const std = @import("std");
const GeneralPurposeAllocator = std.heap.GeneralPurposeAllocator(.{});

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

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

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

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

/// Holds data on what is 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,
    /// Index into `primitives` to specify which primitive to render.
    primitive_index: usize,
};

/// Describes the layout of each vertex that a primitive is made of.
const VertexData = struct {
    position: [3]f32,
};

/// Contains the data to render a primitive (3D shape or model).
const PrimitiveData = struct {
    /// Vertices describe the "points" that a primitive is made out of.
    /// This buffer is of type `[]VertexData`.
    vertex_buffer: *gpu.Buffer,
    vertex_count: u32,

    /// Indices describe what vertices make up the triangles in a primitive.
    /// This buffer is of type `[]u32`.
    index_buffer: *gpu.Buffer,
    index_count: u32,

    // For example, `vertex_buffer` may have 4 points defining a square, but
    // since it needs to be rendered using 2 triangles, `index_buffer` will
    // contain 6 entries, `0, 1, 2` and `3, 2, 1` making up one triangle each.
};

pub const App = @This();

gpa: GeneralPurposeAllocator,
allocator: std.mem.Allocator,
random: std.rand.Random,

app_timer: core.Timer,
title_timer: core.Timer,

depth_texture: *gpu.Texture,
depth_texture_view: *gpu.TextureView,
pipeline: *gpu.RenderPipeline,
scene_uniform_buffer: *gpu.Buffer,
scene_uniform_bind_group: *gpu.BindGroup,
primitives: []PrimitiveData,
object_data: []ObjectData,

pub fn init(app: *App) !void {
    try core.init(.{});

    // Set up a "general purpose allocator" that will handle allocations for
    // the lifetime of the application, for which a more specific allocation
    // strategy is not necessary.
    //
    // Here, `gpa` is an instance of this allocator, which handles the
    // allocation logic, while `allocator` is the interface through which
    // functions such as `alloc` and `free` are called.
    app.gpa = GeneralPurposeAllocator{};
    app.allocator = app.gpa.allocator();

    // Create a pseudo-random number generator, but initialize it with
    // a constant seed so we always get the same result when launching.
    var prng = std.rand.DefaultPrng.init(0);
    app.random = prng.random();

    app.app_timer = try core.Timer.start();
    app.title_timer = try core.Timer.start();

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

    const shader_module = core.device.createShaderModuleWGSL("shader.wgsl", @embedFile("shader.wgsl"));
    defer shader_module.release();

    // Set up rendering pipeline.
    app.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 result = createAndWriteUniformBuffer(
            app.pipeline.getBindGroupLayout(0),
            SceneUniformBuffer{ .view_proj_matrix = zm.identity() },
        );

        app.scene_uniform_buffer = result.buffer;
        app.scene_uniform_bind_group = result.bind_group;
    }

    // Set up the primitives we want to render.
    app.primitives = try app.allocator.alloc(PrimitiveData, 2);
    // Triangle
    app.primitives[0] = createPrimitive(
        &.{
            // zig fmt: off
            .{ .position = .{  0.0,  0.5, 0.0 } },
            .{ .position = .{  0.5, -0.5, 0.0 } },
            .{ .position = .{ -0.5, -0.5, 0.0 } },
            // zig fmt: on
        },
        // Note that the back faces of triangles are "culled", and thus not visible.
        // We need to take care to specify the vertices in counter-clock orientation.
        &.{
            0, 1, 2,
        },
    );
    // Square
    app.primitives[1] = createPrimitive(
        // A square is made up of 4 vertices.
        //   0--2
        //   |  |
        //   |  |
        //   1--3
        &.{
            // zig fmt: off
            .{ .position = .{ -0.5, -0.5, 0.0 } },
            .{ .position = .{ -0.5,  0.5, 0.0 } },
            .{ .position = .{  0.5, -0.5, 0.0 } },
            .{ .position = .{  0.5,  0.5, 0.0 } },
            // zig fmt: on
        },
        // But it has to be split up into 2 triangles,
        // specified in a counter-clockwise orientation.
        //   0--2  4
        //   | /  /|
        //   |/  / |
        //   1  5--3
        &.{
            0, 1, 2,
            3, 2, 1,
        },
    );

    // 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.
        app.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 (app.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 result = createAndWriteUniformBuffer(
                app.pipeline.getBindGroupLayout(1),
                ObjectUniformBuffer{
                    .model_matrix = zm.transpose(model_matrix),
                    .color = color,
                },
            );

            // 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 = result.buffer,
                .uniform_bind_group = result.bind_group,
                .primitive_index = app.random.int(usize) % app.primitives.len,
            };
        }
    }
}

/// 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.
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.
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;
}

/// Creates a primitive from the provided vertices and indices,
/// and uploads the buffers necessary to render it to the GPU.
fn createPrimitive(
    vertices: []const VertexData,
    indices: []const u32,
) PrimitiveData {
    return .{
        .vertex_buffer = createAndWriteBuffer(VertexData, vertices, .{ .vertex = true, .copy_dst = true }),
        .vertex_count = @intCast(vertices.len),
        .index_buffer = createAndWriteBuffer(u32, indices, .{ .index = true, .copy_dst = true }),
        .index_count = @intCast(indices.len),
    };
}

pub fn deinit(app: *App) void {
    // Using `defer` here, so we can specify them
    // in the order they were created in `init`.
    defer core.deinit();
    defer _ = app.gpa.deinit(); // TODO: Check for memory leaks?
    defer app.depth_texture.release();
    defer app.depth_texture_view.release();
    defer app.pipeline.release();
    defer app.scene_uniform_buffer.release();
    defer app.scene_uniform_bind_group.release();
    defer app.allocator.free(app.primitives);
    defer for (app.primitives) |p| {
        p.vertex_buffer.release();
        p.index_buffer.release();
    };
    defer app.allocator.free(app.object_data);
    defer for (app.object_data) |o| {
        o.uniform_buffer.release();
        o.uniform_bind_group.release();
    };
}

pub fn update(app: *App) !bool {
    var iter = core.pollEvents();
    while (iter.next()) |event| {
        switch (event) {
            .close => return true,
            else => {},
        }
    }

    // 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 = 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 = app.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(app.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(app.pipeline);
        pass.setBindGroup(0, app.scene_uniform_bind_group, &.{});

        for (app.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 primitive_index = object.primitive_index;
            const primitive = app.primitives[primitive_index];
            pass.setVertexBuffer(0, primitive.vertex_buffer, 0, primitive.vertex_count * @sizeOf(VertexData));
            pass.setIndexBuffer(primitive.index_buffer, .uint32, 0, primitive.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(primitive.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});

    // Update the window title to show FPS and input frequency.
    if (app.title_timer.read() >= 1.0) {
        app.title_timer.reset();
        try core.printTitle("Triangle [ {d}fps ] [ Input {d}hz ]", .{ core.frameRate(), core.inputRate() });
    }

    return false;
}