#include "common_uniforms"

@group(0) @binding(0) var<uniform> globals: GlobalUniforms;
@group(0) @binding(1) var<storage, read> object_data: ObjectsBuffer;

// Binding 2 is reserved for BVH buffer when enabled

@group(0) @binding(3) var noise_tex: texture_2d<f32>;
@group(0) @binding(4) var noise_sampler: sampler;
@group(0) @binding(5) var sky_tex: texture_2d<f32>;

struct VertexOutput {
    @builtin(position) position: vec4<f32>,
    @location(0) local_pos: vec3<f32>,
    @location(1) color: vec4<f32>,
    @location(2) @interpolate(flat) instance_index: u32,
    @location(3) world_pos: vec3<f32>,
    @location(4) transformed_normal: vec3<f32>,
};

@vertex
fn vs_main(@builtin(vertex_index) vertex_index: u32, 
           @builtin(instance_index) instance_index: u32) -> VertexOutput {
    
    var pos = array<vec3<f32>, 36>(
        vec3(-1.0, -1.0,  1.0), vec3( 1.0, -1.0,  1.0), vec3( 1.0,  1.0,  1.0),
        vec3(-1.0, -1.0,  1.0), vec3( 1.0,  1.0,  1.0), vec3(-1.0,  1.0,  1.0),
        vec3(-1.0, -1.0, -1.0), vec3(-1.0,  1.0, -1.0), vec3( 1.0,  1.0, -1.0),
        vec3(-1.0, -1.0, -1.0), vec3( 1.0,  1.0, -1.0), vec3( 1.0, -1.0, -1.0),
        vec3(-1.0,  1.0, -1.0), vec3(-1.0,  1.0,  1.0), vec3( 1.0,  1.0,  1.0),
        vec3(-1.0,  1.0, -1.0), vec3( 1.0,  1.0,  1.0), vec3( 1.0,  1.0, -1.0),
        vec3(-1.0, -1.0, -1.0), vec3( 1.0, -1.0, -1.0), vec3( 1.0, -1.0,  1.0),
        vec3(-1.0, -1.0, -1.0), vec3( 1.0, -1.0,  1.0), vec3(-1.0, -1.0,  1.0),
        vec3( 1.0, -1.0, -1.0), vec3( 1.0,  1.0, -1.0), vec3( 1.0,  1.0,  1.0),
        vec3( 1.0, -1.0, -1.0), vec3( 1.0,  1.0,  1.0), vec3( 1.0, -1.0,  1.0),
        vec3(-1.0, -1.0, -1.0), vec3(-1.0, -1.0,  1.0), vec3(-1.0,  1.0,  1.0),
        vec3(-1.0, -1.0, -1.0), vec3(-1.0,  1.0,  1.0), vec3(-1.0,  1.0, -1.0)
    );

    var p = pos[vertex_index];
    let obj = object_data.objects[instance_index];
    let obj_type = obj.params.x;

    if (obj_type == 5.0) { // MESH
        // For meshes, we use the actual vertex data, not proxy geometry.
        // The position here is a placeholder, the real mesh data is handled by mesh_pipeline_.
        var out: VertexOutput;
        out.position = vec4<f32>(0.0, 0.0, 2.0, 1.0); // Outside far plane, so it's not rendered by this pipeline.
        return out;
    }

    // Tight fit for Torus proxy hull (major radius 1.0, minor 0.4)
    if (obj_type == 3.0) {
        p.x = p.x * 1.5;
        p.z = p.z * 1.5;
        p.y = p.y * 0.5;
    }

    let world_pos = obj.model * vec4<f32>(p, 1.0);
    let clip_pos = globals.view_proj * world_pos;

    var out: VertexOutput;
    out.position = clip_pos;
    out.local_pos = p; 
    out.color = obj.color;
    out.instance_index = instance_index;
    out.world_pos = world_pos.xyz;
    
    // Correct normal transformation for meshes: transpose of inverse of model matrix
    // For non-uniform scaling, this is necessary. For other primitives, we use their analytical normals.
    if (obj_type == 5.0) {
        // Calculate inverse transpose of the model matrix (upper 3x3 part)
        let model_matrix = mat3x3<f32>(obj.model[0].xyz, obj.model[1].xyz, obj.model[2].xyz);
        let normal_matrix = transpose(inverse(model_matrix));
        out.transformed_normal = normalize(normal_matrix * in.normal);
    } else {
        // For SDF primitives, we don't use vertex normals directly here; they are computed in the fragment shader.
        // However, we still need to output a normal for the fragment shader to use if it were a rasterized primitive.
        // The transformed_normal is not used by the SDF fragment shader, but for correctness, we'll pass it.
        // If this were a rasterized mesh, it would be used.
        out.transformed_normal = normalize(vec3<f32>(0.0, 1.0, 0.0)); // Placeholder for non-mesh types
    }

    return out;
}

#include "render/scene_query_mode"
#include "render/shadows"
#include "render/lighting_utils"
#include "ray_box"

struct FragmentOutput {
    @location(0) color: vec4<f32>,
    @builtin(frag_depth) depth: f32,
};

@fragment
fn fs_main(in: VertexOutput) -> FragmentOutput {
    let obj = object_data.objects[in.instance_index];
    let obj_type = obj.params.x;

    var p: vec3<f32>;
    var normal: vec3<f32>;
    var base_color = in.color.rgb;
    let light_dir = normalize(vec3<f32>(1.0, 1.0, 1.0)); 

    if (obj_type <= 0.0) { // Raster path (legacy or generic)
        p = in.world_pos;
        // Use the transformed normal passed from the vertex shader for rasterized objects
        normal = normalize(in.transformed_normal);

        // Apply grid pattern to floor
        let uv = p.xz * 0.5;
        let grid = 0.5 + 0.5 * sin(uv.x * 3.14) * sin(uv.y * 3.14);
        let grid_val = smoothstep(0.45, 0.55, grid);
        base_color = base_color * (0.5 + 0.5 * grid_val);
    } else { // SDF path
        let ro_world = globals.camera_pos_time.xyz;
        let rd_world = normalize(in.world_pos - ro_world);
        
        // Ray-Box Intersection in local space to find tight bounds
        let ro_local = (obj.inv_model * vec4<f32>(ro_world, 1.0)).xyz;
        let rd_local = normalize((obj.inv_model * vec4<f32>(rd_world, 0.0)).xyz);
        
        // Proxy box extent (matches vs_main)
        // MESHES use obj.params.yzw for extent
        var extent = vec3<f32>(1.0);
        if (obj.params.x == 3.0) { extent = vec3<f32>(1.5, 0.5, 1.5); } // Torus
        else if (obj.params.x == 5.0) { extent = obj.params.yzw; } // MESH extent
        
        let bounds = ray_box_intersection(ro_local, rd_local, extent);

        if (!bounds.hit) { discard; }

        var t = bounds.t_entry;
        var hit = false;
        for (var i = 0; i < 64; i = i + 1) {
            let q = ro_local + rd_local * t;
            let d_local = get_dist(q, obj.params);
            if (d_local < 0.0005) { hit = true; break; }
            t = t + d_local;
            if (t > bounds.t_exit) { break; }
        }
        if (!hit) { discard; }
        
        let q_hit = ro_local + rd_local * t;
        p = (obj.model * vec4<f32>(q_hit, 1.0)).xyz; // Correct world position
        
        // Calculate normal with bump mapping
        let e = vec2<f32>(0.005, 0.0);
        let disp_strength = 0.05;
        
        let q_x1 = q_hit + e.xyy;
        let uv_x1 = vec2<f32>(atan2(q_x1.x, q_x1.z) / 6.28 + 0.5, acos(clamp(q_x1.y / length(q_x1), -1.0, 1.0)) / 3.14);
        let h_x1 = textureSample(noise_tex, noise_sampler, uv_x1).r;
        let d_x1 = get_dist(q_x1, obj.params) - disp_strength * h_x1;
        
        let q_x2 = q_hit - e.xyy;
        let uv_x2 = vec2<f32>(atan2(q_x2.x, q_x2.z) / 6.28 + 0.5, acos(clamp(q_x2.y / length(q_x2), -1.0, 1.0)) / 3.14);
        let h_x2 = textureSample(noise_tex, noise_sampler, uv_x2).r;
        let d_x2 = get_dist(q_x2, obj.params) - disp_strength * h_x2;
        
        let q_y1 = q_hit + e.yxy;
        let uv_y1 = vec2<f32>(atan2(q_y1.x, q_y1.z) / 6.28 + 0.5, acos(clamp(q_y1.y / length(q_y1), -1.0, 1.0)) / 3.14);
        let h_y1 = textureSample(noise_tex, noise_sampler, uv_y1).r;
        let d_y1 = get_dist(q_y1, obj.params) - disp_strength * h_y1;
        
        let q_y2 = q_hit - e.yxy;
        let uv_y2 = vec2<f32>(atan2(q_y2.x, q_y2.z) / 6.28 + 0.5, acos(clamp(q_y2.y / length(q_y2), -1.0, 1.0)) / 3.14);
        let h_y2 = textureSample(noise_tex, noise_sampler, uv_y2).r;
        let d_y2 = get_dist(q_y2, obj.params) - disp_strength * h_y2;
        
        let q_z1 = q_hit + e.yyx;
        let uv_z1 = vec2<f32>(atan2(q_z1.x, q_z1.z) / 6.28 + 0.5, acos(clamp(q_z1.y / length(q_z1), -1.0, 1.0)) / 3.14);
        let h_z1 = textureSample(noise_tex, noise_sampler, uv_z1).r;
        let d_z1 = get_dist(q_z1, obj.params) - disp_strength * h_z1;
        
        let q_z2 = q_hit - e.yyx;
        let uv_z2 = vec2<f32>(atan2(q_z2.x, q_z2.z) / 6.28 + 0.5, acos(clamp(q_z2.y / length(q_z2), -1.0, 1.0)) / 3.14);
        let h_z2 = textureSample(noise_tex, noise_sampler, uv_z2).r;
        let d_z2 = get_dist(q_z2, obj.params) - disp_strength * h_z2;
        
        let n_local = normalize(vec3<f32>(d_x1 - d_x2, d_y1 - d_y2, d_z1 - d_z2));
        let normal_matrix = mat3x3<f32>(obj.inv_model[0].xyz, obj.inv_model[1].xyz, obj.inv_model[2].xyz);
        normal = normalize(transpose(normal_matrix) * n_local);

        // Apply texture to SDF color
        if (in.instance_index == 0u || obj_type == 4.0) { // Floor (index 0) or PLANE
            let uv_grid = p.xz * 0.5;
            let grid = 0.5 + 0.5 * sin(uv_grid.x * 3.14) * sin(uv_grid.y * 3.14);
            let grid_val = smoothstep(0.45, 0.55, grid);
            base_color = base_color * (0.5 + 0.5 * grid_val);
        } else {
            let uv_hit = vec2<f32>(atan2(q_hit.x, q_hit.z) / 6.28 + 0.5, acos(clamp(q_hit.y / length(q_hit), -1.0, 1.0)) / 3.14);
            let tex_val = textureSample(noise_tex, noise_sampler, uv_hit).r;
            base_color = base_color * (0.7 + 0.3 * tex_val);
        }
    }

    let shadow = calc_shadow(p, light_dir, 0.05, 20.0, in.instance_index);
    let lit_color = calculate_lighting(base_color, normal, p, shadow);
    
    var out: FragmentOutput;
    out.color = vec4<f32>(lit_color, 1.0);
    
    // Calculate and write correct depth
    let clip_pos = globals.view_proj * vec4<f32>(p, 1.0);
    out.depth = clip_pos.z / clip_pos.w;
    
    return out;
}