use crate::{CompositorGpuHint, WgpuAtlas, WgpuContext, WgpuDeviceRequirements};
use bytemuck::{Pod, Zeroable};
use gpui::{
	AtlasTextureId, BackdropFilter, Background, Bounds, DevicePixels, FilterBoundary, GpuSpecs,
	MonochromeSprite, PaintSurface, Path, Point, PolychromeSprite, PrimitiveBatch, Quad,
	ScaledFilter, ScaledPixels, Scene, Shadow, Size, SubpixelSprite, Underline,
	get_gamma_correction_ratios,
};
use log::warn;

/// The largest blur radius in a scene-space filter chain, in device pixels — used to size the
/// blur kernel and the dilated region the blur passes are scissored to.
///
/// The `match` is exhaustive on purpose: adding a [`ScaledFilter`] variant breaks it here,
/// forcing this backend to handle (or deliberately ignore) the new filter rather than silently
/// dropping it.
fn max_blur_radius(filters: &[ScaledFilter]) -> f32 {
	filters.iter().fold(0.0, |acc, filter| match filter {
		ScaledFilter::Blur(radius) => acc.max(radius.0),
	})
}
#[cfg(not(target_family = "wasm"))]
use raw_window_handle::{HasDisplayHandle, HasWindowHandle};
use std::cell::RefCell;
use std::num::NonZeroU64;
use std::rc::Rc;
use std::sync::{Arc, Mutex};

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GlobalParams {
	viewport_size: [f32; 2],
	premultiplied_alpha: u32,
	pad: u32,
}

#[repr(C)]
#[derive(Clone, Copy, Default, Pod, Zeroable)]
struct PodBounds {
	origin: [f32; 2],
	size: [f32; 2],
}

impl From<Bounds<ScaledPixels>> for PodBounds {
	fn from(bounds: Bounds<ScaledPixels>) -> Self {
		Self {
			origin: [bounds.origin.x.0, bounds.origin.y.0],
			size: [bounds.size.width.0, bounds.size.height.0],
		}
	}
}

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct SurfaceParams {
	bounds: PodBounds,
	content_mask: PodBounds,
}

/// Uniform passed to the blur pipelines. The same struct drives the downsample, separable
/// gaussian, and composite passes; fields not relevant to a given pass are left zero.
#[repr(C)]
#[derive(Clone, Copy, Default, Pod, Zeroable)]
struct BlurParams {
	/// Composite target rectangle, in device pixels (composite pass only).
	bounds: PodBounds,
	/// Clip rectangle, in device pixels (composite pass only).
	content_mask: PodBounds,
	/// Rounded-corner radii (tl, tr, br, bl), in device pixels (composite pass only).
	corner_radii: [f32; 4],
	/// Per-tap sampling step in UV space (gaussian passes only): (1/width, 0) or (0, 1/height).
	direction: [f32; 2],
	/// Gaussian sigma, in the (half-resolution) blur texture's pixels.
	sigma: f32,
	/// Element opacity, multiplied into the composited result.
	opacity: f32,
	/// Number of taps to each side of center (gaussian passes only).
	tap_count: f32,
	/// Spacing between taps in pixels; >1 lets `tap_count` taps span very large radii without
	/// truncating the gaussian (see #6 in review).
	tap_step: f32,
	/// 1.0 to clip the composite to the rounded rect (backdrop — the panel has a defined shape),
	/// 0.0 to let the blurred result fade out on its own (content `filter` — it bleeds past the
	/// element bounds like CSS, so the fade isn't sharply truncated at the box edge).
	clip_rounded: f32,
	/// 1.0 = snapped 2:1 box downsample (anchor the half-res grid to a fixed 2px grid at the
	/// origin, so a stationary element blurs identically at every window size); 0.0 = 1:1 copy
	/// (the scene blit, which must not downsample). Downsample pass only.
	downsample: f32,
}

#[repr(C)]
#[derive(Clone, Copy, Pod, Zeroable)]
struct GammaParams {
	gamma_ratios: [f32; 4],
	grayscale_enhanced_contrast: f32,
	subpixel_enhanced_contrast: f32,
	is_bgr: u32,
	_pad: u32,
}

#[derive(Clone, Debug)]
#[repr(C)]
struct PathSprite {
	bounds: Bounds<ScaledPixels>,
}

#[derive(Clone, Debug)]
#[repr(C)]
struct PathRasterizationVertex {
	xy_position: Point<ScaledPixels>,
	st_position: Point<f32>,
	color: Background,
	bounds: Bounds<ScaledPixels>,
}

pub struct WgpuSurfaceConfig {
	pub size: Size<DevicePixels>,
	pub transparent: bool,
	/// Preferred presentation mode. When `Some`, the renderer will use this
	/// mode if supported by the surface, falling back to `Fifo`.
	/// When `None`, defaults to `Fifo` (VSync).
	///
	/// Mobile platforms may prefer `Mailbox` (triple-buffering) to avoid
	/// blocking in `get_current_texture()` during lifecycle transitions.
	pub preferred_present_mode: Option<wgpu::PresentMode>,
}

struct WgpuPipelines {
	quads: wgpu::RenderPipeline,
	shadows: wgpu::RenderPipeline,
	path_rasterization: wgpu::RenderPipeline,
	paths: wgpu::RenderPipeline,
	underlines: wgpu::RenderPipeline,
	mono_sprites: wgpu::RenderPipeline,
	subpixel_sprites: Option<wgpu::RenderPipeline>,
	poly_sprites: wgpu::RenderPipeline,
	#[allow(dead_code)]
	surfaces: wgpu::RenderPipeline,
	/// Copies a source texture into the (smaller) target with one bilinear tap. Used both to
	/// downsample the scene into the half-resolution blur texture and to blit the offscreen
	/// scene into the swapchain at the end of the frame.
	blur_downsample: wgpu::RenderPipeline,
	/// One axis of a separable gaussian blur; direction is supplied per draw via [`BlurParams`].
	blur: wgpu::RenderPipeline,
	/// Composites a blurred texture into a rounded rectangle (with clip + opacity).
	blur_composite: wgpu::RenderPipeline,
}

struct WgpuBindGroupLayouts {
	globals: wgpu::BindGroupLayout,
	instances: wgpu::BindGroupLayout,
	instances_with_texture: wgpu::BindGroupLayout,
	surfaces: wgpu::BindGroupLayout,
	blur: wgpu::BindGroupLayout,
}

/// Shared GPU context reference, used to coordinate device recovery across multiple windows.
pub type GpuContext = Rc<RefCell<Option<WgpuContext>>>;

/// GPU resources that must be dropped together during device recovery.
struct WgpuResources {
	device: Arc<wgpu::Device>,
	queue: Arc<wgpu::Queue>,
	surface: wgpu::Surface<'static>,
	pipelines: WgpuPipelines,
	bind_group_layouts: WgpuBindGroupLayouts,
	atlas_sampler: wgpu::Sampler,
	surface_sampler: wgpu::Sampler,
	#[allow(dead_code)]
	surface_uniform_buffer: wgpu::Buffer,
	/// One reused uniform buffer holding [`BlurParams`] for every blur pass in a frame, each at a
	/// distinct (alignment-strided) offset. Avoids allocating a buffer per pass; distinct offsets
	/// mean `write_buffer`'s last-write-at-submit semantics don't clobber earlier passes.
	blur_params_buffer: wgpu::Buffer,
	globals_buffer: wgpu::Buffer,
	globals_bind_group: wgpu::BindGroup,
	path_globals_bind_group: wgpu::BindGroup,
	instance_buffer: wgpu::Buffer,
	path_intermediate_texture: Option<wgpu::Texture>,
	path_intermediate_view: Option<wgpu::TextureView>,
	path_msaa_texture: Option<wgpu::Texture>,
	path_msaa_view: Option<wgpu::TextureView>,
	/// Blur offscreen targets. Allocated lazily (only when a frame actually uses a blur filter)
	/// so apps that never blur pay no extra VRAM. `None`/empty until first use.
	///
	/// Full-resolution offscreen color target the scene is rendered into so that blur passes
	/// can sample already-painted content; blitted to the swapchain at the end of the frame.
	scene_color_texture: Option<wgpu::Texture>,
	scene_color_view: Option<wgpu::TextureView>,
	/// Half-resolution ping/pong targets for the downsample + separable gaussian passes.
	blur_ping_texture: Option<wgpu::Texture>,
	blur_ping_view: Option<wgpu::TextureView>,
	blur_pong_texture: Option<wgpu::Texture>,
	blur_pong_view: Option<wgpu::TextureView>,
	/// Full-resolution offscreen targets a content-filter (`filter`) group renders into before
	/// being blurred and composited back. One per nesting level (indexed by depth) so nested
	/// content blurs isolate correctly, up to [`MAX_FILTER_DEPTH`]; deeper nests render inline.
	group_textures: Vec<wgpu::Texture>,
	group_views: Vec<wgpu::TextureView>,
}

impl WgpuResources {
	fn invalidate_intermediate_textures(&mut self) {
		self.path_intermediate_texture = None;
		self.path_intermediate_view = None;
		self.path_msaa_texture = None;
		self.path_msaa_view = None;
		self.scene_color_texture = None;
		self.scene_color_view = None;
		self.blur_ping_texture = None;
		self.blur_ping_view = None;
		self.blur_pong_texture = None;
		self.blur_pong_view = None;
		self.group_textures.clear();
		self.group_views.clear();
	}
}

/// Number of content-filter (`filter`) nesting levels that get their own isolated group texture.
/// Two covers the realistic "a blurred element inside another blurred element" case; deeper nests
/// render inline (unblurred at the inner level) rather than allocating unbounded VRAM.
const MAX_FILTER_DEPTH: usize = 2;

/// Number of [`BlurParams`] slots in the shared blur-params buffer (one per blur pass per frame).
/// Each frame uses 4 passes per backdrop/group plus one blit; 256 covers dozens of filters.
const BLUR_PARAMS_SLOTS: u64 = 256;

pub struct WgpuRenderer {
	/// Shared GPU context for device recovery coordination (unused on WASM).
	#[allow(dead_code)]
	context: Option<GpuContext>,
	/// Compositor GPU hint for adapter selection (unused on WASM).
	#[allow(dead_code)]
	compositor_gpu: Option<CompositorGpuHint>,
	/// Application-requested extra wgpu features/limits, stored for device recovery.
	#[allow(dead_code)]
	extra_requirements: Option<WgpuDeviceRequirements>,
	resources: Option<WgpuResources>,
	surface_config: wgpu::SurfaceConfiguration,
	atlas: Arc<WgpuAtlas>,
	path_globals_offset: u64,
	gamma_offset: u64,
	instance_buffer_capacity: u64,
	max_buffer_size: u64,
	storage_buffer_alignment: u64,
	/// Stride between [`BlurParams`] slots in `blur_params_buffer`, and a per-frame bump cursor
	/// (in slots) handed out to blur passes. Cell so the `&self` blur helpers can advance it.
	blur_params_stride: u64,
	blur_params_slot: std::cell::Cell<u64>,
	rendering_params: RenderingParameters,
	is_bgr: bool,
	dual_source_blending: bool,
	adapter_info: wgpu::AdapterInfo,
	transparent_alpha_mode: wgpu::CompositeAlphaMode,
	opaque_alpha_mode: wgpu::CompositeAlphaMode,
	max_texture_size: u32,
	last_error: Arc<Mutex<Option<String>>>,
	failed_frame_count: u32,
	device_lost: std::sync::Arc<std::sync::atomic::AtomicBool>,
	surface_configured: bool,
	needs_redraw: bool,
}

impl WgpuRenderer {
	fn resources(&self) -> &WgpuResources {
		self.resources
			.as_ref()
			.expect("GPU resources not available")
	}

	fn resources_mut(&mut self) -> &mut WgpuResources {
		self.resources
			.as_mut()
			.expect("GPU resources not available")
	}

	/// Creates a new WgpuRenderer from raw window handles.
	///
	/// The `gpu_context` is a shared reference that coordinates GPU context across
	/// multiple windows. The first window to create a renderer will initialize the
	/// context; subsequent windows will share it.
	///
	/// # Safety
	/// The caller must ensure that the window handle remains valid for the lifetime
	/// of the returned renderer.
	#[cfg(not(target_family = "wasm"))]
	pub fn new<W>(
		gpu_context: GpuContext,
		window: &W,
		config: WgpuSurfaceConfig,
		compositor_gpu: Option<CompositorGpuHint>,
		extra_requirements: Option<WgpuDeviceRequirements>,
	) -> anyhow::Result<Self>
	where
		W: HasWindowHandle + HasDisplayHandle + std::fmt::Debug + Send + Sync + Clone + 'static,
	{
		let window_handle = window
			.window_handle()
			.map_err(|e| anyhow::anyhow!("Failed to get window handle: {e}"))?;

		let target = wgpu::SurfaceTargetUnsafe::RawHandle {
			// Fall back to the display handle already provided via InstanceDescriptor::display.
			raw_display_handle: None,
			raw_window_handle: window_handle.as_raw(),
		};

		// Use the existing context's instance if available, otherwise create a new one.
		// The surface must be created with the same instance that will be used for
		// adapter selection, otherwise wgpu will panic.
		let instance = gpu_context
			.borrow()
			.as_ref()
			.map(|ctx| ctx.instance.clone())
			.unwrap_or_else(|| WgpuContext::instance(Box::new(window.clone())));

		// Safety: The caller guarantees that the window handle is valid for the
		// lifetime of this renderer. In practice, the RawWindow struct is created
		// from the native window handles and the surface is dropped before the window.
		let surface = unsafe {
			instance
				.create_surface_unsafe(target)
				.map_err(|e| anyhow::anyhow!("Failed to create surface: {e}"))?
		};

		let mut ctx_ref = gpu_context.borrow_mut();
		let context = match ctx_ref.as_mut() {
			Some(context) => {
				context.check_compatible_with_surface(&surface)?;
				context
			}
			None => ctx_ref.insert(WgpuContext::new(
				instance,
				&surface,
				compositor_gpu,
				extra_requirements.as_ref(),
			)?),
		};

		let atlas = Arc::new(WgpuAtlas::from_context(context));

		Self::new_internal(
			Some(Rc::clone(&gpu_context)),
			context,
			surface,
			config,
			compositor_gpu,
			extra_requirements,
			atlas,
		)
	}

	#[cfg(target_family = "wasm")]
	pub fn new_from_canvas(
		context: &WgpuContext,
		canvas: &web_sys::HtmlCanvasElement,
		config: WgpuSurfaceConfig,
	) -> anyhow::Result<Self> {
		let surface = context
			.instance
			.create_surface(wgpu::SurfaceTarget::Canvas(canvas.clone()))
			.map_err(|e| anyhow::anyhow!("Failed to create surface: {e}"))?;

		let atlas = Arc::new(WgpuAtlas::from_context(context));

		Self::new_internal(None, context, surface, config, None, None, atlas)
	}

	fn new_internal(
		gpu_context: Option<GpuContext>,
		context: &WgpuContext,
		surface: wgpu::Surface<'static>,
		config: WgpuSurfaceConfig,
		compositor_gpu: Option<CompositorGpuHint>,
		extra_requirements: Option<WgpuDeviceRequirements>,
		atlas: Arc<WgpuAtlas>,
	) -> anyhow::Result<Self> {
		let surface_caps = surface.get_capabilities(&context.adapter);
		let preferred_formats = [
			wgpu::TextureFormat::Bgra8Unorm,
			wgpu::TextureFormat::Rgba8Unorm,
		];
		let surface_format = preferred_formats
			.iter()
			.find(|f| surface_caps.formats.contains(f))
			.copied()
			.or_else(|| surface_caps.formats.iter().find(|f| !f.is_srgb()).copied())
			.or_else(|| surface_caps.formats.first().copied())
			.ok_or_else(|| {
				anyhow::anyhow!(
					"Surface reports no supported texture formats for adapter {:?}",
					context.adapter.get_info().name
				)
			})?;

		let pick_alpha_mode =
			|preferences: &[wgpu::CompositeAlphaMode]| -> anyhow::Result<wgpu::CompositeAlphaMode> {
				preferences
					.iter()
					.find(|p| surface_caps.alpha_modes.contains(p))
					.copied()
					.or_else(|| surface_caps.alpha_modes.first().copied())
					.ok_or_else(|| {
						anyhow::anyhow!(
							"Surface reports no supported alpha modes for adapter {:?}",
							context.adapter.get_info().name
						)
					})
			};

		let transparent_alpha_mode = pick_alpha_mode(&[
			wgpu::CompositeAlphaMode::PreMultiplied,
			wgpu::CompositeAlphaMode::Inherit,
		])?;

		let opaque_alpha_mode = pick_alpha_mode(&[
			wgpu::CompositeAlphaMode::Opaque,
			wgpu::CompositeAlphaMode::Inherit,
		])?;

		let alpha_mode = if config.transparent {
			transparent_alpha_mode
		} else {
			opaque_alpha_mode
		};

		let device = Arc::clone(&context.device);
		let max_texture_size = device.limits().max_texture_dimension_2d;

		let requested_width = config.size.width.0 as u32;
		let requested_height = config.size.height.0 as u32;
		let clamped_width = requested_width.min(max_texture_size);
		let clamped_height = requested_height.min(max_texture_size);

		if clamped_width != requested_width || clamped_height != requested_height {
			warn!(
				"Requested surface size ({}, {}) exceeds maximum texture dimension {}. \
                 Clamping to ({}, {}). Window content may not fill the entire window.",
				requested_width, requested_height, max_texture_size, clamped_width, clamped_height
			);
		}

		let surface_config = wgpu::SurfaceConfiguration {
			usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
			format: surface_format,
			width: clamped_width.max(1),
			height: clamped_height.max(1),
			present_mode: config
				.preferred_present_mode
				.filter(|mode| surface_caps.present_modes.contains(mode))
				.unwrap_or(wgpu::PresentMode::Fifo),
			desired_maximum_frame_latency: 2,
			alpha_mode,
			view_formats: vec![],
		};
		// Configure the surface immediately. The adapter selection process already validated
		// that this adapter can successfully configure this surface.
		surface.configure(&context.device, &surface_config);

		let queue = Arc::clone(&context.queue);
		let dual_source_blending = context.supports_dual_source_blending();

		let rendering_params = RenderingParameters::new(&context.adapter, surface_format);
		let bind_group_layouts = Self::create_bind_group_layouts(&device);
		let pipelines = Self::create_pipelines(
			&device,
			&bind_group_layouts,
			surface_format,
			alpha_mode,
			rendering_params.path_sample_count,
			dual_source_blending,
		);

		let atlas_sampler = device.create_sampler(&wgpu::SamplerDescriptor {
			label: Some("atlas_sampler"),
			mag_filter: wgpu::FilterMode::Linear,
			min_filter: wgpu::FilterMode::Linear,
			..Default::default()
		});

		let surface_sampler = device.create_sampler(&wgpu::SamplerDescriptor {
			label: Some("surface_sampler"),
			mag_filter: wgpu::FilterMode::Linear,
			min_filter: wgpu::FilterMode::Linear,
			..Default::default()
		});

		let surface_uniform_buffer = device.create_buffer(&wgpu::BufferDescriptor {
			label: Some("surface_uniform_buffer"),
			size: std::mem::size_of::<SurfaceParams>() as u64,
			usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
			mapped_at_creation: false,
		});

		let uniform_alignment = device.limits().min_uniform_buffer_offset_alignment as u64;
		// Shared blur-params buffer: BLUR_PARAMS_SLOTS slots, each one alignment stride apart.
		let blur_params_stride =
			(std::mem::size_of::<BlurParams>() as u64).next_multiple_of(uniform_alignment);
		let blur_params_buffer = device.create_buffer(&wgpu::BufferDescriptor {
			label: Some("blur_params_buffer"),
			size: blur_params_stride * BLUR_PARAMS_SLOTS,
			usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
			mapped_at_creation: false,
		});

		let globals_size = std::mem::size_of::<GlobalParams>() as u64;
		let gamma_size = std::mem::size_of::<GammaParams>() as u64;
		let path_globals_offset = globals_size.next_multiple_of(uniform_alignment);
		let gamma_offset = (path_globals_offset + globals_size).next_multiple_of(uniform_alignment);

		let globals_buffer = device.create_buffer(&wgpu::BufferDescriptor {
			label: Some("globals_buffer"),
			size: gamma_offset + gamma_size,
			usage: wgpu::BufferUsages::UNIFORM | wgpu::BufferUsages::COPY_DST,
			mapped_at_creation: false,
		});

		let max_buffer_size = device.limits().max_buffer_size;
		let storage_buffer_alignment = device.limits().min_storage_buffer_offset_alignment as u64;
		let initial_instance_buffer_capacity = 2 * 1024 * 1024;
		let instance_buffer = device.create_buffer(&wgpu::BufferDescriptor {
			label: Some("instance_buffer"),
			size: initial_instance_buffer_capacity,
			usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_DST,
			mapped_at_creation: false,
		});

		let globals_bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
			label: Some("globals_bind_group"),
			layout: &bind_group_layouts.globals,
			entries: &[
				wgpu::BindGroupEntry {
					binding: 0,
					resource: wgpu::BindingResource::Buffer(wgpu::BufferBinding {
						buffer: &globals_buffer,
						offset: 0,
						size: Some(NonZeroU64::new(globals_size).unwrap()),
					}),
				},
				wgpu::BindGroupEntry {
					binding: 1,
					resource: wgpu::BindingResource::Buffer(wgpu::BufferBinding {
						buffer: &globals_buffer,
						offset: gamma_offset,
						size: Some(NonZeroU64::new(gamma_size).unwrap()),
					}),
				},
			],
		});

		let path_globals_bind_group = device.create_bind_group(&wgpu::BindGroupDescriptor {
			label: Some("path_globals_bind_group"),
			layout: &bind_group_layouts.globals,
			entries: &[
				wgpu::BindGroupEntry {
					binding: 0,
					resource: wgpu::BindingResource::Buffer(wgpu::BufferBinding {
						buffer: &globals_buffer,
						offset: path_globals_offset,
						size: Some(NonZeroU64::new(globals_size).unwrap()),
					}),
				},
				wgpu::BindGroupEntry {
					binding: 1,
					resource: wgpu::BindingResource::Buffer(wgpu::BufferBinding {
						buffer: &globals_buffer,
						offset: gamma_offset,
						size: Some(NonZeroU64::new(gamma_size).unwrap()),
					}),
				},
			],
		});

		let adapter_info = context.adapter.get_info();

		let last_error: Arc<Mutex<Option<String>>> = Arc::new(Mutex::new(None));
		let last_error_clone = Arc::clone(&last_error);
		device.on_uncaptured_error(Arc::new(move |error| {
			let mut guard = last_error_clone.lock().unwrap();
			*guard = Some(error.to_string());
		}));

		let resources = WgpuResources {
			device,
			queue,
			surface,
			pipelines,
			bind_group_layouts,
			atlas_sampler,
			surface_sampler,
			surface_uniform_buffer,
			blur_params_buffer,
			globals_buffer,
			globals_bind_group,
			path_globals_bind_group,
			instance_buffer,
			// Defer intermediate texture creation to first draw call via ensure_intermediate_textures().
			// This avoids panics when the device/surface is in an invalid state during initialization.
			path_intermediate_texture: None,
			path_intermediate_view: None,
			path_msaa_texture: None,
			path_msaa_view: None,
			scene_color_texture: None,
			scene_color_view: None,
			blur_ping_texture: None,
			blur_ping_view: None,
			blur_pong_texture: None,
			blur_pong_view: None,
			group_textures: Vec::new(),
			group_views: Vec::new(),
		};

		Ok(Self {
			context: gpu_context,
			compositor_gpu,
			extra_requirements,
			resources: Some(resources),
			surface_config,
			atlas,
			path_globals_offset,
			gamma_offset,
			instance_buffer_capacity: initial_instance_buffer_capacity,
			max_buffer_size,
			storage_buffer_alignment,
			blur_params_stride,
			blur_params_slot: std::cell::Cell::new(0),
			rendering_params,
			is_bgr: false,
			dual_source_blending,
			adapter_info,
			transparent_alpha_mode,
			opaque_alpha_mode,
			max_texture_size,
			last_error,
			failed_frame_count: 0,
			device_lost: context.device_lost_flag(),
			surface_configured: true,
			needs_redraw: false,
		})
	}

	fn create_bind_group_layouts(device: &wgpu::Device) -> WgpuBindGroupLayouts {
		let globals = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
			label: Some("globals_layout"),
			entries: &[
				wgpu::BindGroupLayoutEntry {
					binding: 0,
					visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
					ty: wgpu::BindingType::Buffer {
						ty: wgpu::BufferBindingType::Uniform,
						has_dynamic_offset: false,
						min_binding_size: NonZeroU64::new(
							std::mem::size_of::<GlobalParams>() as u64
						),
					},
					count: None,
				},
				wgpu::BindGroupLayoutEntry {
					binding: 1,
					visibility: wgpu::ShaderStages::FRAGMENT,
					ty: wgpu::BindingType::Buffer {
						ty: wgpu::BufferBindingType::Uniform,
						has_dynamic_offset: false,
						min_binding_size: NonZeroU64::new(std::mem::size_of::<GammaParams>() as u64),
					},
					count: None,
				},
			],
		});

		let storage_buffer_entry = |binding: u32| wgpu::BindGroupLayoutEntry {
			binding,
			visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
			ty: wgpu::BindingType::Buffer {
				ty: wgpu::BufferBindingType::Storage { read_only: true },
				has_dynamic_offset: false,
				min_binding_size: None,
			},
			count: None,
		};

		let instances = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
			label: Some("instances_layout"),
			entries: &[storage_buffer_entry(0)],
		});

		let instances_with_texture =
			device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
				label: Some("instances_with_texture_layout"),
				entries: &[
					storage_buffer_entry(0),
					wgpu::BindGroupLayoutEntry {
						binding: 1,
						visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
						ty: wgpu::BindingType::Texture {
							sample_type: wgpu::TextureSampleType::Float { filterable: true },
							view_dimension: wgpu::TextureViewDimension::D2,
							multisampled: false,
						},
						count: None,
					},
					wgpu::BindGroupLayoutEntry {
						binding: 2,
						visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
						ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::Filtering),
						count: None,
					},
				],
			});

		let surfaces = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
			label: Some("surfaces_layout"),
			entries: &[
				wgpu::BindGroupLayoutEntry {
					binding: 0,
					visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
					ty: wgpu::BindingType::Buffer {
						ty: wgpu::BufferBindingType::Uniform,
						has_dynamic_offset: false,
						min_binding_size: NonZeroU64::new(
							std::mem::size_of::<SurfaceParams>() as u64
						),
					},
					count: None,
				},
				wgpu::BindGroupLayoutEntry {
					binding: 1,
					visibility: wgpu::ShaderStages::FRAGMENT,
					ty: wgpu::BindingType::Texture {
						sample_type: wgpu::TextureSampleType::Float { filterable: true },
						view_dimension: wgpu::TextureViewDimension::D2,
						multisampled: false,
					},
					count: None,
				},
				wgpu::BindGroupLayoutEntry {
					binding: 2,
					visibility: wgpu::ShaderStages::FRAGMENT,
					ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::Filtering),
					count: None,
				},
			],
		});

		let blur = device.create_bind_group_layout(&wgpu::BindGroupLayoutDescriptor {
			label: Some("blur_layout"),
			entries: &[
				wgpu::BindGroupLayoutEntry {
					binding: 0,
					visibility: wgpu::ShaderStages::VERTEX_FRAGMENT,
					ty: wgpu::BindingType::Buffer {
						ty: wgpu::BufferBindingType::Uniform,
						has_dynamic_offset: false,
						min_binding_size: NonZeroU64::new(std::mem::size_of::<BlurParams>() as u64),
					},
					count: None,
				},
				wgpu::BindGroupLayoutEntry {
					binding: 1,
					visibility: wgpu::ShaderStages::FRAGMENT,
					ty: wgpu::BindingType::Texture {
						sample_type: wgpu::TextureSampleType::Float { filterable: true },
						view_dimension: wgpu::TextureViewDimension::D2,
						multisampled: false,
					},
					count: None,
				},
				wgpu::BindGroupLayoutEntry {
					binding: 2,
					visibility: wgpu::ShaderStages::FRAGMENT,
					ty: wgpu::BindingType::Sampler(wgpu::SamplerBindingType::Filtering),
					count: None,
				},
			],
		});

		WgpuBindGroupLayouts {
			globals,
			instances,
			instances_with_texture,
			surfaces,
			blur,
		}
	}

	fn create_pipelines(
		device: &wgpu::Device,
		layouts: &WgpuBindGroupLayouts,
		surface_format: wgpu::TextureFormat,
		alpha_mode: wgpu::CompositeAlphaMode,
		path_sample_count: u32,
		dual_source_blending: bool,
	) -> WgpuPipelines {
		// Diagnostic guard: verify the device actually has
		// DUAL_SOURCE_BLENDING. We have a crash report (ZED-5G1) where a
		// feature mismatch caused a wgpu-hal abort, but we haven't
		// identified the code path that produces the mismatch. This
		// guard prevents the crash and logs more evidence.
		// Remove this check once:
		// a) We find and fix the root cause, or
		// b) There are no reports of this warning appearing for some time.
		let device_has_feature = device
			.features()
			.contains(wgpu::Features::DUAL_SOURCE_BLENDING);
		if dual_source_blending && !device_has_feature {
			log::error!(
				"BUG: dual_source_blending flag is true but device does not \
                 have DUAL_SOURCE_BLENDING enabled (device features: {:?}). \
                 Falling back to mono text rendering. Please report this at \
                 https://github.com/zed-industries/zed/issues",
				device.features(),
			);
		}
		let dual_source_blending = dual_source_blending && device_has_feature;

		let base_shader_source = include_str!("shaders.wgsl");
		let shader_module = device.create_shader_module(wgpu::ShaderModuleDescriptor {
			label: Some("gpui_shaders"),
			source: wgpu::ShaderSource::Wgsl(std::borrow::Cow::Borrowed(base_shader_source)),
		});

		let subpixel_shader_source = include_str!("shaders_subpixel.wgsl");
		let subpixel_shader_module = if dual_source_blending {
			let combined = format!(
				"enable dual_source_blending;\n{base_shader_source}\n{subpixel_shader_source}"
			);
			Some(device.create_shader_module(wgpu::ShaderModuleDescriptor {
				label: Some("gpui_subpixel_shaders"),
				source: wgpu::ShaderSource::Wgsl(std::borrow::Cow::Owned(combined)),
			}))
		} else {
			None
		};

		let blend_mode = match alpha_mode {
			wgpu::CompositeAlphaMode::PreMultiplied => {
				wgpu::BlendState::PREMULTIPLIED_ALPHA_BLENDING
			}
			_ => wgpu::BlendState::ALPHA_BLENDING,
		};

		let color_target = wgpu::ColorTargetState {
			format: surface_format,
			blend: Some(blend_mode),
			write_mask: wgpu::ColorWrites::ALL,
		};

		let create_pipeline = |name: &str,
		                       vs_entry: &str,
		                       fs_entry: &str,
		                       globals_layout: &wgpu::BindGroupLayout,
		                       data_layout: &wgpu::BindGroupLayout,
		                       topology: wgpu::PrimitiveTopology,
		                       color_targets: &[Option<wgpu::ColorTargetState>],
		                       sample_count: u32,
		                       module: &wgpu::ShaderModule| {
			let pipeline_layout = device.create_pipeline_layout(&wgpu::PipelineLayoutDescriptor {
				label: Some(&format!("{name}_layout")),
				bind_group_layouts: &[Some(globals_layout), Some(data_layout)],
				immediate_size: 0,
			});

			device.create_render_pipeline(&wgpu::RenderPipelineDescriptor {
				label: Some(name),
				layout: Some(&pipeline_layout),
				vertex: wgpu::VertexState {
					module,
					entry_point: Some(vs_entry),
					buffers: &[],
					compilation_options: wgpu::PipelineCompilationOptions::default(),
				},
				fragment: Some(wgpu::FragmentState {
					module,
					entry_point: Some(fs_entry),
					targets: color_targets,
					compilation_options: wgpu::PipelineCompilationOptions::default(),
				}),
				primitive: wgpu::PrimitiveState {
					topology,
					strip_index_format: None,
					front_face: wgpu::FrontFace::Ccw,
					cull_mode: None,
					polygon_mode: wgpu::PolygonMode::Fill,
					unclipped_depth: false,
					conservative: false,
				},
				depth_stencil: None,
				multisample: wgpu::MultisampleState {
					count: sample_count,
					mask: !0,
					alpha_to_coverage_enabled: false,
				},
				multiview_mask: None,
				cache: None,
			})
		};

		let quads = create_pipeline(
			"quads",
			"vs_quad",
			"fs_quad",
			&layouts.globals,
			&layouts.instances,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target.clone())],
			1,
			&shader_module,
		);

		let shadows = create_pipeline(
			"shadows",
			"vs_shadow",
			"fs_shadow",
			&layouts.globals,
			&layouts.instances,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target.clone())],
			1,
			&shader_module,
		);

		let path_rasterization = create_pipeline(
			"path_rasterization",
			"vs_path_rasterization",
			"fs_path_rasterization",
			&layouts.globals,
			&layouts.instances,
			wgpu::PrimitiveTopology::TriangleList,
			&[Some(wgpu::ColorTargetState {
				format: surface_format,
				blend: Some(wgpu::BlendState::PREMULTIPLIED_ALPHA_BLENDING),
				write_mask: wgpu::ColorWrites::ALL,
			})],
			path_sample_count,
			&shader_module,
		);

		let paths_blend = wgpu::BlendState {
			color: wgpu::BlendComponent {
				src_factor: wgpu::BlendFactor::One,
				dst_factor: wgpu::BlendFactor::OneMinusSrcAlpha,
				operation: wgpu::BlendOperation::Add,
			},
			alpha: wgpu::BlendComponent {
				src_factor: wgpu::BlendFactor::One,
				dst_factor: wgpu::BlendFactor::One,
				operation: wgpu::BlendOperation::Add,
			},
		};

		let paths = create_pipeline(
			"paths",
			"vs_path",
			"fs_path",
			&layouts.globals,
			&layouts.instances_with_texture,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(wgpu::ColorTargetState {
				format: surface_format,
				blend: Some(paths_blend),
				write_mask: wgpu::ColorWrites::ALL,
			})],
			1,
			&shader_module,
		);

		let underlines = create_pipeline(
			"underlines",
			"vs_underline",
			"fs_underline",
			&layouts.globals,
			&layouts.instances,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target.clone())],
			1,
			&shader_module,
		);

		let mono_sprites = create_pipeline(
			"mono_sprites",
			"vs_mono_sprite",
			"fs_mono_sprite",
			&layouts.globals,
			&layouts.instances_with_texture,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target.clone())],
			1,
			&shader_module,
		);

		let subpixel_sprites = if let Some(subpixel_module) = &subpixel_shader_module {
			let subpixel_blend = wgpu::BlendState {
				color: wgpu::BlendComponent {
					src_factor: wgpu::BlendFactor::Src1,
					dst_factor: wgpu::BlendFactor::OneMinusSrc1,
					operation: wgpu::BlendOperation::Add,
				},
				alpha: wgpu::BlendComponent {
					src_factor: wgpu::BlendFactor::One,
					dst_factor: wgpu::BlendFactor::OneMinusSrcAlpha,
					operation: wgpu::BlendOperation::Add,
				},
			};

			Some(create_pipeline(
				"subpixel_sprites",
				"vs_subpixel_sprite",
				"fs_subpixel_sprite",
				&layouts.globals,
				&layouts.instances_with_texture,
				wgpu::PrimitiveTopology::TriangleStrip,
				&[Some(wgpu::ColorTargetState {
					format: surface_format,
					blend: Some(subpixel_blend),
					write_mask: wgpu::ColorWrites::COLOR,
				})],
				1,
				subpixel_module,
			))
		} else {
			None
		};

		let poly_sprites = create_pipeline(
			"poly_sprites",
			"vs_poly_sprite",
			"fs_poly_sprite",
			&layouts.globals,
			&layouts.instances_with_texture,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target.clone())],
			1,
			&shader_module,
		);

		let surfaces = create_pipeline(
			"surfaces",
			"vs_surface",
			"fs_surface",
			&layouts.globals,
			&layouts.surfaces,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(color_target)],
			1,
			&shader_module,
		);

		// Blur pipelines all sample one texture into another; the downsample and gaussian passes
		// overwrite their (intermediate) target, while the composite blends over the scene.
		let no_blend_target = wgpu::ColorTargetState {
			format: surface_format,
			blend: None,
			write_mask: wgpu::ColorWrites::ALL,
		};

		let blur_downsample = create_pipeline(
			"blur_downsample",
			"vs_blur_fullscreen",
			"fs_blur_downsample",
			&layouts.globals,
			&layouts.blur,
			wgpu::PrimitiveTopology::TriangleList,
			&[Some(no_blend_target.clone())],
			1,
			&shader_module,
		);

		let blur = create_pipeline(
			"blur",
			"vs_blur_fullscreen",
			"fs_blur",
			&layouts.globals,
			&layouts.blur,
			wgpu::PrimitiveTopology::TriangleList,
			&[Some(no_blend_target)],
			1,
			&shader_module,
		);

		// The blurred sample is premultiplied (blurring against the transparent, rgb=0 region
		// around the source scales rgb with the fading alpha), so the composite outputs
		// premultiplied and blends premultiplied — straight alpha blending would multiply rgb by
		// alpha a second time and darken the faded edges. Independent of the window's alpha mode.
		let premultiplied_target = wgpu::ColorTargetState {
			format: surface_format,
			blend: Some(wgpu::BlendState::PREMULTIPLIED_ALPHA_BLENDING),
			write_mask: wgpu::ColorWrites::ALL,
		};
		let blur_composite = create_pipeline(
			"blur_composite",
			"vs_blur_composite",
			"fs_blur_composite",
			&layouts.globals,
			&layouts.blur,
			wgpu::PrimitiveTopology::TriangleStrip,
			&[Some(premultiplied_target)],
			1,
			&shader_module,
		);

		WgpuPipelines {
			quads,
			shadows,
			path_rasterization,
			paths,
			underlines,
			mono_sprites,
			subpixel_sprites,
			poly_sprites,
			surfaces,
			blur_downsample,
			blur,
			blur_composite,
		}
	}

	fn create_path_intermediate(
		device: &wgpu::Device,
		format: wgpu::TextureFormat,
		width: u32,
		height: u32,
	) -> (wgpu::Texture, wgpu::TextureView) {
		let texture = device.create_texture(&wgpu::TextureDescriptor {
			label: Some("path_intermediate"),
			size: wgpu::Extent3d {
				width: width.max(1),
				height: height.max(1),
				depth_or_array_layers: 1,
			},
			mip_level_count: 1,
			sample_count: 1,
			dimension: wgpu::TextureDimension::D2,
			format,
			usage: wgpu::TextureUsages::RENDER_ATTACHMENT | wgpu::TextureUsages::TEXTURE_BINDING,
			view_formats: &[],
		});
		let view = texture.create_view(&wgpu::TextureViewDescriptor::default());
		(texture, view)
	}

	fn create_msaa_if_needed(
		device: &wgpu::Device,
		format: wgpu::TextureFormat,
		width: u32,
		height: u32,
		sample_count: u32,
	) -> Option<(wgpu::Texture, wgpu::TextureView)> {
		if sample_count <= 1 {
			return None;
		}
		let texture = device.create_texture(&wgpu::TextureDescriptor {
			label: Some("path_msaa"),
			size: wgpu::Extent3d {
				width: width.max(1),
				height: height.max(1),
				depth_or_array_layers: 1,
			},
			mip_level_count: 1,
			sample_count,
			dimension: wgpu::TextureDimension::D2,
			format,
			usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
			view_formats: &[],
		});
		let view = texture.create_view(&wgpu::TextureViewDescriptor::default());
		Some((texture, view))
	}

	pub fn update_drawable_size(&mut self, size: Size<DevicePixels>) {
		let width = size.width.0 as u32;
		let height = size.height.0 as u32;

		if width != self.surface_config.width || height != self.surface_config.height {
			let clamped_width = width.min(self.max_texture_size);
			let clamped_height = height.min(self.max_texture_size);

			if clamped_width != width || clamped_height != height {
				warn!(
					"Requested surface size ({}, {}) exceeds maximum texture dimension {}. \
                     Clamping to ({}, {}). Window content may not fill the entire window.",
					width, height, self.max_texture_size, clamped_width, clamped_height
				);
			}

			self.surface_config.width = clamped_width.max(1);
			self.surface_config.height = clamped_height.max(1);
			let surface_config = self.surface_config.clone();

			// GPU resources may not exist yet, skip rather than panicking
			let Some(resources) = self.resources.as_mut() else {
				return;
			};

			// Wait for any in-flight GPU work to complete before destroying textures
			if let Err(e) = resources.device.poll(wgpu::PollType::Wait {
				submission_index: None,
				timeout: None,
			}) {
				warn!("Failed to poll device during resize: {e:?}");
			}

			// Destroy old textures before allocating new ones to avoid GPU memory spikes
			if let Some(ref texture) = resources.path_intermediate_texture {
				texture.destroy();
			}
			if let Some(ref texture) = resources.path_msaa_texture {
				texture.destroy();
			}
			for texture in [
				&resources.scene_color_texture,
				&resources.blur_ping_texture,
				&resources.blur_pong_texture,
			]
			.into_iter()
			.flatten()
			{
				texture.destroy();
			}
			for texture in &resources.group_textures {
				texture.destroy();
			}

			resources
				.surface
				.configure(&resources.device, &surface_config);

			// Invalidate intermediate textures - they will be lazily recreated
			// in draw() after we confirm the surface is healthy. This avoids
			// panics when the device/surface is in an invalid state during resize.
			resources.invalidate_intermediate_textures();
		}
	}

	fn ensure_intermediate_textures(&mut self) {
		if self.resources().path_intermediate_texture.is_some() {
			return;
		}

		let format = self.surface_config.format;
		let width = self.surface_config.width;
		let height = self.surface_config.height;
		let path_sample_count = self.rendering_params.path_sample_count;
		let resources = self.resources_mut();

		let (t, v) = Self::create_path_intermediate(&resources.device, format, width, height);
		resources.path_intermediate_texture = Some(t);
		resources.path_intermediate_view = Some(v);

		let (path_msaa_texture, path_msaa_view) = Self::create_msaa_if_needed(
			&resources.device,
			format,
			width,
			height,
			path_sample_count,
		)
		.map(|(t, v)| (Some(t), Some(v)))
		.unwrap_or((None, None));
		resources.path_msaa_texture = path_msaa_texture;
		resources.path_msaa_view = path_msaa_view;
	}

	/// Lazily allocate the blur offscreen targets — the full-res scene texture, half-res
	/// ping/pong, and one full-res group texture per nesting level. Called only on frames that
	/// actually use a blur filter, so non-blurring apps never pay this VRAM. A no-op once
	/// allocated (invalidated alongside the path intermediates on resize / device loss).
	fn ensure_blur_textures(&mut self) {
		if self.resources().scene_color_texture.is_some() {
			return;
		}
		let format = self.surface_config.format;
		let width = self.surface_config.width;
		let height = self.surface_config.height;
		let blur_width = (width / 2).max(1);
		let blur_height = (height / 2).max(1);
		let resources = self.resources_mut();

		let (t, v) = Self::create_path_intermediate(&resources.device, format, width, height);
		resources.scene_color_texture = Some(t);
		resources.scene_color_view = Some(v);
		let (t, v) =
			Self::create_path_intermediate(&resources.device, format, blur_width, blur_height);
		resources.blur_ping_texture = Some(t);
		resources.blur_ping_view = Some(v);
		let (t, v) =
			Self::create_path_intermediate(&resources.device, format, blur_width, blur_height);
		resources.blur_pong_texture = Some(t);
		resources.blur_pong_view = Some(v);

		for _ in 0..MAX_FILTER_DEPTH {
			let (t, v) = Self::create_path_intermediate(&resources.device, format, width, height);
			resources.group_textures.push(t);
			resources.group_views.push(v);
		}
	}

	pub fn set_subpixel_layout(&mut self, is_bgr: bool) {
		self.is_bgr = is_bgr;
	}

	pub fn update_transparency(&mut self, transparent: bool) {
		let new_alpha_mode = if transparent {
			self.transparent_alpha_mode
		} else {
			self.opaque_alpha_mode
		};

		if new_alpha_mode != self.surface_config.alpha_mode {
			self.surface_config.alpha_mode = new_alpha_mode;
			let surface_config = self.surface_config.clone();
			let path_sample_count = self.rendering_params.path_sample_count;
			let dual_source_blending = self.dual_source_blending;
			let resources = self.resources_mut();
			resources
				.surface
				.configure(&resources.device, &surface_config);
			resources.pipelines = Self::create_pipelines(
				&resources.device,
				&resources.bind_group_layouts,
				surface_config.format,
				surface_config.alpha_mode,
				path_sample_count,
				dual_source_blending,
			);
		}
	}

	#[allow(dead_code)]
	pub fn viewport_size(&self) -> Size<DevicePixels> {
		Size {
			width: DevicePixels(self.surface_config.width as i32),
			height: DevicePixels(self.surface_config.height as i32),
		}
	}

	pub fn sprite_atlas(&self) -> &Arc<WgpuAtlas> {
		&self.atlas
	}

	pub fn supports_dual_source_blending(&self) -> bool {
		self.dual_source_blending
	}

	pub fn gpu_context(&self) -> (Arc<wgpu::Device>, Arc<wgpu::Queue>) {
		let resources = self.resources();
		(resources.device.clone(), resources.queue.clone())
	}

	pub fn gpu_specs(&self) -> GpuSpecs {
		GpuSpecs {
			is_software_emulated: self.adapter_info.device_type == wgpu::DeviceType::Cpu,
			device_name: self.adapter_info.name.clone(),
			driver_name: self.adapter_info.driver.clone(),
			driver_info: self.adapter_info.driver_info.clone(),
		}
	}

	pub fn max_texture_size(&self) -> u32 {
		self.max_texture_size
	}

	pub fn draw(&mut self, scene: &Scene) -> bool {
		// Bail out early if the surface has been unconfigured (e.g. during
		// Android background/rotation transitions).  Attempting to acquire
		// a texture from an unconfigured surface can block indefinitely on
		// some drivers (Adreno).
		if !self.surface_configured {
			return false;
		}

		let last_error = self.last_error.lock().unwrap().take();
		if let Some(error) = last_error {
			self.failed_frame_count += 1;
			log::error!(
				"GPU error during frame (failure {} of 10): {error}",
				self.failed_frame_count
			);

			// TBD. Does retrying more actually help?
			if self.failed_frame_count > 10 {
				panic!("Too many consecutive GPU errors. Last error: {error}");
			} else if self.failed_frame_count > 5 {
				if let Some(res) = self.resources.as_mut() {
					res.invalidate_intermediate_textures();
				}
				self.atlas.clear();
				self.needs_redraw = true;
				self.failed_frame_count = 0;
				return false;
			}
		} else {
			self.failed_frame_count = 0;
		}

		self.atlas.before_frame();

		let frame = match self.resources().surface.get_current_texture() {
			wgpu::CurrentSurfaceTexture::Success(frame) => frame,
			wgpu::CurrentSurfaceTexture::Suboptimal(frame) => {
				// Textures must be destroyed before the surface can be reconfigured.
				drop(frame);
				let surface_config = self.surface_config.clone();
				let resources = self.resources_mut();
				resources
					.surface
					.configure(&resources.device, &surface_config);
				return false;
			}
			wgpu::CurrentSurfaceTexture::Lost | wgpu::CurrentSurfaceTexture::Outdated => {
				let surface_config = self.surface_config.clone();
				let resources = self.resources_mut();
				resources
					.surface
					.configure(&resources.device, &surface_config);
				return false;
			}
			wgpu::CurrentSurfaceTexture::Timeout | wgpu::CurrentSurfaceTexture::Occluded => {
				return false;
			}
			wgpu::CurrentSurfaceTexture::Validation => {
				*self.last_error.lock().unwrap() =
					Some("Surface texture validation error".to_string());
				return false;
			}
		};

		// Now that we know the surface is healthy, ensure intermediate textures exist
		self.ensure_intermediate_textures();

		// Blur is the only thing that needs the offscreen scene texture; allocate it (and the
		// ping/pong/group targets) lazily so non-blurring apps pay no extra VRAM or blit.
		let use_offscreen =
			!scene.backdrop_filters.is_empty() || !scene.filter_boundaries.is_empty();
		if use_offscreen {
			self.ensure_blur_textures();
		}

		let frame_view = frame
			.texture
			.create_view(&wgpu::TextureViewDescriptor::default());

		let gamma_params = GammaParams {
			gamma_ratios: self.rendering_params.gamma_ratios,
			grayscale_enhanced_contrast: self.rendering_params.grayscale_enhanced_contrast,
			subpixel_enhanced_contrast: self.rendering_params.subpixel_enhanced_contrast,
			is_bgr: self.is_bgr as u32,
			_pad: 0,
		};

		let globals = GlobalParams {
			viewport_size: [
				self.surface_config.width as f32,
				self.surface_config.height as f32,
			],
			premultiplied_alpha: if self.surface_config.alpha_mode
				== wgpu::CompositeAlphaMode::PreMultiplied
			{
				1
			} else {
				0
			},
			pad: 0,
		};

		let path_globals = GlobalParams {
			premultiplied_alpha: 0,
			..globals
		};

		{
			let resources = self.resources();
			resources.queue.write_buffer(
				&resources.globals_buffer,
				0,
				bytemuck::bytes_of(&globals),
			);
			resources.queue.write_buffer(
				&resources.globals_buffer,
				self.path_globals_offset,
				bytemuck::bytes_of(&path_globals),
			);
			resources.queue.write_buffer(
				&resources.globals_buffer,
				self.gamma_offset,
				bytemuck::bytes_of(&gamma_params),
			);
		}

		loop {
			let mut instance_offset: u64 = 0;
			// Reset the blur-params bump cursor each (re)render of the scene.
			self.blur_params_slot.set(0);
			let mut overflow = false;

			let mut encoder =
				self.resources()
					.device
					.create_command_encoder(&wgpu::CommandEncoderDescriptor {
						label: Some("main_encoder"),
					});

			// When the scene contains blur filters, render into the offscreen scene texture (so
			// filters can sample already-painted content mid-frame) and blit to the swapchain at
			// the end; otherwise render straight to the swapchain. `use_offscreen` and the blur
			// textures were computed/allocated above.
			let scene_color_view = if use_offscreen {
				Some(
					self.resources()
						.scene_color_view
						.as_ref()
						.expect("scene_color_view allocated by ensure_blur_textures")
						.clone(),
				)
			} else {
				None
			};
			// The active render target. While inside a content-filter (`filter`) group it points
			// at a group texture so the group renders in isolation.
			let mut current_target = match &scene_color_view {
				Some(view) => view.clone(),
				None => frame_view.clone(),
			};
			// One group texture per nesting depth; empty when not blurring.
			let group_views = if use_offscreen {
				self.resources().group_views.clone()
			} else {
				Vec::new()
			};
			// (boundary, parent target to composite back into, whether this level is isolated).
			let mut filter_stack: Vec<(FilterBoundary, wgpu::TextureView, bool)> = Vec::new();

			{
				let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
					label: Some("main_pass"),
					color_attachments: &[Some(wgpu::RenderPassColorAttachment {
						view: &current_target,
						resolve_target: None,
						ops: wgpu::Operations {
							load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
							store: wgpu::StoreOp::Store,
						},
						depth_slice: None,
					})],
					depth_stencil_attachment: None,
					..Default::default()
				});

				for batch in scene.batches() {
					let ok = match batch {
						PrimitiveBatch::Quads(range) => {
							self.draw_quads(&scene.quads[range], &mut instance_offset, &mut pass)
						}
						PrimitiveBatch::Shadows(range) => self.draw_shadows(
							&scene.shadows[range],
							&mut instance_offset,
							&mut pass,
						),
						PrimitiveBatch::Paths(range) => {
							let paths = &scene.paths[range];
							if paths.is_empty() {
								continue;
							}

							drop(pass);

							let did_draw = self.draw_paths_to_intermediate(
								&mut encoder,
								paths,
								&mut instance_offset,
							);

							pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
								label: Some("main_pass_continued"),
								color_attachments: &[Some(wgpu::RenderPassColorAttachment {
									view: &current_target,
									resolve_target: None,
									ops: wgpu::Operations {
										load: wgpu::LoadOp::Load,
										store: wgpu::StoreOp::Store,
									},
									depth_slice: None,
								})],
								depth_stencil_attachment: None,
								..Default::default()
							});

							if did_draw {
								self.draw_paths_from_intermediate(
									paths,
									&mut instance_offset,
									&mut pass,
								)
							} else {
								false
							}
						}
						PrimitiveBatch::Underlines(range) => self.draw_underlines(
							&scene.underlines[range],
							&mut instance_offset,
							&mut pass,
						),
						PrimitiveBatch::MonochromeSprites { texture_id, range } => self
							.draw_monochrome_sprites(
								&scene.monochrome_sprites[range],
								texture_id,
								&mut instance_offset,
								&mut pass,
							),
						PrimitiveBatch::SubpixelSprites { texture_id, range } => self
							.draw_subpixel_sprites(
								&scene.subpixel_sprites[range],
								texture_id,
								&mut instance_offset,
								&mut pass,
							),
						PrimitiveBatch::PolychromeSprites { texture_id, range } => self
							.draw_polychrome_sprites(
								&scene.polychrome_sprites[range],
								texture_id,
								&mut instance_offset,
								&mut pass,
							),
						PrimitiveBatch::Surfaces(range) => {
							self.draw_surfaces(&scene.surfaces[range], &mut pass)
						}
						PrimitiveBatch::BackdropFilters(range) => {
							// Interrupt the current pass, blur the content painted so far behind
							// each backdrop's rounded rect, then resume drawing on top.
							drop(pass);
							for filter in &scene.backdrop_filters[range] {
								self.draw_backdrop_filter(&mut encoder, filter, &current_target);
							}
							pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
								label: Some("main_pass_continued"),
								color_attachments: &[Some(wgpu::RenderPassColorAttachment {
									view: &current_target,
									resolve_target: None,
									ops: wgpu::Operations {
										load: wgpu::LoadOp::Load,
										store: wgpu::StoreOp::Store,
									},
									depth_slice: None,
								})],
								depth_stencil_attachment: None,
								..Default::default()
							});
							true
						}
						PrimitiveBatch::FilterBoundary(ix) => {
							let boundary = scene.filter_boundaries[ix].clone();
							if boundary.is_start {
								// Each isolated nesting level uses its own group texture from the
								// pool (indexed by current isolation depth). Beyond the pool size
								// (MAX_FILTER_DEPTH) deeper filters render inline without isolation
								// rather than corrupting an outer group.
								let depth = filter_stack.iter().filter(|entry| entry.2).count();
								if depth < group_views.len() {
									drop(pass);
									let parent = current_target.clone();
									current_target = group_views[depth].clone();
									filter_stack.push((boundary, parent, true));
									pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
										label: Some("filter_group"),
										color_attachments: &[Some(
											wgpu::RenderPassColorAttachment {
												view: &current_target,
												resolve_target: None,
												ops: wgpu::Operations {
													load: wgpu::LoadOp::Clear(
														wgpu::Color::TRANSPARENT,
													),
													store: wgpu::StoreOp::Store,
												},
												depth_slice: None,
											},
										)],
										depth_stencil_attachment: None,
										..Default::default()
									});
								} else {
									filter_stack.push((boundary, current_target.clone(), false));
								}
							} else if let Some((boundary, parent, isolated)) = filter_stack.pop() {
								if isolated {
									drop(pass);
									self.blur_and_composite(
										&mut encoder,
										&current_target,
										&parent,
										boundary.bounds,
										boundary.content_mask.bounds,
										[
											boundary.corner_radii.top_left.0,
											boundary.corner_radii.top_right.0,
											boundary.corner_radii.bottom_right.0,
											boundary.corner_radii.bottom_left.0,
										],
										max_blur_radius(&boundary.filters),
										boundary.opacity,
										false,
									);
									current_target = parent;
									pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
										label: Some("main_pass_continued"),
										color_attachments: &[Some(
											wgpu::RenderPassColorAttachment {
												view: &current_target,
												resolve_target: None,
												ops: wgpu::Operations {
													load: wgpu::LoadOp::Load,
													store: wgpu::StoreOp::Store,
												},
												depth_slice: None,
											},
										)],
										depth_stencil_attachment: None,
										..Default::default()
									});
								}
							}
							true
						}
					};
					if !ok {
						overflow = true;
						break;
					}
				}
			}

			if overflow {
				drop(encoder);
				if self.instance_buffer_capacity >= self.max_buffer_size {
					log::error!(
						"instance buffer size grew too large: {}",
						self.instance_buffer_capacity
					);
					frame.present();
					return true;
				}
				self.grow_instance_buffer();
				continue;
			}

			// Present the offscreen scene by copying it into the swapchain texture. Skipped when
			// rendering went straight to the swapchain (no filters this frame).
			if let Some(scene_color_view) = &scene_color_view {
				self.blit_to_frame(&mut encoder, scene_color_view, &frame_view);
			}

			self.resources()
				.queue
				.submit(std::iter::once(encoder.finish()));
			frame.present();
			return true;
		}
	}

	fn draw_quads(
		&self,
		quads: &[Quad],
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let data = unsafe { Self::instance_bytes(quads) };
		self.draw_instances(
			data,
			quads.len() as u32,
			&self.resources().pipelines.quads,
			instance_offset,
			pass,
		)
	}

	fn draw_shadows(
		&self,
		shadows: &[Shadow],
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let data = unsafe { Self::instance_bytes(shadows) };
		self.draw_instances(
			data,
			shadows.len() as u32,
			&self.resources().pipelines.shadows,
			instance_offset,
			pass,
		)
	}

	fn draw_underlines(
		&self,
		underlines: &[Underline],
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let data = unsafe { Self::instance_bytes(underlines) };
		self.draw_instances(
			data,
			underlines.len() as u32,
			&self.resources().pipelines.underlines,
			instance_offset,
			pass,
		)
	}

	fn draw_monochrome_sprites(
		&self,
		sprites: &[MonochromeSprite],
		texture_id: AtlasTextureId,
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let tex_info = self.atlas.get_texture_info(texture_id);
		let data = unsafe { Self::instance_bytes(sprites) };
		self.draw_instances_with_texture(
			data,
			sprites.len() as u32,
			&tex_info.view,
			&self.resources().pipelines.mono_sprites,
			instance_offset,
			pass,
		)
	}

	fn draw_subpixel_sprites(
		&self,
		sprites: &[SubpixelSprite],
		texture_id: AtlasTextureId,
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let tex_info = self.atlas.get_texture_info(texture_id);
		let data = unsafe { Self::instance_bytes(sprites) };
		let resources = self.resources();
		let pipeline = resources
			.pipelines
			.subpixel_sprites
			.as_ref()
			.unwrap_or(&resources.pipelines.mono_sprites);
		self.draw_instances_with_texture(
			data,
			sprites.len() as u32,
			&tex_info.view,
			pipeline,
			instance_offset,
			pass,
		)
	}

	#[cfg(any(target_os = "linux", target_os = "freebsd"))]
	fn draw_surfaces(&self, surfaces: &[PaintSurface], pass: &mut wgpu::RenderPass<'_>) -> bool {
		let resources = self.resources();
		for surface in surfaces {
			let Some(wgpu_texture) = surface.texture.downcast_ref::<wgpu::Texture>() else {
				continue;
			};

			let texture_view = wgpu_texture.create_view(&wgpu::TextureViewDescriptor::default());

			let params = SurfaceParams {
				bounds: surface.bounds.into(),
				content_mask: surface.content_mask.bounds.into(),
			};

			resources.queue.write_buffer(
				&resources.surface_uniform_buffer,
				0,
				bytemuck::bytes_of(&params),
			);

			let bind_group = resources
				.device
				.create_bind_group(&wgpu::BindGroupDescriptor {
					label: Some("surface_bind_group"),
					layout: &resources.bind_group_layouts.surfaces,
					entries: &[
						wgpu::BindGroupEntry {
							binding: 0,
							resource: resources.surface_uniform_buffer.as_entire_binding(),
						},
						wgpu::BindGroupEntry {
							binding: 1,
							resource: wgpu::BindingResource::TextureView(&texture_view),
						},
						wgpu::BindGroupEntry {
							binding: 2,
							resource: wgpu::BindingResource::Sampler(&resources.surface_sampler),
						},
					],
				});

			pass.set_pipeline(&resources.pipelines.surfaces);
			pass.set_bind_group(0, &resources.globals_bind_group, &[]);
			pass.set_bind_group(1, &bind_group, &[]);
			pass.draw(0..4, 0..1);
		}
		true
	}

	#[cfg(not(any(target_os = "linux", target_os = "freebsd")))]
	fn draw_surfaces(&self, _surfaces: &[PaintSurface], _pass: &mut wgpu::RenderPass<'_>) -> bool {
		true
	}

	/// Build a bind group for a blur pass. Writes `params` into the next slot of the shared
	/// `blur_params_buffer` (no per-pass allocation) and references that slot, the source texture,
	/// and the filtering sampler. Distinct per-pass offsets keep `write_buffer`'s
	/// last-write-at-submit semantics from clobbering earlier passes within a frame.
	fn make_blur_bind_group(
		&self,
		params: BlurParams,
		source: &wgpu::TextureView,
	) -> wgpu::BindGroup {
		let resources = self.resources();
		let slot = self.blur_params_slot.get() % BLUR_PARAMS_SLOTS;
		self.blur_params_slot.set(slot + 1);
		let offset = slot * self.blur_params_stride;
		resources.queue.write_buffer(
			&resources.blur_params_buffer,
			offset,
			bytemuck::bytes_of(&params),
		);
		resources
			.device
			.create_bind_group(&wgpu::BindGroupDescriptor {
				label: Some("blur_bind_group"),
				layout: &resources.bind_group_layouts.blur,
				entries: &[
					wgpu::BindGroupEntry {
						binding: 0,
						resource: wgpu::BindingResource::Buffer(wgpu::BufferBinding {
							buffer: &resources.blur_params_buffer,
							offset,
							size: NonZeroU64::new(std::mem::size_of::<BlurParams>() as u64),
						}),
					},
					wgpu::BindGroupEntry {
						binding: 1,
						resource: wgpu::BindingResource::TextureView(source),
					},
					wgpu::BindGroupEntry {
						binding: 2,
						resource: wgpu::BindingResource::Sampler(&resources.surface_sampler),
					},
				],
			})
	}

	/// Run a full-screen (3-vertex) blur pass that overwrites `target` by sampling `source`.
	/// `scissor` (x, y, w, h, in `target` pixels) limits fragment work to the region that
	/// actually feeds the composite — the element bounds dilated by the kernel radius.
	fn run_blur_pass(
		&self,
		encoder: &mut wgpu::CommandEncoder,
		label: &str,
		pipeline: &wgpu::RenderPipeline,
		target: &wgpu::TextureView,
		source: &wgpu::TextureView,
		params: BlurParams,
		scissor: [u32; 4],
	) {
		let bind_group = self.make_blur_bind_group(params, source);
		let resources = self.resources();
		let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
			label: Some(label),
			color_attachments: &[Some(wgpu::RenderPassColorAttachment {
				view: target,
				resolve_target: None,
				ops: wgpu::Operations {
					load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
					store: wgpu::StoreOp::Store,
				},
				depth_slice: None,
			})],
			depth_stencil_attachment: None,
			..Default::default()
		});
		pass.set_pipeline(pipeline);
		pass.set_bind_group(0, &resources.globals_bind_group, &[]);
		pass.set_bind_group(1, &bind_group, &[]);
		pass.set_scissor_rect(scissor[0], scissor[1], scissor[2], scissor[3]);
		pass.draw(0..3, 0..1);
	}

	/// Blur `source` (full-resolution) and composite the result into `target`, clipped to
	/// `bounds`/`corner_radii`/`content_mask` and modulated by `opacity`. Shared by the backdrop
	/// and content-filter paths. Uses the half-resolution ping/pong textures as scratch.
	#[allow(clippy::too_many_arguments)]
	fn blur_and_composite(
		&self,
		encoder: &mut wgpu::CommandEncoder,
		source: &wgpu::TextureView,
		target: &wgpu::TextureView,
		bounds: Bounds<ScaledPixels>,
		content_mask: Bounds<ScaledPixels>,
		corner_radii: [f32; 4],
		blur_radius: f32,
		opacity: f32,
		// Backdrop clips to the rounded rect; content (`filter`) bleeds past its bounds.
		clip_rounded: bool,
	) {
		// Sigma is halved because the blur runs at half resolution.
		let sigma = (blur_radius * 0.5).max(0.0);
		if sigma <= 0.0 {
			return;
		}
		// Span ±3σ. If that needs more than 32 taps, spread the taps apart (tap_step > 1) rather
		// than truncating the kernel — keeps very large radii from clipping (review #6).
		let ideal_taps = (3.0 * sigma).ceil();
		let tap_count = ideal_taps.clamp(1.0, 32.0);
		let tap_step = (ideal_taps / tap_count).max(1.0);
		let full_w = self.surface_config.width;
		let full_h = self.surface_config.height;
		let blur_width = (full_w / 2).max(1) as f32;
		let blur_height = (full_h / 2).max(1) as f32;

		// Limit the half-res passes to the element bounds dilated by the kernel radius (3·sigma,
		// full-res) — outside that the composite never samples, so there's no reason to blur it.
		let dilation = 3.0 * blur_radius;
		let hw = (full_w / 2).max(1);
		let hh = (full_h / 2).max(1);
		let x0 = (((bounds.origin.x.0 - dilation) * 0.5).floor().max(0.0) as u32).min(hw);
		let y0 = (((bounds.origin.y.0 - dilation) * 0.5).floor().max(0.0) as u32).min(hh);
		let x1 = ((((bounds.origin.x.0 + bounds.size.width.0 + dilation) * 0.5)
			.ceil()
			.max(0.0) as u32)
			.min(hw))
		.max(x0);
		let y1 = ((((bounds.origin.y.0 + bounds.size.height.0 + dilation) * 0.5)
			.ceil()
			.max(0.0) as u32)
			.min(hh))
		.max(y0);
		let scissor = [x0, y0, x1 - x0, y1 - y0];
		if scissor[2] == 0 || scissor[3] == 0 {
			return;
		}

		// Owned handles so the passes below don't borrow `self`.
		let (ping, pong) = {
			let resources = self.resources();
			match (
				resources.blur_ping_view.as_ref(),
				resources.blur_pong_view.as_ref(),
			) {
				(Some(ping), Some(pong)) => (ping.clone(), pong.clone()),
				_ => return,
			}
		};

		// Downsample source -> ping, then separable gaussian ping -> pong -> ping.
		self.run_blur_pass(
			encoder,
			"blur_downsample",
			&self.resources().pipelines.blur_downsample,
			&ping,
			source,
			BlurParams {
				downsample: 1.0,
				..Default::default()
			},
			scissor,
		);
		self.run_blur_pass(
			encoder,
			"blur_horizontal",
			&self.resources().pipelines.blur,
			&pong,
			&ping,
			BlurParams {
				direction: [1.0 / blur_width, 0.0],
				sigma,
				tap_count,
				tap_step,
				..Default::default()
			},
			scissor,
		);
		self.run_blur_pass(
			encoder,
			"blur_vertical",
			&self.resources().pipelines.blur,
			&ping,
			&pong,
			BlurParams {
				direction: [0.0, 1.0 / blur_height],
				sigma,
				tap_count,
				tap_step,
				..Default::default()
			},
			scissor,
		);

		// Composite the blurred result into the target (loads existing content). For content blur
		// the quad covers the dilated region so the blur can fade out past the element box (no
		// sharp clip); for backdrop the quad is the element bounds and the shader clips to the
		// rounded rect.
		let composite_bounds = if clip_rounded {
			bounds
		} else {
			bounds.dilate(ScaledPixels(dilation))
		};
		let params = BlurParams {
			bounds: composite_bounds.into(),
			content_mask: content_mask.into(),
			corner_radii,
			opacity,
			clip_rounded: if clip_rounded { 1.0 } else { 0.0 },
			..Default::default()
		};
		let bind_group = self.make_blur_bind_group(params, &ping);
		let resources = self.resources();
		let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
			label: Some("blur_composite"),
			color_attachments: &[Some(wgpu::RenderPassColorAttachment {
				view: target,
				resolve_target: None,
				ops: wgpu::Operations {
					load: wgpu::LoadOp::Load,
					store: wgpu::StoreOp::Store,
				},
				depth_slice: None,
			})],
			depth_stencil_attachment: None,
			..Default::default()
		});
		pass.set_pipeline(&resources.pipelines.blur_composite);
		pass.set_bind_group(0, &resources.globals_bind_group, &[]);
		pass.set_bind_group(1, &bind_group, &[]);
		pass.draw(0..4, 0..1);
	}

	/// Blur the scene painted so far behind `filter.bounds` and composite it back as frosted glass.
	fn draw_backdrop_filter(
		&self,
		encoder: &mut wgpu::CommandEncoder,
		filter: &BackdropFilter,
		scene_color_view: &wgpu::TextureView,
	) {
		self.blur_and_composite(
			encoder,
			scene_color_view,
			scene_color_view,
			filter.bounds,
			filter.content_mask.bounds,
			[
				filter.corner_radii.top_left.0,
				filter.corner_radii.top_right.0,
				filter.corner_radii.bottom_right.0,
				filter.corner_radii.bottom_left.0,
			],
			max_blur_radius(&filter.filters),
			filter.opacity,
			true,
		);
	}

	/// Copy the offscreen scene texture into the swapchain texture.
	fn blit_to_frame(
		&self,
		encoder: &mut wgpu::CommandEncoder,
		source: &wgpu::TextureView,
		frame_view: &wgpu::TextureView,
	) {
		let bind_group = self.make_blur_bind_group(BlurParams::default(), source);
		let resources = self.resources();
		let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
			label: Some("scene_blit"),
			color_attachments: &[Some(wgpu::RenderPassColorAttachment {
				view: frame_view,
				resolve_target: None,
				ops: wgpu::Operations {
					load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
					store: wgpu::StoreOp::Store,
				},
				depth_slice: None,
			})],
			depth_stencil_attachment: None,
			..Default::default()
		});
		pass.set_pipeline(&resources.pipelines.blur_downsample);
		pass.set_bind_group(0, &resources.globals_bind_group, &[]);
		pass.set_bind_group(1, &bind_group, &[]);
		pass.draw(0..3, 0..1);
	}

	fn draw_polychrome_sprites(
		&self,
		sprites: &[PolychromeSprite],
		texture_id: AtlasTextureId,
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let tex_info = self.atlas.get_texture_info(texture_id);
		let data = unsafe { Self::instance_bytes(sprites) };
		self.draw_instances_with_texture(
			data,
			sprites.len() as u32,
			&tex_info.view,
			&self.resources().pipelines.poly_sprites,
			instance_offset,
			pass,
		)
	}

	fn draw_instances(
		&self,
		data: &[u8],
		instance_count: u32,
		pipeline: &wgpu::RenderPipeline,
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		if instance_count == 0 {
			return true;
		}
		let Some((offset, size)) = self.write_to_instance_buffer(instance_offset, data) else {
			return false;
		};
		let resources = self.resources();
		let bind_group = resources
			.device
			.create_bind_group(&wgpu::BindGroupDescriptor {
				label: None,
				layout: &resources.bind_group_layouts.instances,
				entries: &[wgpu::BindGroupEntry {
					binding: 0,
					resource: self.instance_binding(offset, size),
				}],
			});
		pass.set_pipeline(pipeline);
		pass.set_bind_group(0, &resources.globals_bind_group, &[]);
		pass.set_bind_group(1, &bind_group, &[]);
		pass.draw(0..4, 0..instance_count);
		true
	}

	fn draw_instances_with_texture(
		&self,
		data: &[u8],
		instance_count: u32,
		texture_view: &wgpu::TextureView,
		pipeline: &wgpu::RenderPipeline,
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		if instance_count == 0 {
			return true;
		}
		let Some((offset, size)) = self.write_to_instance_buffer(instance_offset, data) else {
			return false;
		};
		let resources = self.resources();
		let bind_group = resources
			.device
			.create_bind_group(&wgpu::BindGroupDescriptor {
				label: None,
				layout: &resources.bind_group_layouts.instances_with_texture,
				entries: &[
					wgpu::BindGroupEntry {
						binding: 0,
						resource: self.instance_binding(offset, size),
					},
					wgpu::BindGroupEntry {
						binding: 1,
						resource: wgpu::BindingResource::TextureView(texture_view),
					},
					wgpu::BindGroupEntry {
						binding: 2,
						resource: wgpu::BindingResource::Sampler(&resources.atlas_sampler),
					},
				],
			});
		pass.set_pipeline(pipeline);
		pass.set_bind_group(0, &resources.globals_bind_group, &[]);
		pass.set_bind_group(1, &bind_group, &[]);
		pass.draw(0..4, 0..instance_count);
		true
	}

	unsafe fn instance_bytes<T>(instances: &[T]) -> &[u8] {
		unsafe {
			std::slice::from_raw_parts(
				instances.as_ptr() as *const u8,
				std::mem::size_of_val(instances),
			)
		}
	}

	fn draw_paths_from_intermediate(
		&self,
		paths: &[Path<ScaledPixels>],
		instance_offset: &mut u64,
		pass: &mut wgpu::RenderPass<'_>,
	) -> bool {
		let first_path = &paths[0];
		let sprites: Vec<PathSprite> = if paths.last().map(|p| &p.order) == Some(&first_path.order)
		{
			paths
				.iter()
				.map(|p| PathSprite {
					bounds: p.clipped_bounds(),
				})
				.collect()
		} else {
			let mut bounds = first_path.clipped_bounds();
			for path in paths.iter().skip(1) {
				bounds = bounds.union(&path.clipped_bounds());
			}
			vec![PathSprite { bounds }]
		};

		let resources = self.resources();
		let Some(path_intermediate_view) = resources.path_intermediate_view.as_ref() else {
			return true;
		};

		let sprite_data = unsafe { Self::instance_bytes(&sprites) };
		self.draw_instances_with_texture(
			sprite_data,
			sprites.len() as u32,
			path_intermediate_view,
			&resources.pipelines.paths,
			instance_offset,
			pass,
		)
	}

	fn draw_paths_to_intermediate(
		&self,
		encoder: &mut wgpu::CommandEncoder,
		paths: &[Path<ScaledPixels>],
		instance_offset: &mut u64,
	) -> bool {
		let mut vertices = Vec::new();
		for path in paths {
			let bounds = path.clipped_bounds();
			vertices.extend(path.vertices.iter().map(|v| PathRasterizationVertex {
				xy_position: v.xy_position,
				st_position: v.st_position,
				color: path.color,
				bounds,
			}));
		}

		if vertices.is_empty() {
			return true;
		}

		let vertex_data = unsafe { Self::instance_bytes(&vertices) };
		let Some((vertex_offset, vertex_size)) =
			self.write_to_instance_buffer(instance_offset, vertex_data)
		else {
			return false;
		};

		let resources = self.resources();
		let data_bind_group = resources
			.device
			.create_bind_group(&wgpu::BindGroupDescriptor {
				label: Some("path_rasterization_bind_group"),
				layout: &resources.bind_group_layouts.instances,
				entries: &[wgpu::BindGroupEntry {
					binding: 0,
					resource: self.instance_binding(vertex_offset, vertex_size),
				}],
			});

		let Some(path_intermediate_view) = resources.path_intermediate_view.as_ref() else {
			return true;
		};

		let (target_view, resolve_target) = if let Some(ref msaa_view) = resources.path_msaa_view {
			(msaa_view, Some(path_intermediate_view))
		} else {
			(path_intermediate_view, None)
		};

		{
			let mut pass = encoder.begin_render_pass(&wgpu::RenderPassDescriptor {
				label: Some("path_rasterization_pass"),
				color_attachments: &[Some(wgpu::RenderPassColorAttachment {
					view: target_view,
					resolve_target,
					ops: wgpu::Operations {
						load: wgpu::LoadOp::Clear(wgpu::Color::TRANSPARENT),
						store: wgpu::StoreOp::Store,
					},
					depth_slice: None,
				})],
				depth_stencil_attachment: None,
				..Default::default()
			});

			pass.set_pipeline(&resources.pipelines.path_rasterization);
			pass.set_bind_group(0, &resources.path_globals_bind_group, &[]);
			pass.set_bind_group(1, &data_bind_group, &[]);
			pass.draw(0..vertices.len() as u32, 0..1);
		}

		true
	}

	fn grow_instance_buffer(&mut self) {
		let new_capacity = (self.instance_buffer_capacity * 2).min(self.max_buffer_size);
		log::info!("increased instance buffer size to {}", new_capacity);
		let resources = self.resources_mut();
		resources.instance_buffer = resources.device.create_buffer(&wgpu::BufferDescriptor {
			label: Some("instance_buffer"),
			size: new_capacity,
			usage: wgpu::BufferUsages::STORAGE | wgpu::BufferUsages::COPY_DST,
			mapped_at_creation: false,
		});
		self.instance_buffer_capacity = new_capacity;
	}

	fn write_to_instance_buffer(
		&self,
		instance_offset: &mut u64,
		data: &[u8],
	) -> Option<(u64, NonZeroU64)> {
		let offset = (*instance_offset).next_multiple_of(self.storage_buffer_alignment);
		let size = (data.len() as u64).max(16);
		if offset + size > self.instance_buffer_capacity {
			return None;
		}
		let resources = self.resources();
		resources
			.queue
			.write_buffer(&resources.instance_buffer, offset, data);
		*instance_offset = offset + size;
		Some((offset, NonZeroU64::new(size).expect("size is at least 16")))
	}

	fn instance_binding(&self, offset: u64, size: NonZeroU64) -> wgpu::BindingResource<'_> {
		wgpu::BindingResource::Buffer(wgpu::BufferBinding {
			buffer: &self.resources().instance_buffer,
			offset,
			size: Some(size),
		})
	}

	/// Mark the surface as unconfigured so rendering is skipped until a new
	/// surface is provided via [`replace_surface`](Self::replace_surface).
	///
	/// This does **not** drop the renderer — the device, queue, atlas, and
	/// pipelines stay alive.  Use this when the native window is destroyed
	/// (e.g. Android `TerminateWindow`) but you intend to re-create the
	/// surface later without losing cached atlas textures.
	pub fn unconfigure_surface(&mut self) {
		self.surface_configured = false;
		// Drop intermediate textures since they reference the old surface size.
		if let Some(res) = self.resources.as_mut() {
			res.invalidate_intermediate_textures();
		}
	}

	/// Replace the wgpu surface with a new one (e.g. after Android destroys
	/// and recreates the native window).  Keeps the device, queue, atlas, and
	/// all pipelines intact so cached `AtlasTextureId`s remain valid.
	///
	/// The `instance` **must** be the same [`wgpu::Instance`] that was used to
	/// create the adapter and device (i.e. from the [`WgpuContext`]).  Using a
	/// different instance will cause a "Device does not exist" panic because
	/// the wgpu device is bound to its originating instance.
	#[cfg(not(target_family = "wasm"))]
	pub fn replace_surface<W: HasWindowHandle>(
		&mut self,
		window: &W,
		config: WgpuSurfaceConfig,
		instance: &wgpu::Instance,
	) -> anyhow::Result<()> {
		let window_handle = window
			.window_handle()
			.map_err(|e| anyhow::anyhow!("Failed to get window handle: {e}"))?;

		let surface = create_surface(instance, window_handle.as_raw())?;

		let width = (config.size.width.0 as u32).max(1);
		let height = (config.size.height.0 as u32).max(1);

		let alpha_mode = if config.transparent {
			self.transparent_alpha_mode
		} else {
			self.opaque_alpha_mode
		};

		self.surface_config.width = width;
		self.surface_config.height = height;
		self.surface_config.alpha_mode = alpha_mode;
		if let Some(mode) = config.preferred_present_mode {
			self.surface_config.present_mode = mode;
		}

		{
			let res = self
				.resources
				.as_mut()
				.expect("GPU resources not available");
			surface.configure(&res.device, &self.surface_config);
			res.surface = surface;

			// Invalidate intermediate textures — they'll be recreated lazily.
			res.invalidate_intermediate_textures();
		}

		self.surface_configured = true;

		Ok(())
	}

	pub fn destroy(&mut self) {
		// Release surface-bound GPU resources eagerly so the underlying native
		// window can be destroyed before the renderer itself is dropped.
		self.resources.take();
	}

	/// Returns true if the GPU device was lost and recovery is needed.
	pub fn device_lost(&self) -> bool {
		self.device_lost.load(std::sync::atomic::Ordering::SeqCst)
	}

	/// Returns true if a redraw is needed because GPU state was cleared.
	/// Calling this method clears the flag.
	pub fn needs_redraw(&mut self) -> bool {
		std::mem::take(&mut self.needs_redraw)
	}

	/// Recovers from a lost GPU device by recreating the renderer with a new context.
	///
	/// Call this after detecting `device_lost()` returns true.
	///
	/// This method coordinates recovery across multiple windows:
	/// - The first window to call this will recreate the shared context
	/// - Subsequent windows will adopt the already-recovered context
	#[cfg(not(target_family = "wasm"))]
	pub fn recover<W>(&mut self, window: &W) -> anyhow::Result<()>
	where
		W: HasWindowHandle + HasDisplayHandle + std::fmt::Debug + Send + Sync + Clone + 'static,
	{
		let gpu_context = self.context.as_ref().expect("recover requires gpu_context");

		// Check if another window already recovered the context
		let needs_new_context = gpu_context
			.borrow()
			.as_ref()
			.is_none_or(|ctx| ctx.device_lost());

		let window_handle = window
			.window_handle()
			.map_err(|e| anyhow::anyhow!("Failed to get window handle: {e}"))?;

		let surface = if needs_new_context {
			log::warn!("GPU device lost, recreating context...");

			// Drop old resources to release Arc<Device>/Arc<Queue> and GPU resources
			self.resources = None;
			*gpu_context.borrow_mut() = None;

			// Wait briefly for the GPU driver to stabilize, then try to
			// recreate the context without software renderers. If this fails
			// the caller should request another frame and retry — the real GPU
			// may need more time to come back (e.g. after suspend/resume).
			std::thread::sleep(std::time::Duration::from_millis(350));

			let instance = WgpuContext::instance(Box::new(window.clone()));
			let surface = create_surface(&instance, window_handle.as_raw())?;
			let new_context = WgpuContext::new_rejecting_software(
				instance,
				&surface,
				self.compositor_gpu,
				self.extra_requirements.as_ref(),
			)?;
			*gpu_context.borrow_mut() = Some(new_context);
			surface
		} else {
			let ctx_ref = gpu_context.borrow();
			let instance = &ctx_ref.as_ref().unwrap().instance;
			create_surface(instance, window_handle.as_raw())?
		};

		let config = WgpuSurfaceConfig {
			size: gpui::Size {
				width: gpui::DevicePixels(self.surface_config.width as i32),
				height: gpui::DevicePixels(self.surface_config.height as i32),
			},
			transparent: self.surface_config.alpha_mode != wgpu::CompositeAlphaMode::Opaque,
			preferred_present_mode: Some(self.surface_config.present_mode),
		};
		let gpu_context = Rc::clone(gpu_context);
		let ctx_ref = gpu_context.borrow();
		let context = ctx_ref.as_ref().expect("context should exist");

		self.resources = None;
		self.atlas.handle_device_lost(context);

		let extra_reqs = self.extra_requirements.clone();
		*self = Self::new_internal(
			Some(gpu_context.clone()),
			context,
			surface,
			config,
			self.compositor_gpu,
			extra_reqs,
			self.atlas.clone(),
		)?;

		log::info!("GPU recovery complete");
		Ok(())
	}
}

#[cfg(not(target_family = "wasm"))]
fn create_surface(
	instance: &wgpu::Instance,
	raw_window_handle: raw_window_handle::RawWindowHandle,
) -> anyhow::Result<wgpu::Surface<'static>> {
	unsafe {
		instance
			.create_surface_unsafe(wgpu::SurfaceTargetUnsafe::RawHandle {
				// Fall back to the display handle already provided via InstanceDescriptor::display.
				raw_display_handle: None,
				raw_window_handle,
			})
			.map_err(|e| anyhow::anyhow!("{e}"))
	}
}

struct RenderingParameters {
	path_sample_count: u32,
	gamma_ratios: [f32; 4],
	grayscale_enhanced_contrast: f32,
	subpixel_enhanced_contrast: f32,
}

impl RenderingParameters {
	fn new(adapter: &wgpu::Adapter, surface_format: wgpu::TextureFormat) -> Self {
		use std::env;

		let format_features = adapter.get_texture_format_features(surface_format);
		let path_sample_count = [4, 2, 1]
			.into_iter()
			.find(|&n| format_features.flags.sample_count_supported(n))
			.unwrap_or(1);

		let gamma = env::var("ZED_FONTS_GAMMA")
			.ok()
			.and_then(|v| v.parse().ok())
			.unwrap_or(1.8_f32)
			.clamp(1.0, 2.2);
		let gamma_ratios = get_gamma_correction_ratios(gamma);

		let grayscale_enhanced_contrast = env::var("ZED_FONTS_GRAYSCALE_ENHANCED_CONTRAST")
			.ok()
			.and_then(|v| v.parse().ok())
			.unwrap_or(1.0_f32)
			.max(0.0);

		let subpixel_enhanced_contrast = env::var("ZED_FONTS_SUBPIXEL_ENHANCED_CONTRAST")
			.ok()
			.and_then(|v| v.parse().ok())
			.unwrap_or(0.5_f32)
			.max(0.0);

		Self {
			path_sample_count,
			gamma_ratios,
			grayscale_enhanced_contrast,
			subpixel_enhanced_contrast,
		}
	}
}