const std = @import("std");

const core = @import("mach-core");
const gpu = core.gpu;

const zm = @import("zmath");
const vec = zm.f32x4;
const Mat = zm.Mat;

const App = @import("./main.zig");

const primitives = @import("./primitives.zig");
const VertexData = primitives.VertexData;
const PrimitiveData = primitives.PrimitiveData;

/// Holds information about how a perticular scene should be rendered.
const SceneUniformBuffer = struct {
    view_proj_matrix: Mat,
};

/// Holds information about where and how an object should be rendered.
const ObjectUniformBuffer = struct {
    model_matrix: Mat,
    color: [3]f32,
};

/// Holds data needed to render an object in a rendering pass.
const ObjectData = struct {
    /// Reference to data stored on the GPU of type `ObjectUniformBuffer`.
    uniform_buffer: *gpu.Buffer,
    /// Bind group used to associate the buffer to the `object` shader parameter.
    uniform_bind_group: *gpu.BindGroup,
    /// Reference to the primitive (shape or model) to render for this object.
    primitive: *PrimitiveData,
};

const Renderer = @This();

app: *App,

pipeline: *gpu.RenderPipeline,
scene_uniform_buffer: *gpu.Buffer,
scene_uniform_bind_group: *gpu.BindGroup,

depth_texture: ?*gpu.Texture = null,
depth_texture_view: ?*gpu.TextureView = null,

primitive_data: []PrimitiveData,
object_data: []ObjectData,

pub fn init(app: *App) !*Renderer {
    const shader_module = core.device.createShaderModuleWGSL("shader.wgsl", @embedFile("shader.wgsl"));
    defer shader_module.release();

    // Set up rendering pipeline.
    const pipeline = core.device.createRenderPipeline(&.{
        .vertex = gpu.VertexState.init(.{
            .module = shader_module,
            .entry_point = "vertex_main",
            .buffers = &.{
                gpu.VertexBufferLayout.init(.{
                    .array_stride = @sizeOf(VertexData),
                    .step_mode = .vertex,
                    .attributes = &.{
                        .{ .format = .float32x3, .shader_location = 0, .offset = @offsetOf(VertexData, "position") },
                    },
                }),
            },
        }),
        .primitive = .{
            .topology = .triangle_list,
            .front_face = .ccw,
            .cull_mode = .back,
        },
        .depth_stencil = &.{
            .format = .depth24_plus,
            .depth_write_enabled = .true,
            .depth_compare = .less,
        },
        .fragment = &gpu.FragmentState.init(.{
            .module = shader_module,
            .entry_point = "frag_main",
            .targets = &.{.{ .format = core.descriptor.format }},
        }),
    });

    // Set up scene related uniform buffers and bind groups.
    const scene_uniform = createAndWriteUniformBuffer(
        pipeline.getBindGroupLayout(0),
        SceneUniformBuffer{ .view_proj_matrix = zm.identity() },
    );

    // Set up the primitives we want to render.
    //
    // Using `dupe` to allocate a slice here allows easily adjusting the
    // primitives to use, without changing the type of `primitive_data`.
    const primitive_data = try app.allocator.dupe(PrimitiveData, &.{
        // primitives.createTrianglePrimitive(1.0),
        // primitives.createSquarePrimitive(0.8),
        // primitives.createCirclePrimitive(0.5, 24),
        primitives.createCubePrimitive(0.65),
        primitives.createPyramidPrimitive(0.75),
    });

    // Set up object related uniform buffers and bind groups.
    // This uploads data to the GPU about all the object we
    // want to render, such as their location and color.
    const grid_size = 8;

    // Allocate a slice to store as many ObjectData as we want to create.
    //
    // Using a slice instead of an array means that we could change how
    // many object we want to render at compile time, however it requires
    // allocating, and later freeing, memory to store the slice.
    const object_data = try app.allocator.alloc(ObjectData, grid_size * grid_size);

    // Note that for loops in Zig are a little different than you might
    // know from other languages. They only look over arrays, slices,
    // tuples and ranges, potentially multiple at once.
    for (object_data, 0..) |*object, i| {
        const grid_max: f32 = @floatFromInt(grid_size - 1);
        const x = @as(f32, @floatFromInt(i % grid_size)) / grid_max;
        const z = @as(f32, @floatFromInt(i / grid_size)) / grid_max;

        const rotation = zm.rotationY(std.math.tau * (x + z) / 2.0);
        const translation = zm.translation((x - 0.5) * grid_size, 0, (z - 0.5) * grid_size);
        const model_matrix = zm.mul(rotation, translation);

        // Make the object have a color depending on its location in the grid.
        // These values are layed out so each corner is red, green, blue and black.
        const color = .{
            std.math.clamp(1.0 - x - z, 0.0, 1.0),
            std.math.clamp(x - z, 0.0, 1.0),
            std.math.clamp(z - x, 0.0, 1.0),
        };

        const object_uniform = createAndWriteUniformBuffer(
            pipeline.getBindGroupLayout(1),
            ObjectUniformBuffer{
                .model_matrix = zm.transpose(model_matrix),
                .color = color,
            },
        );

        // Pick a "random" primitive to use for this object.
        const primitive_index = app.random.int(usize) % primitive_data.len;
        const primitive = &primitive_data[primitive_index];

        // The `*object` syntax gets us a pointer to each element in the
        // `object_data` slice, allowing us to override it within the loop.
        object.* = .{
            .uniform_buffer = object_uniform.buffer,
            .uniform_bind_group = object_uniform.bind_group,
            .primitive = primitive,
        };
    }

    const result = try app.allocator.create(Renderer);
    result.* = .{
        .app = app,
        .pipeline = pipeline,
        .scene_uniform_buffer = scene_uniform.buffer,
        .scene_uniform_bind_group = scene_uniform.bind_group,
        .primitive_data = primitive_data,
        .object_data = object_data,
    };

    // Initialize the depth texture.
    // This is called also whenever the window is resized.
    result.recreateDepthTexture();

    return result;
}

pub fn deinit(self: *Renderer) void {
    // Using `defer` here, so we can specify them
    // in the order they were created in `init`.
    defer self.app.allocator.destroy(self);

    defer self.pipeline.release();
    defer self.scene_uniform_buffer.release();
    defer self.scene_uniform_bind_group.release();

    defer self.app.allocator.free(self.primitive_data);
    defer for (self.primitive_data) |p| {
        p.vertex_buffer.release();
        p.index_buffer.release();
    };
    defer self.app.allocator.free(self.object_data);
    defer for (self.object_data) |o| {
        o.uniform_buffer.release();
        o.uniform_bind_group.release();
    };

    defer if (self.depth_texture) |t| t.release();
    defer if (self.depth_texture_view) |v| v.release();
}

pub fn resize(self: *Renderer) void {
    // Recreate depth texture with the proper size, otherwise
    // the application may crash when the window is resized.
    self.recreateDepthTexture();
}

pub fn update(self: *Renderer) void {
    // Set up a view matrix from the camera transform.
    // This moves everything to be relative to the camera.
    // TODO: Actually implement camera transform instead of hardcoding a look-at matrix.
    // const view_matrix = zm.inverse(app.camera_transform);
    const time = self.app.app_timer.read();
    const camera_distance = 8.0;
    const x = @cos(time * std.math.tau / 20) * camera_distance;
    const z = @sin(time * std.math.tau / 20) * camera_distance;
    const camera_pos = vec(x, 2.0, z, 1.0);
    const view_matrix = zm.lookAtLh(camera_pos, vec(0, 0, 0, 1), vec(0, 1, 0, 1));

    // Set up a projection matrix using the size of the window.
    // The perspective projection will make things further away appear smaller.
    const width: f32 = @floatFromInt(core.descriptor.width);
    const height: f32 = @floatFromInt(core.descriptor.height);
    const field_of_view = std.math.degreesToRadians(f32, 45.0);
    const proj_matrix = zm.perspectiveFovLh(field_of_view, width / height, 0.05, 80.0);

    const view_proj_matrix = zm.mul(view_matrix, proj_matrix);

    // Get back buffer texture to render to.
    const back_buffer_view = core.swap_chain.getCurrentTextureView().?;
    defer back_buffer_view.release();
    // Once rendering is done (hence `defer`), swap back buffer to the front to display.
    defer core.swap_chain.present();

    const render_pass_info = gpu.RenderPassDescriptor.init(.{
        .color_attachments = &.{.{
            .view = back_buffer_view,
            .clear_value = std.mem.zeroes(gpu.Color),
            .load_op = .clear,
            .store_op = .store,
        }},
        .depth_stencil_attachment = &.{
            .view = self.depth_texture_view.?,
            .depth_load_op = .clear,
            .depth_store_op = .store,
            .depth_clear_value = 1.0,
        },
    });

    // Create a `WGPUCommandEncoder` which provides an interface for recording GPU commands.
    const encoder = core.device.createCommandEncoder(null);
    defer encoder.release();

    // Write to the scene uniform buffer for this set of commands.
    encoder.writeBuffer(self.scene_uniform_buffer, 0, &[_]SceneUniformBuffer{.{
        // All matrices the GPU has to work with need to be transposed,
        // because WebGPU uses column-major matrices while zmath is row-major.
        .view_proj_matrix = zm.transpose(view_proj_matrix),
    }});

    {
        const pass = encoder.beginRenderPass(&render_pass_info);
        defer pass.release();
        defer pass.end();

        pass.setPipeline(self.pipeline);
        pass.setBindGroup(0, self.scene_uniform_bind_group, &.{});

        for (self.object_data) |object| {
            // Set the vertex and index buffer used to render this object
            // to the primitive it wants to use (either triangle or square).
            const prim = object.primitive;
            pass.setVertexBuffer(0, prim.vertex_buffer, 0, prim.vertex_count * @sizeOf(VertexData));
            pass.setIndexBuffer(prim.index_buffer, .uint32, 0, prim.index_count * @sizeOf(u32));

            // Set the bind group for an object we want to render.
            pass.setBindGroup(1, object.uniform_bind_group, &.{});

            // Draw a number of triangles as specified in the index buffer.
            pass.drawIndexed(prim.index_count, 1, 0, 0, 0);
        }
    }

    // Finish recording commands, creating a `WGPUCommandBuffer`.
    var command = encoder.finish(null);
    defer command.release();

    // Submit the command(s) to the GPU.
    core.queue.submit(&.{command});
}

/// Creates a depth texture. This is used to ensure that when things are
/// rendered, an object behind another won't draw over one in front, simply
/// because it was rendered at a later point in time.
pub fn recreateDepthTexture(self: *Renderer) void {
    // Release previous depth butter and view, if any.
    if (self.depth_texture) |t| t.release();
    if (self.depth_texture_view) |v| v.release();

    self.depth_texture = core.device.createTexture(&.{
        .usage = .{ .render_attachment = true },
        .size = .{ .width = core.descriptor.width, .height = core.descriptor.height },
        .format = .depth24_plus,
    });
    self.depth_texture_view = self.depth_texture.?.createView(null);
}

/// Creates a buffer on the GPU to store uniform parameter information as
/// well as a bind group with the specified layout pointing to that buffer.
/// Additionally, immediately fills the buffer with the provided data.
pub fn createAndWriteUniformBuffer(
    layout: *gpu.BindGroupLayout,
    data: anytype,
) struct {
    buffer: *gpu.Buffer,
    bind_group: *gpu.BindGroup,
} {
    const T = @TypeOf(data);
    const usage = gpu.Buffer.UsageFlags{ .copy_dst = true, .uniform = true };
    const buffer = createAndWriteBuffer(T, &.{data}, usage);

    // "Bind groups" are used to associate data from buffers with shader parameters.
    // So for example the `scene_uniform_bind_group` is accessible via `scene` in our shader.
    // Essentially, buffer = data, and bind group = binding parameter to that data.
    const bind_group_entry = gpu.BindGroup.Entry.buffer(0, buffer, 0, @sizeOf(T));
    const bind_group_desc = gpu.BindGroup.Descriptor.init(.{ .layout = layout, .entries = &.{bind_group_entry} });
    const bind_group = core.device.createBindGroup(&bind_group_desc);

    return .{ .buffer = buffer, .bind_group = bind_group };
}

/// Creates a buffer on the GPU with the specified usage
/// flags and immediately fills it with the provided data.
pub fn createAndWriteBuffer(
    comptime T: type,
    data: []const T,
    usage: gpu.Buffer.UsageFlags,
) *gpu.Buffer {
    const buffer = core.device.createBuffer(&.{
        .size = data.len * @sizeOf(T),
        .usage = usage,
        .mapped_at_creation = .false,
    });
    core.queue.writeBuffer(buffer, 0, data);
    return buffer;
}