roxlap-render 0.11.0

Unified CPU/GPU renderer facade for the roxlap scene-graph engine — one SceneRenderer over roxlap-core opticast (softbuffer) and roxlap-gpu (wgpu), with automatic CPU fallback.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
//! CPU backend — `roxlap-core` opticast presented via `softbuffer`.
//!
//! RF.1: owns the software surface + the per-frame [`ScratchPool`] and
//! z-buffer, and runs the multi-grid opticast compositor
//! ([`render_scene_composed`]). Mirrors the scene-demo's old `redraw`
//! world pass. Sprites land in RF.3.

#[cfg(not(target_arch = "wasm32"))]
use std::num::NonZeroU32;
#[cfg(not(target_arch = "wasm32"))]
use std::sync::Arc;

use roxlap_core::camera_math;
use roxlap_core::kfa_draw::solve_kfa_limbs;
use roxlap_core::rasterizer::ScratchPool;
use roxlap_core::sprite::{draw_sprite, DrawTarget};
use roxlap_core::Camera;
use roxlap_formats::kv6::Kv6;
use roxlap_formats::sprite::Sprite;
use roxlap_scene::render::render_scene_composed;
use roxlap_scene::Scene;

#[cfg(not(target_arch = "wasm32"))]
use crate::{DynDisplay, DynWindow, HasDisplayHandle, HasWindowHandle};
use crate::{FrameParams, KfaSprite, Line3, RenderOptions, SpriteSet};

/// Near plane (camera-forward distance, voxel units) below which a
/// [`Line3`] endpoint is clipped — keeps the pinhole divide finite and
/// stops points behind the camera from wrapping onto the screen.
const NEAR_Z: f32 = 0.0625;

/// Depth-test slack (perpendicular distance) so a line resting on the
/// surface it traces doesn't z-fight against that surface.
const DEPTH_BIAS: f32 = 0.5;

/// Alpha-blend `rgb` (`0x__RRGGBB`) over `dst` (`0x00RRGGBB`) by `alpha`
/// (`0..=255`). Returns `0x00RRGGBB`, matching the framebuffer packing.
fn blend_rgb(dst: u32, rgb: u32, alpha: u32) -> u32 {
    if alpha >= 255 {
        return rgb & 0x00ff_ffff;
    }
    let ia = 255 - alpha;
    let r = (((rgb >> 16) & 0xff) * alpha + ((dst >> 16) & 0xff) * ia) / 255;
    let g = (((rgb >> 8) & 0xff) * alpha + ((dst >> 8) & 0xff) * ia) / 255;
    let b = ((rgb & 0xff) * alpha + (dst & 0xff) * ia) / 255;
    (r << 16) | (g << 8) | b
}

/// The CPU backend's framebuffer presenter. Native blits into a
/// `softbuffer` window surface; wasm uploads to a WebGL2 texture +
/// fullscreen quad on the canvas (no softbuffer in the browser).
#[cfg(not(target_arch = "wasm32"))]
type Presenter = softbuffer::Surface<Arc<DynDisplay>, Arc<DynWindow>>;
#[cfg(target_arch = "wasm32")]
type Presenter = crate::cpu_blit::WebGlBlit;

/// World-space view-ray direction (un-normalised) for window pixel
/// `(x, y)` under the CPU opticast projection (voxlap `setcamera`):
/// `(x − hx)·right + (y − hy)·down + hz·forward` — `camera_math`'s
/// `corn[0]` plus the per-pixel `right`/`down` steps. Standalone so
/// it's unit-testable without a window.
#[must_use]
#[allow(clippy::too_many_arguments)]
pub(crate) fn setcamera_pixel_ray(
    right: [f64; 3],
    down: [f64; 3],
    forward: [f64; 3],
    x: f64,
    y: f64,
    hx: f32,
    hy: f32,
    hz: f32,
) -> [f64; 3] {
    let (a, b, c) = (x - f64::from(hx), y - f64::from(hy), f64::from(hz));
    [
        a * right[0] + b * down[0] + c * forward[0],
        a * right[1] + b * down[1] + c * forward[1],
        a * right[2] + b * down[2] + c * forward[2],
    ]
}

pub(crate) struct CpuBackend {
    /// Framebuffer presenter — native `softbuffer` window surface, or
    /// the wasm WebGL2 canvas blitter (see [`Presenter`]). On native,
    /// `softbuffer::Context` is dropped after surface creation; the
    /// surface keeps its own clone of the type-erased `Arc<dyn …>`
    /// display/window handles so the backend stays generic-free over
    /// the host's windowing library.
    present_target: Presenter,
    /// Current framebuffer size in physical pixels. Seeded at
    /// construction, updated by [`Self::resize`] — replaces the old
    /// per-frame `window.inner_size()` poll so the backend never
    /// touches a concrete window type.
    current_dims: (u32, u32),
    pool: ScratchPool,
    zbuffer: Vec<f32>,
    /// Framebuffer dimensions of the last `render` — the `zbuffer`
    /// stride for [`Self::pick_depth`].
    last_dims: (u32, u32),
    /// Opticast projection params `(hx, hy, hz)` of the last `render`,
    /// from its [`OpticastSettings`] — the CPU unproject for
    /// [`Self::pixel_ray`].
    last_hxyz: (f32, f32, f32),
    /// Widest combined-grid `vsid` the pool's `lastx` is sized for;
    /// kept so a window grow can re-create the pool.
    max_grid_vsid: u32,
    n_threads: usize,
    clear_sky: u32,
    /// Pre-built per-instance sprites (one per [`SpriteSet`] instance,
    /// model KV6 cloned once at `set_sprites`), drawn each frame after
    /// the world via `draw_sprite`.
    sprites: Vec<Sprite>,
    /// GPU.12 incremental — source [`SpriteSet::models`] index per entry
    /// in [`sprites`](Self::sprites), so
    /// [`update_sprite_model`](Self::update_sprite_model) can swap one
    /// model's `kv6` into every instance of it without a full rebuild.
    sprite_models: Vec<usize>,
    /// Posed KFA limbs (flattened across all registered KFA sprites),
    /// refreshed by [`Self::update_kfa_poses`] and drawn after the
    /// static sprites each frame via `draw_sprite`.
    kfa_limbs: Vec<Sprite>,
    /// `F`-capture: when set, the next frame copies its composited
    /// buffer into `captured` before presenting.
    capture_next: bool,
    captured: Option<(Vec<u32>, u32, u32)>,
    /// Owned composited frame (`0x00RRGGBB`), sized `width*height` of the
    /// last [`Self::render`]. `render` composites the scene + sprites
    /// here without touching the window; [`Self::present`] blits it into
    /// the softbuffer surface and presents, and [`Self::paint_egui`]
    /// rasterises egui over it first. Decoupling the composite from the
    /// present lets a host slot a UI pass between them.
    framebuffer: Vec<u32>,
    /// egui atlas cache + software rasteriser (`hud` feature).
    #[cfg(feature = "hud")]
    egui_raster: crate::cpu_egui::EguiRaster,
}

impl CpuBackend {
    /// Shared construction: build the pool / z-buffer / framebuffer
    /// around an already-created `present_target` (native softbuffer
    /// surface or wasm WebGL2 blitter).
    fn assemble(present_target: Presenter, size: (u32, u32), opts: &RenderOptions) -> Self {
        let (w, h) = (size.0.max(1), size.1.max(1));
        let n_threads = opts
            .cpu_render_threads
            .clamp(1, rayon::current_num_threads().max(1));
        let pool = ScratchPool::new_parallel(w, h, opts.cpu_max_grid_vsid, n_threads);
        let zbuffer = vec![f32::INFINITY; (w as usize) * (h as usize)];
        let framebuffer = vec![opts.clear_sky; (w as usize) * (h as usize)];

        Self {
            present_target,
            current_dims: (w, h),
            pool,
            zbuffer,
            last_dims: (w, h),
            last_hxyz: (0.0, 0.0, 0.0),
            max_grid_vsid: opts.cpu_max_grid_vsid,
            n_threads,
            clear_sky: opts.clear_sky,
            sprites: Vec::new(),
            sprite_models: Vec::new(),
            kfa_limbs: Vec::new(),
            capture_next: false,
            captured: None,
            framebuffer,
            #[cfg(feature = "hud")]
            egui_raster: crate::cpu_egui::EguiRaster::default(),
        }
    }

    /// Native: present into a `softbuffer` surface bound to `window`.
    #[cfg(not(target_arch = "wasm32"))]
    pub(crate) fn new<W>(window: Arc<W>, size: (u32, u32), opts: &RenderOptions) -> Self
    where
        W: HasWindowHandle + HasDisplayHandle + Send + Sync + 'static,
    {
        // Erase the concrete window type behind two `Arc<dyn …>`
        // handles. `raw-window-handle` implements `HasDisplayHandle` /
        // `HasWindowHandle` for `Arc<H>` with `H: ?Sized`, and a bare
        // trait object implements its own (object-safe) trait, so both
        // erased Arcs satisfy softbuffer's bounds.
        let display: Arc<DynDisplay> = window.clone();
        let window: Arc<DynWindow> = window;
        let context = softbuffer::Context::new(display).expect("softbuffer: Context::new");
        let surface = softbuffer::Surface::new(&context, window).expect("softbuffer: Surface::new");
        Self::assemble(surface, size, opts)
    }

    /// wasm: present into a WebGL2 blitter over `canvas` (no softbuffer
    /// in the browser).
    #[cfg(target_arch = "wasm32")]
    pub(crate) fn new_from_canvas(
        canvas: web_sys::HtmlCanvasElement,
        size: (u32, u32),
        opts: &RenderOptions,
    ) -> Self {
        let (w, h) = (size.0.max(1), size.1.max(1));
        let blit = crate::cpu_blit::WebGlBlit::new(&canvas, w, h)
            .expect("roxlap-render: WebGL2 blit init");
        Self::assemble(blit, size, opts)
    }

    /// Request that the next rendered frame be captured for readback.
    pub(crate) fn request_capture(&mut self) {
        self.capture_next = true;
    }

    /// Take the most recently captured frame, if any.
    pub(crate) fn take_capture(&mut self) -> Option<(Vec<u32>, u32, u32)> {
        self.captured.take()
    }

    /// World-space view-ray direction (un-normalised) for pixel
    /// `(x, y)` under the CPU opticast projection (voxlap `setcamera`):
    /// `(x - hx)·right + (y - hy)·down + hz·forward`, using the last
    /// frame's `(hx, hy, hz)`. `None` before the first render.
    pub(crate) fn pixel_ray(&self, camera: &Camera, x: f64, y: f64) -> Option<[f64; 3]> {
        let (hx, hy, hz) = self.last_hxyz;
        if hz <= 0.0 {
            return None;
        }
        Some(setcamera_pixel_ray(
            camera.right,
            camera.down,
            camera.forward,
            x,
            y,
            hx,
            hy,
            hz,
        ))
    }

    /// World-t depth at pixel `(x, y)` from the last frame's z-buffer
    /// (already in CPU memory — no readback). `None` for out-of-bounds
    /// or sky (`+INF`). See [`SceneRenderer::pick_depth`].
    pub(crate) fn pick_depth(&self, x: u32, y: u32) -> Option<f32> {
        let (w, h) = self.last_dims;
        if x >= w || y >= h {
            return None;
        }
        let t = *self.zbuffer.get((y * w + x) as usize)?;
        if t.is_finite() {
            Some(t)
        } else {
            None
        }
    }

    /// Pre-build one [`Sprite`] per instance (model KV6 cloned, the
    /// instance position applied) so per-frame drawing never re-clones.
    pub(crate) fn set_sprites(&mut self, set: &SpriteSet) {
        let mut sprites = Vec::with_capacity(set.instances.len());
        let mut sprite_models = Vec::with_capacity(set.instances.len());
        for inst in &set.instances {
            if let Some(model) = set.models.get(inst.model) {
                let mut s = model.clone();
                s.p = inst.pos;
                sprites.push(s);
                sprite_models.push(inst.model);
            }
        }
        self.sprites = sprites;
        self.sprite_models = sprite_models;
    }

    /// GPU.12 incremental — swap the edited `kv6` into every cached
    /// instance of host model `model_index`, keeping each instance's
    /// world position. Mirrors the GPU backend's single-model update on
    /// the software path (where "rebuild" is just a kv6 clone per
    /// instance, so the win is parity rather than bandwidth). No-op if no
    /// instance references `model_index`.
    pub(crate) fn update_sprite_model(&mut self, model_index: usize, kv6: &Kv6) {
        for (s, &m) in self.sprites.iter_mut().zip(&self.sprite_models) {
            if m == model_index {
                s.kv6 = kv6.clone();
            }
        }
    }

    /// Register KFA sprites — for the CPU backend this is the same as a
    /// pose refresh: solve every limb's world transform from its
    /// current `kfaval[]` and cache the resulting [`Sprite`]s.
    pub(crate) fn set_kfa_sprites(&mut self, kfas: &mut [KfaSprite]) {
        self.update_kfa_poses(kfas);
    }

    /// Re-solve every KFA limb's world transform and cache the posed
    /// [`Sprite`]s for the next [`Self::render`].
    pub(crate) fn update_kfa_poses(&mut self, kfas: &mut [KfaSprite]) {
        self.kfa_limbs.clear();
        for kfa in kfas.iter_mut() {
            solve_kfa_limbs(kfa);
            self.kfa_limbs.extend(kfa.limbs.iter().cloned());
        }
    }

    pub(crate) fn resize(&mut self, width: u32, height: u32) {
        // softbuffer + the pool resize lazily inside `render`; we just
        // record the new size the host reported (replacing the old
        // per-frame `window.inner_size()` poll). The WebGL2 blitter's
        // texture, by contrast, must be re-allocated eagerly.
        self.current_dims = (width.max(1), height.max(1));
        #[cfg(target_arch = "wasm32")]
        self.present_target
            .resize(self.current_dims.0, self.current_dims.1);
    }

    pub(crate) fn render(&mut self, scene: &mut Scene, camera: &Camera, frame: &FrameParams) {
        let (width, height) = self.current_dims;
        if width == 0 || height == 0 {
            return;
        }
        let pixel_count = (width as usize) * (height as usize);
        self.last_dims = (width, height);
        self.last_hxyz = (frame.settings.hx, frame.settings.hy, frame.settings.hz);

        // Grow the z-buffer + pool to follow a window resize.
        if self.zbuffer.len() < pixel_count {
            self.zbuffer.resize(pixel_count, f32::INFINITY);
        }
        if self.pool.slot(0).uurend_half_stride < width as usize {
            self.pool =
                ScratchPool::new_parallel(width, height, self.max_grid_vsid, self.n_threads);
        }

        // Per-frame pool config (engine sky/fog → rasterizer). The
        // rasterizer takes packed colours as `i32`; reinterpret the
        // bits (not a numeric cast).
        let sky_i = i32::from_ne_bytes(frame.sky_color.to_ne_bytes());
        self.pool.set_skycast(sky_i, 0);
        let fog_i = i32::from_ne_bytes(frame.fog_color.to_ne_bytes());
        self.pool.set_fog(fog_i, frame.fog_max_scan_dist);
        self.pool.set_treat_z_max_as_air(frame.treat_z_max_as_air);
        // Per-face grid shading (voxlap setsideshades) — the grid-scan
        // analogue of sprite_lighting. Default [0;6] keeps sideshademode
        // off (byte-identical to the no-side-shade path).
        let [top, bot, left, right, up, down] = frame.side_shades;
        self.pool.set_side_shades(top, bot, left, right, up, down);

        // Composite into the owned framebuffer (not the window) so the
        // present can be deferred — a host may paint a UI over it first.
        // `render_scene_composed` convention: caller pre-fills the
        // framebuffer with sky + the z-buffer with +INF, then it
        // z-merges every grid in.
        if self.framebuffer.len() < pixel_count {
            self.framebuffer.resize(pixel_count, self.clear_sky);
        }
        let fb = &mut self.framebuffer[..pixel_count];
        for px in fb.iter_mut() {
            *px = self.clear_sky;
        }
        for z in &mut self.zbuffer[..pixel_count] {
            *z = f32::INFINITY;
        }

        let _outcome = render_scene_composed(
            fb,
            &mut self.zbuffer[..pixel_count],
            width as usize,
            width,
            height,
            &mut self.pool,
            scene,
            camera,
            frame.settings,
            frame.sky_color,
            frame.sky,
        );

        // Sprites layer on top of the heightmap world, z-tested against
        // the same z-buffer (camera-facing voxel splat). Needs the
        // host-built lighting; skipped if absent or no sprites.
        if let Some(lighting) = frame.sprite_lighting {
            if !self.sprites.is_empty() || !self.kfa_limbs.is_empty() {
                let cam_state = camera_math::derive(
                    camera,
                    width,
                    height,
                    frame.settings.hx,
                    frame.settings.hy,
                    frame.settings.hz,
                );
                let mut target = DrawTarget::new(
                    fb,
                    &mut self.zbuffer[..pixel_count],
                    width as usize,
                    width,
                    height,
                );
                // Static sprites, then the posed KFA limbs (already
                // solved by `update_kfa_poses`); both z-test against the
                // shared buffer so order doesn't affect the result.
                for sprite in self.sprites.iter().chain(self.kfa_limbs.iter()) {
                    let _written =
                        draw_sprite(&mut target, &cam_state, frame.settings, lighting, sprite);
                }
            }
        }

        if self.capture_next {
            self.capture_next = false;
            self.captured = Some((fb.to_vec(), width, height));
        }
        // No present here — the host calls `present` or `paint_egui`.
    }

    /// Blit the composited [`Self::framebuffer`] into the softbuffer
    /// surface and present it. The no-UI counterpart to
    /// [`Self::paint_egui`]; both finish the frame `render` started.
    pub(crate) fn present(&mut self) {
        self.blit_and_present(self.last_dims);
    }

    /// Rasterise depth-tested world-space [`Line3`] segments over the
    /// framebuffer the last [`render`](Self::render) composited. Uses that
    /// frame's pinhole projection (`last_hxyz` / `last_dims`) and z-buffer
    /// (perpendicular distance, smaller = closer, sky = `+inf`), so the
    /// rendered terrain occludes lines behind it. Call after `render`,
    /// before `present` / `paint_egui`.
    pub(crate) fn draw_lines(&mut self, camera: &Camera, lines: &[Line3]) {
        let (w, h) = self.last_dims;
        let (hx, hy, hz) = self.last_hxyz;
        if w == 0 || h == 0 || hz <= 0.0 {
            return; // nothing rendered yet — no projection to reuse
        }
        let pixel_count = (w as usize) * (h as usize);
        if self.framebuffer.len() < pixel_count || self.zbuffer.len() < pixel_count {
            return;
        }
        let cam = camera_math::derive(camera, w, h, hx, hy, hz);
        // World point → camera-relative (right, down, forward) coords.
        // The forward component is the CPU z-buffer metric (perpendicular
        // distance); right/down drive the pinhole screen projection.
        let cam_coords = |p: [f32; 3]| -> [f32; 3] {
            let d = [p[0] - cam.pos[0], p[1] - cam.pos[1], p[2] - cam.pos[2]];
            [
                cam.right[0] * d[0] + cam.right[1] * d[1] + cam.right[2] * d[2],
                cam.down[0] * d[0] + cam.down[1] * d[1] + cam.down[2] * d[2],
                cam.forward[0] * d[0] + cam.forward[1] * d[1] + cam.forward[2] * d[2],
            ]
        };

        let fb = &mut self.framebuffer[..pixel_count];
        let zb = &self.zbuffer[..pixel_count];
        let (wi, hi) = (w as i32, h as i32);

        for line in lines {
            let a = [line.a[0] as f32, line.a[1] as f32, line.a[2] as f32];
            let b = [line.b[0] as f32, line.b[1] as f32, line.b[2] as f32];
            let ca = cam_coords(a);
            let cb = cam_coords(b);

            // Near-plane clip in segment-parameter space (forward depth
            // `cz >= NEAR_Z`) so the pinhole divide stays finite and points
            // behind the camera don't wrap. Both behind → invisible.
            let (cza, czb) = (ca[2], cb[2]);
            if cza < NEAR_Z && czb < NEAR_Z {
                continue;
            }
            let (mut t0, mut t1) = (0.0f32, 1.0f32);
            let dz = czb - cza;
            if dz.abs() > f32::EPSILON {
                let t_near = (NEAR_Z - cza) / dz;
                if dz > 0.0 {
                    t0 = t0.max(t_near); // a is behind: enter at the near plane
                } else {
                    t1 = t1.min(t_near); // b is behind: leave at the near plane
                }
            }
            if t0 > t1 {
                continue;
            }
            let lerp3 = |t: f32| {
                [
                    ca[0] + (cb[0] - ca[0]) * t,
                    ca[1] + (cb[1] - ca[1]) * t,
                    ca[2] + (cb[2] - ca[2]) * t,
                ]
            };
            let p0 = lerp3(t0);
            let p1 = lerp3(t1);

            // Pinhole project; carry 1/cz for perspective-correct depth
            // (1/cz is linear in screen space, cz is not).
            let inv0 = 1.0 / p0[2];
            let inv1 = 1.0 / p1[2];
            let sx0 = hx + p0[0] * hz * inv0;
            let sy0 = hy + p0[1] * hz * inv0;
            let sx1 = hx + p1[0] * hz * inv1;
            let sy1 = hy + p1[1] * hz * inv1;

            let alpha = (line.color >> 24) & 0xff;
            if alpha == 0 {
                continue; // fully transparent
            }
            let rgb = line.color & 0x00ff_ffff;

            // DDA along the dominant screen axis, stamping `width_px`
            // pixels perpendicular to the segment.
            let dx = sx1 - sx0;
            let dy = sy1 - sy0;
            let steps = dx.abs().max(dy.abs()).ceil().max(1.0);
            let len = (dx * dx + dy * dy).sqrt().max(1e-6);
            let (perp_x, perp_y) = (-dy / len, dx / len);
            let half = ((line.width_px - 1.0).max(0.0) * 0.5).round() as i32;

            let nsteps = steps as i32;
            for s in 0..=nsteps {
                let t = s as f32 / steps;
                let inv_z = inv0 + (inv1 - inv0) * t;
                let depth = 1.0 / inv_z; // perpendicular distance at this pixel
                let cx = sx0 + dx * t;
                let cy = sy0 + dy * t;
                for woff in -half..=half {
                    let px = (cx + perp_x * woff as f32).round() as i32;
                    let py = (cy + perp_y * woff as f32).round() as i32;
                    if px < 0 || py < 0 || px >= wi || py >= hi {
                        continue;
                    }
                    let idx = (py as usize) * (w as usize) + (px as usize);
                    if line.depth_test && depth > zb[idx] + DEPTH_BIAS {
                        continue; // occluded by nearer rendered geometry
                    }
                    fb[idx] = blend_rgb(fb[idx], rgb, alpha);
                }
            }
        }
    }

    /// Shared tail of `present` / `paint_egui`: copy the framebuffer to
    /// the window surface at `(width, height)` and present.
    #[cfg(not(target_arch = "wasm32"))]
    fn blit_and_present(&mut self, dims: (u32, u32)) {
        let (width, height) = dims;
        let (Some(w_nz), Some(h_nz)) = (NonZeroU32::new(width), NonZeroU32::new(height)) else {
            return;
        };
        let pixel_count = (width as usize) * (height as usize);
        if self.framebuffer.len() < pixel_count {
            return;
        }
        self.present_target
            .resize(w_nz, h_nz)
            .expect("softbuffer: resize");
        let mut buffer = self
            .present_target
            .buffer_mut()
            .expect("softbuffer: buffer_mut");
        buffer[..pixel_count].copy_from_slice(&self.framebuffer[..pixel_count]);
        buffer.present().expect("softbuffer: present");
    }

    /// wasm counterpart: upload the framebuffer to the WebGL2 texture
    /// and draw the fullscreen quad on the canvas.
    #[cfg(target_arch = "wasm32")]
    fn blit_and_present(&mut self, dims: (u32, u32)) {
        let (width, height) = dims;
        let pixel_count = (width as usize) * (height as usize);
        if width == 0 || height == 0 || self.framebuffer.len() < pixel_count {
            return;
        }
        self.present_target.resize(width, height);
        self.present_target
            .present(&self.framebuffer[..pixel_count]);
    }

    /// Software-rasterise the egui `jobs` over the composited
    /// framebuffer, then present (`hud` feature). Replaces
    /// [`Self::present`] for the UI-overlay path.
    #[cfg(feature = "hud")]
    pub(crate) fn paint_egui(
        &mut self,
        jobs: &[egui::ClippedPrimitive],
        textures: &egui::TexturesDelta,
        pixels_per_point: f32,
    ) {
        let (width, height) = self.last_dims;
        let pixel_count = (width as usize) * (height as usize);
        if self.framebuffer.len() < pixel_count {
            return;
        }
        self.egui_raster
            .update_textures(&textures.set, &textures.free);
        self.egui_raster.paint(
            &mut self.framebuffer[..pixel_count],
            width,
            height,
            jobs,
            pixels_per_point,
        );
        self.blit_and_present((width, height));
    }
}

#[cfg(test)]
mod cpu_ray_tests {
    use super::setcamera_pixel_ray;

    const RIGHT: [f64; 3] = [1.0, 0.0, 0.0];
    const DOWN: [f64; 3] = [0.0, 1.0, 0.0];
    const FWD: [f64; 3] = [0.0, 0.0, 1.0]; // voxlap z-down "look down"

    // Centre pixel (hx, hy) → straight along `forward`.
    #[test]
    fn centre_pixel_is_forward() {
        let d = setcamera_pixel_ray(RIGHT, DOWN, FWD, 320.0, 240.0, 320.0, 240.0, 320.0);
        assert_eq!(d, [0.0, 0.0, 320.0]);
    }

    // Off-centre pixel tilts proportionally: (px-hx, py-hy, hz).
    #[test]
    fn offcentre_pixel_tilts_linearly() {
        let d = setcamera_pixel_ray(RIGHT, DOWN, FWD, 384.0, 272.0, 320.0, 240.0, 320.0);
        assert_eq!(d, [64.0, 32.0, 320.0]);
    }
}

#[cfg(test)]
mod blend_tests {
    use super::blend_rgb;

    #[test]
    fn opaque_replaces_destination() {
        // alpha = 255 → source colour, ignoring the destination.
        assert_eq!(blend_rgb(0x00_12_34_56, 0xAA_BB_CC, 255), 0x00_AA_BB_CC);
    }

    #[test]
    fn zero_alpha_keeps_destination() {
        // alpha = 0 → 100% destination (the caller skips alpha==0, but
        // the blend itself must still be a no-op).
        assert_eq!(blend_rgb(0x00_12_34_56, 0xAA_BB_CC, 0), 0x00_12_34_56);
    }

    #[test]
    fn half_alpha_is_midpoint() {
        // white over black at ~50% → mid grey, per channel (255*128/255).
        let out = blend_rgb(0x00_00_00_00, 0x00_FF_FF_FF, 128);
        assert_eq!(out, 0x00_80_80_80);
    }

    #[test]
    fn result_has_no_high_byte() {
        // Output must stay 0x00RRGGBB to match the framebuffer packing.
        assert_eq!(
            blend_rgb(0x00_FF_FF_FF, 0xFF_FF_FF_FF, 200) & 0xFF00_0000,
            0
        );
    }
}