_set_viewport(
        m_handle,
        x,
        y,
        width,
        height,
        minDepth,
        maxDepth);
  }
  void setScissorRect(uint32_t x, uint32_t y, uint32_t width, uint32_t height)
      override
  {
    rive_gpu_pass_set_scissor(m_handle, x, y, width, height);
  }
  void setStencilReference(uint32_t ref) override
  {
    rive_gpu_pass_set_stencil_reference(m_handle, ref);
  }
  void setBlendColor(float r, float g, float b, float a) override
  {
    rive_gpu_pass_set_blend_color(m_handle, r, g, b, a);
  }
  void draw(
      uint32_t vertexCount,
      uint32_t instanceCount,
      uint32_t firstVertex,
      uint32_t firstInstance) override
  {
    rive_gpu_pass_draw(
        m_handle,
        vertexCount,
        instanceCount,
        firstVertex,
        firstInstance);
  }
  void drawIndexed(
      uint32_t indexCount,
      uint32_t instanceCount,
      uint32_t firstIndex,
      int32_t baseVertex,
      uint32_t firstInstance) override
  {
    rive_gpu_pass_draw_indexed(
        m_handle,
        indexCount,
        instanceCount,
        firstIndex,
        baseVertex,
        firstInstance);
  }
  void finish() override
  {
    if (m_finished)
    {
      return;
    }
    rive_gpu_pass_finish(m_handle);
    m_finished = true;
    for (uint32_t i = 0; i < ore::kMaxBindGroups; ++i)
    {
      m_boundGroups[i] = nullptr;
    }
  }

private:
  uint32_t m_handle;
};

// A view over a host color target, described by the props the host fills:
// width, height, format and sample count. The texture is metadata only, so
// attachment validation sees the real format and sample count.
rcp<ore::TextureView> makeModuleColorView(
    uint32_t handle,
    const uint32_t props[4])
{
  ore::TextureDesc textureDesc;
  textureDesc.width = props[0];
  textureDesc.height = props[1];
  textureDesc.format = (ore::TextureFormat)props[2];
  textureDesc.sampleCount = props[3];
  textureDesc.renderTarget = true;
  ore::TextureViewDesc viewDesc;
  auto texture = make_rcp<ModuleOreTexture>(0, textureDesc);
  viewDesc.texture = texture.get();
  return make_rcp<ModuleOreTextureView>(handle, std::move(texture), viewDesc);
}

class ModuleOreContext : public ore::Context
{
public:
  ModuleOreContext() : ore::Context(nullptr) {}

  rcp<ore::Buffer> makeBuffer(const ore::BufferDesc& desc) override
  {
    uint32_t handle = rive_gpu_buffer_new(
        (uint32_t)desc.usage,
        desc.size,
        desc.immutable ? 1 : 0,
        static_cast<const uint8_t*>(desc.data),
        desc.data != nullptr ? desc.size : 0);
    if (handle == 0)
    {
      setLastError("host rejected buffer");
      return nullptr;
    }
    return make_rcp<ModuleOreBuffer>(handle, desc.size, desc.usage);
  }

  rcp<ore::Texture> makeTexture(const ore::TextureDesc& desc) override
  {
    rive_gpu_texture_desc_v1 pod;
    pod.width = desc.width;
    pod.height = desc.height;
    pod.depthOrArrayLayers = desc.depthOrArrayLayers;
    pod.format = (uint32_t)desc.format;
    pod.textureType = (uint32_t)desc.type;
    pod.renderTarget = desc.renderTarget ? 1 : 0;
    pod.numMipmaps = desc.numMipmaps;
    pod.sampleCount = desc.sampleCount;
    uint32_t handle = rive_gpu_texture_new(&pod, sizeof(pod));
    if (handle == 0)
    {
      setLastError("host rejected texture");
      return nullptr;
    }
    return make_rcp<ModuleOreTexture>(handle, desc);
  }

  rcp<ore::TextureView> makeTextureViewImpl(
      const ore::TextureViewDesc& desc) override
  {
    auto* texture = static_cast<ModuleOreTexture*>(desc.texture);
    if (texture == nullptr)
    {
      setLastError("texture view requires a texture");
      return nullptr;
    }
    rive_gpu_texture_view_desc_v1 pod;
    pod.dimension = (uint32_t)desc.dimension;
    pod.aspect = (uint32_t)desc.aspect;
    pod.baseMipLevel = desc.baseMipLevel;
    pod.mipCount = desc.mipCount;
    pod.baseLayer = desc.baseLayer;
    pod.layerCount = desc.layerCount;
    uint32_t handle = rive_gpu_texture_view_new(
        texture->handle(),
        &pod,
        sizeof(pod));
    if (handle == 0)
    {
      setLastError("host rejected texture view");
      return nullptr;
    }
    return make_rcp<ModuleOreTextureView>(
        handle,
        ref_rcp<ore::Texture>(texture),
        desc);
  }
  rcp<ore::Sampler> makeSampler(const ore::SamplerDesc& desc) override
  {
    rive_gpu_sampler_desc_v1 pod;
    pod.minFilter = (uint32_t)desc.minFilter;
    pod.magFilter = (uint32_t)desc.magFilter;
    pod.mipmapFilter = (uint32_t)desc.mipmapFilter;
    pod.wrapU = (uint32_t)desc.wrapU;
    pod.wrapV = (uint32_t)desc.wrapV;
    pod.wrapW = (uint32_t)desc.wrapW;
    pod.compare = (uint32_t)desc.compare;
    pod.minLod = desc.minLod;
    pod.maxLod = desc.maxLod;
    pod.maxAnisotropy = desc.maxAnisotropy;
    uint32_t handle = rive_gpu_sampler_new(&pod, sizeof(pod));
    if (handle == 0)
    {
      setLastError("host rejected sampler");
      return nullptr;
    }
    return ore::describedSampler(make_rcp<ModuleOreSampler>(handle), desc);
  }
  // The host builds shaders only from admitted assets, through
  // wasmModuleShaderModule.
  rcp<ore::ShaderModule> makeShaderModule(const ore::ShaderModuleDesc&) override
  {
    setLastError("host builds shader modules only from shader assets");
    return nullptr;
  }
  rcp<ore::BindGroupLayout> makeBindGroupLayout(
      const ore::BindGroupLayoutDesc& desc) override
  {
    std::vector<rive_gpu_bind_group_layout_entry_v1> pods(desc.entryCount);
    for (uint32_t i = 0; i < desc.entryCount; i++)
    {
      auto& pod = pods[i];
      const auto& entry = desc.entries[i];
      pod.binding = entry.binding;
      pod.kind = (uint32_t)entry.kind;
      pod.visibility = entry.visibility.mask;
      pod.hasDynamicOffset = entry.hasDynamicOffset ? 1 : 0;
      pod.textureViewDim = (uint32_t)entry.textureViewDim;
      pod.textureSampleType = (uint32_t)entry.textureSampleType;
      pod.textureMultisampled = entry.textureMultisampled ? 1 : 0;
      pod.minBindingSize = entry.minBindingSize;
      pod.nativeSlotVS = entry.nativeSlotVS;
      pod.nativeSlotFS = entry.nativeSlotFS;
      pod.nativeSlotCS = entry.nativeSlotCS;
      pod.samplerNonFiltering = entry.samplerNonFiltering ? 1 : 0;
    }
    uint32_t handle = rive_gpu_bind_group_layout_new(
        desc.groupIndex,
        pods.data(),
        (uint32_t)(pods.size() * sizeof(pods[0])));
    if (handle == 0)
    {
      setLastError("host rejected bind group layout");
      return nullptr;
    }
    return make_rcp<ModuleOreBindGroupLayout>(handle, desc);
  }
  rcp<ore::Pipeline> makePipeline(
      const ore::PipelineDesc& desc,
      std::string* outError) override
  {
    rive_gpu_pipeline_desc_v1 pod = {};
    auto moduleHandle = [](ore::ShaderModule* module) -> uint32_t {
      return module != nullptr
                 ? static_cast<ModuleOreShaderModule*>(module)->handle()
                 : 0;
    };
    pod.vertexModule = moduleHandle(desc.vertexModule);
    pod.fragmentModule = moduleHandle(desc.fragmentModule);
    pod.vertexEntrySize = desc.vertexEntryPoint != nullptr
                              ? (uint32_t)strlen(desc.vertexEntryPoint)
                              : 0;
    pod.fragmentEntrySize = desc.fragmentEntryPoint != nullptr
                                ? (uint32_t)strlen(desc.fragmentEntryPoint)
                                : 0;
    pod.colorCount = desc.colorCount;
    pod.vertexBufferCount = desc.vertexBufferCount;
    uint32_t attributeCount = 0;
    for (uint32_t i = 0; i < desc.vertexBufferCount; i++)
    {
      attributeCount += desc.vertexBuffers[i].attributeCount;
    }
    pod.attributeCount = attributeCount;
    pod.bindGroupLayoutCount = desc.bindGroupLayoutCount;
    pod.topology = (uint32_t)desc.topology;
    pod.indexFormat = (uint32_t)desc.indexFormat;
    pod.cullMode = (uint32_t)desc.cullMode;
    pod.winding = (uint32_t)desc.winding;
    pod.depthFormat = (uint32_t)desc.depthStencil.format;
    pod.depthCompare = (uint32_t)desc.depthStencil.depthCompare;
    pod.depthWriteEnabled = desc.depthStencil.depthWriteEnabled ? 1 : 0;
    pod.depthBias = (uint32_t)desc.depthStencil.depthBias;
    pod.depthBiasSlopeScale = desc.depthStencil.depthBiasSlopeScale;
    pod.depthBiasClamp = desc.depthStencil.depthBiasClamp;
    pod.stencilFrontCompare = (uint32_t)desc.stencilFront.compare;
    pod.stencilFrontFailOp = (uint32_t)desc.stencilFront.failOp;
    pod.stencilFrontDepthFailOp = (uint32_t)desc.stencilFront.depthFailOp;
    pod.stencilFrontPassOp = (uint32_t)desc.stencilFront.passOp;
    pod.stencilBackCompare = (uint32_t)desc.stencilBack.compare;
    pod.stencilBackFailOp = (uint32_t)desc.stencilBack.failOp;
    pod.stencilBackDepthFailOp = (uint32_t)desc.stencilBack.depthFailOp;
    pod.stencilBackPassOp = (uint32_t)desc.stencilBack.passOp;
    pod.stencilReadMask = desc.stencilReadMask;
    pod.stencilWriteMask = desc.stencilWriteMask;
    pod.sampleCount = desc.sampleCount;

    std::vector<uint8_t> blob;
    auto append = [&blob](const void* data, size_t size) {
      if (size != 0)
      {
        const uint8_t* bytes = static_cast<const uint8_t*>(data);
        blob.insert(blob.end(), bytes, bytes + size);
      }
    };
    append(desc.vertexEntryPoint, pod.vertexEntrySize);
    append(desc.fragmentEntryPoint, pod.fragmentEntrySize);
    for (uint32_t i = 0; i < desc.colorCount; i++)
    {
      rive_gpu_color_target_v1 target;
      target.format = (uint32_t)desc.colorTargets[i].format;
      target.blendEnabled = desc.colorTargets[i].blendEnabled ? 1 : 0;
      target.srcColor = (uint32_t)desc.colorTargets[i].blend.srcColor;
      target.dstColor = (uint32_t)desc.colorTargets[i].blend.dstColor;
      target.colorOp = (uint32_t)desc.colorTargets[i].blend.colorOp;
      target.srcAlpha = (uint32_t)desc.colorTargets[i].blend.srcAlpha;
      target.dstAlpha = (uint32_t)desc.colorTargets[i].blend.dstAlpha;
      target.alphaOp = (uint32_t)desc.colorTargets[i].blend.alphaOp;
      target.writeMask = (uint32_t)desc.colorTargets[i].writeMask;
      append(&target, sizeof(target));
    }
    for (uint32_t i = 0; i < desc.vertexBufferCount; i++)
    {
      rive_gpu_vertex_buffer_layout_v1 layout;
      layout.stride = desc.vertexBuffers[i].stride;
      layout.stepMode = (uint32_t)desc.vertexBuffers[i].stepMode;
      layout.attributeCount = desc.vertexBuffers[i].attributeCount;
      append(&layout, sizeof(layout));
    }
    for (uint32_t i = 0; i < desc.vertexBufferCount; i++)
    {
      for (uint32_t a = 0; a < desc.vertexBuffers[i].attributeCount; a++)
      {
        rive_gpu_vertex_attribute_v1 attribute;
        attribute.format = (uint32_t)desc.vertexBuffers[i].attributes[a].format;
        attribute.offset = desc.vertexBuffers[i].attributes[a].offset;
        attribute.shaderSlot = desc.vertexBuffers[i].attributes[a].shaderSlot;
        append(&attribute, sizeof(attribute));
      }
    }
    for (uint32_t i = 0; i < desc.bindGroupLayoutCount; i++)
    {
      uint32_t handle = desc.bindGroupLayouts[i] != nullptr
                            ? static_cast<ModuleOreBindGroupLayout*>(
                                  desc.bindGroupLayouts[i])
                                  ->handle()
                            : 0;
      append(&handle, sizeof(handle));
    }

    uint32_t handle = rive_gpu_pipeline_new(
        &pod,
        sizeof(pod),
        blob.data(),
        (uint32_t)blob.size());
    if (handle == 0)
    {
      if (outError != nullptr)
      {
        *outError = "host rejected pipeline";
      }
      setLastError("host rejected pipeline");
      return nullptr;
    }
    return make_rcp<ModuleOrePipeline>(handle, desc);
  }
  rcp<ore::BindGroup> makeBindGroup(const ore::BindGroupDesc& desc) override
  {
    auto* layout = static_cast<ModuleOreBindGroupLayout*>(desc.layout);
    if (layout == nullptr)
    {
      setLastError("bind group requires a layout");
      return nullptr;
    }
    if (std::string err; !ore::validateBindGroupDesc(desc, &err))
    {
      setLastError("makeBindGroup: %s", err.c_str());
      return nullptr;
    }
    std::vector<rive_gpu_bind_group_ubo_v1> ubos(desc.uboCount);
    for (uint32_t i = 0; i < desc.uboCount; i++)
    {
      ubos[i].slot = desc.ubos[i].slot;
      ubos[i].buffer = static_cast<ModuleOreBuffer*>(desc.ubos[i].buffer)
                           ->handle();
      ubos[i].offset = desc.ubos[i].offset;
      ubos[i].size = desc.ubos[i].size;
    }
    std::vector<rive_gpu_bind_group_texture_v1> textures(desc.textureCount);
    for (uint32_t i = 0; i < desc.textureCount; i++)
    {
      textures[i].slot = desc.textures[i].slot;
      textures[i].view = static_cast<ModuleOreTextureView*>(
                             desc.textures[i].view)
                             ->handle();
    }
    std::vector<rive_gpu_bind_group_sampler_v1> samplers(desc.samplerCount);
    for (uint32_t i = 0; i < desc.samplerCount; i++)
    {
      samplers[i].slot = desc.samplers[i].slot;
      samplers[i].sampler = static_cast<ModuleOreSampler*>(
                                desc.samplers[i].sampler)
                                ->handle();
    }
    uint32_t handle = rive_gpu_bind_group_new(
        layout->handle(),
        ubos.data(),
        (uint32_t)(ubos.size() * sizeof(rive_gpu_bind_group_ubo_v1)),
        textures.data(),
        (uint32_t)(textures.size() * sizeof(rive_gpu_bind_group_texture_v1)),
        samplers.data(),
        (uint32_t)(samplers.size() * sizeof(rive_gpu_bind_group_sampler_v1)));
    if (handle == 0)
    {
      setLastError("host rejected bind group");
      return nullptr;
    }
    return make_rcp<ModuleOreBindGroup>(handle, desc);
  }
  std::unique_ptr<ore::RenderPass> beginRenderPass(
      const ore::RenderPassDesc& desc,
      std::string* outError) override
  {
    auto viewHandle = [](ore::TextureView* view) -> uint32_t {
      return view != nullptr
                 ? static_cast<ModuleOreTextureView*>(view)->handle()
                 : 0;
    };
    rive_gpu_pass_desc_v1 pod = {};
    pod.colorCount = desc.colorCount;
    pod.depthView = viewHandle(desc.depthStencil.view);
    pod.depthLoadOp = (uint32_t)desc.depthStencil.depthLoadOp;
    pod.depthStoreOp = (uint32_t)desc.depthStencil.depthStoreOp;
    pod.depthClearValue = desc.depthStencil.depthClearValue;
    pod.stencilLoadOp = (uint32_t)desc.depthStencil.stencilLoadOp;
    pod.stencilStoreOp = (uint32_t)desc.depthStencil.stencilStoreOp;
    pod.stencilClearValue = desc.depthStencil.stencilClearValue;
    rive_gpu_pass_color_attachment_v1 colors[4] = {};
    for (uint32_t i = 0; i < desc.colorCount && i < 4; i++)
    {
      colors[i].view = viewHandle(desc.colorAttachments[i].view);
      colors[i].resolveTarget = viewHandle(
          desc.colorAttachments[i].resolveTarget);
      colors[i].loadOp = (uint32_t)desc.colorAttachments[i].loadOp;
      colors[i].storeOp = (uint32_t)desc.colorAttachments[i].storeOp;
      colors[i].clearR = desc.colorAttachments[i].clearColor.r;
      colors[i].clearG = desc.colorAttachments[i].clearColor.g;
      colors[i].clearB = desc.colorAttachments[i].clearColor.b;
      colors[i].clearA = desc.colorAttachments[i].clearColor.a;
    }
    pod.labelSize = desc.label != nullptr ? (uint32_t)strlen(desc.label) : 0;
    const uint32_t colorBytes = std::min(desc.colorCount, 4u) *
                                (uint32_t)sizeof(colors[0]);
    auto blob = reinterpret_cast<const uint8_t*>(colors);
    alignas(rive_gpu_pass_color_attachment_v1)
        uint8_t stackBlob[sizeof(colors) + 128];
    std::vector<uint8_t> heapBlob;
    if (pod.labelSize != 0)
    {
      uint8_t* bytes = stackBlob;
      if (colorBytes + pod.labelSize > sizeof(stackBlob))
      {
        heapBlob.resize(colorBytes + pod.labelSize);
        bytes = heapBlob.data();
      }
      memcpy(bytes, colors, colorBytes);
      memcpy(bytes + colorBytes, desc.label, pod.labelSize);
      blob = bytes;
    }
    uint32_t handle = rive_gpu_pass_begin(
        &pod,
        sizeof(pod),
        blob,
        colorBytes + pod.labelSize);
    if (handle == 0)
    {
      if (outError != nullptr)
      {
        *outError = "host rejected render pass";
      }
      setLastError("host rejected render pass");
      return nullptr;
    }
    return std::make_unique<ModuleOreRenderPass>(this, handle, desc);
  }
  rcp<ore::TextureView> wrapCanvasTexture(gpu::RenderCanvas*) override
  {
    return notPortedYet<ore::TextureView>("canvas wraps");
  }

  // The host hands back the same handle until its target changes.
  rcp<ore::TextureView> targetView() override
  {
    uint32_t current = m_targetView != nullptr
                           ? static_cast<ModuleOreTextureView*>(
                                 m_targetView.get())
                                 ->handle()
                           : 0;
    uint32_t props[4] = {};
    uint32_t handle = rive_gpu_target_view(current, props, 4);
    if (handle != current)
    {
      m_targetView = handle != 0 ? makeModuleColorView(handle, props) : nullptr;
    }
    return m_targetView;
  }
  rcp<ore::TextureView> wrapRiveTexture(gpu::Texture*, uint32_t, uint32_t)
      override
  {
    return notPortedYet<ore::TextureView>("image wraps");
  }

  // Module images reach the host by handle; the host reruns the same
  // dispatch the Luau binding does, so a canvas and a foreign image both
  // forward as their image handle and differ only on the host side.
  rcp<ore::TextureView> recordWrapCanvasImage(
      gpu::RenderCanvas* canvas,
      uint32_t width,
      uint32_t height) override
  {
    return recordImageViewByHandle(
        canvas != nullptr ? canvas->renderImage() : nullptr,
        width,
        height);
  }

  rcp<ore::TextureView> recordWrapForeignImageView(
      RenderImage* image,
      uint32_t width,
      uint32_t height) override
  {
    return recordImageViewByHandle(image, width, height);
  }

private:
  rcp<ore::TextureView> recordImageViewByHandle(
      RenderImage* image,
      uint32_t width,
      uint32_t height)
  {
    uint32_t handle = rive_gpu_image_view(
        wasmModuleImageHandle(image),
        width,
        height);
    if (handle == 0)
    {
      setLastError("host rejected image view");
      return nullptr;
    }
    ore::TextureDesc textureDesc;
    textureDesc.width = width;
    textureDesc.height = height;
    ore::TextureViewDesc viewDesc;
    auto texture = make_rcp<ModuleOreTexture>(0, textureDesc);
    viewDesc.texture = texture.get();
    return make_rcp<ModuleOreTextureView>(handle, std::move(texture), viewDesc);
  }

public:
  void beginFrame(const FrameDescriptor&) override {}
  void endFrame() override {}
  void waitForGPU() override {}
  // Mirrors the host answer so RSTB entry selection matches the replay
  // backend.
  ore::ShaderTarget shaderTarget() const override
  {
    if (!m_shaderTargetKnown)
    {
      m_shaderTarget = (ore::ShaderTarget)rive_gpu_shader_target();
      m_shaderTargetKnown = true;
    }
    return m_shaderTarget;
  }
  // Same contract as a recording context with no replay device bound:
  // capability gates cannot be decided module side, so they let the call
  // through and the real backend stays the authority.
  bool featuresKnown() const override { return false; }
  // The host context this proxies is the deferred recorder; binding paths
  // that must not touch a driver gate on this.
  bool isRecording() const override { return true; }

private:
  template <typename T> rcp<T> notPortedYet(const char* what)
  {
    setLastError("%s are not available module side yet", what);
    return nullptr;
  }

  mutable ore::ShaderTarget m_shaderTarget = ore::ShaderTarget::wgsl;
  mutable bool m_shaderTargetKnown = false;
  rcp<ore::TextureView> m_targetView;
};

} // namespace

namespace
{
// Owns the host canvas handle: RenderCanvas has no virtual destructor, so
// release rides the target it retains instead.
class ModuleRenderTarget : public gpu::RenderTarget
{
public:
  ModuleRenderTarget(uint32_t canvasHandle, uint32_t width, uint32_t height)
      : gpu::RenderTarget(width, height)
      , m_handle(canvasHandle)
  {}
  ~ModuleRenderTarget() override
  {
    if (m_handle != 0)
    {
      rive_gpu_canvas_release(m_handle);
    }
  }

  uint32_t handle() const { return m_handle; }
  // Transfers canvas handle ownership to a successor target on resize.
  uint32_t disownHandle()
  {
    uint32_t handle = m_handle;
    m_handle = 0;
    return handle;
  }

private:
  uint32_t m_handle;
};
} // namespace

WasmModuleCanvas wasmModuleWrapCanvas(uint32_t canvasHandle)
{
  uint32_t props[4] = {};
  uint32_t viewHandle = rive_gpu_canvas_color_view(canvasHandle, props, 4);
  if (viewHandle == 0)
  {
    return {};
  }
  WasmModuleCanvas out;
  // Backing installs unbacked (null image texture, module semantics);
  // construction alone no longer carries the render target.
  out.canvas = make_rcp<gpu::RenderCanvas>(props[0], props[1]);
  out.canvas->setBacking(
      nullptr,
      make_rcp<ModuleRenderTarget>(canvasHandle, props[0], props[1]));
  out.colorView = makeModuleColorView(viewHandle, props);
  return out;
}

WasmModuleCanvas wasmModuleResizeCanvas(
    const rcp<gpu::RenderCanvas>& canvas,
    uint32_t width,
    uint32_t height)
{
  if (canvas == nullptr)
  {
    return {};
  }
  auto* target = static_cast<ModuleRenderTarget*>(canvas->renderTarget());
  uint32_t props[4] = {};
  uint32_t viewHandle = rive_gpu_canvas_resize(
      target->handle(),
      width,
      height,
      props,
      4);
  if (viewHandle == 0)
  {
    return {};
  }
  WasmModuleCanvas out;
  out.canvas = make_rcp<gpu::RenderCanvas>(props[0], props[1]);
  out.canvas->setBacking(
      nullptr,
      make_rcp<ModuleRenderTarget>(target->disownHandle(), props[0], props[1]));
  out.colorView = makeModuleColorView(viewHandle, props);
  return out;
}

uint32_t wasmModuleCanvasImageHandle(const rcp<gpu::RenderCanvas>& canvas)
{
  if (canvas == nullptr)
  {
    return 0;
  }
  auto* target = static_cast<ModuleRenderTarget*>(canvas->renderTarget());
  return rive_gpu_canvas_image(target->handle());
}

namespace
{
uint32_t g_shaderObject = 0;
std::string g_shaderName;
} // namespace

void wasmModuleShaderSource(uint32_t object, const std::string& name)
{
  g_shaderObject = object;
  g_shaderName = name;
}

rcp<ore::ShaderModule> wasmModuleShaderModule(
    const ore::ShaderModuleDesc& desc,
    uint32_t entry)
{
  uint32_t handle = rive_gpu_shader_module_from_asset(
      g_shaderObject,
      g_shaderName.data(),
      (uint32_t)g_shaderName.size(),
      entry);
  if (handle == 0)
  {
    return nullptr;
  }
  return make_rcp<ModuleOreShaderModule>(handle, desc);
}

ore::Context* wasmModuleOreContext()
{
  static ModuleOreContext context;
  return &context;
}
} // namespace rive

#endif
#endif