@bornengine/engine 0.4.16

package/native/shared/src/renderer/planar_reflection.rs

@@ -0,0 +1,499 @@
+//! EN-011 — Planar reflection probes.
+//!
+//! A planar reflection probe owns an off-screen RT (`Rgba16Float` HDR
+//! colour + `Depth32Float` depth) into which the engine renders the
+//! scene from a camera mirrored across a flat reflective plane. The
+//! resulting texture is bound at `@group(2) @binding(12)` on materials
+//! that opt in via `Renderer::set_material_reflection_probe`, so e.g.
+//! a water shader can sample the *actual* trees / bridge above the
+//! surface instead of just the static `env_tex` skybox.
+//!
+//! ## V1 design
+//!
+//! - **One probe per plane.** A river is one plane; lakes are one
+//!   plane. Multi-probe blending lives in V2.
+//! - **Rebuilt every frame** at the requested resolution. Half of the
+//!   swapchain width × height is the typical caller-supplied number;
+//!   nothing in this module enforces that — the FFI just defaults to
+//!   it when the game passes 0.
+//! - **Cull list = "everything in the opaque material bucket minus a
+//!   hardcoded exclude list".** Implemented in `Renderer` (this
+//!   module just owns the probe RT + plane parameters); the dispatch
+//!   loop walks `material_system.commands` and skips the same
+//!   material handles for every probe.
+//!
+//! ## Coordinate convention
+//!
+//! The reflection plane is defined by `plane_y` (a single y-value
+//! offset from origin) and a unit `normal`. We mirror world-space
+//! points across plane `n · p = d`, where `d = n · plane_origin` and
+//! `plane_origin = (0, plane_y, 0)`. For the typical horizontal water
+//! surface, `normal = (0, 1, 0)` and `d = plane_y`. Non-axis-aligned
+//! planes work too — the math is general — but the FFI surface
+//! exposes the simpler horizontal case explicitly.
+//!
+//! See `Renderer::dispatch_planar_reflections` for the per-frame
+//! render-graph node that drives all registered probes.
+
+use crate::renderer::util::{mat4_invert, mat4_mul_vec4, mat4_multiply};
+
+/// One planar reflection probe + its dedicated RT pair.
+///
+/// The texture views (`color_view`, `depth_view`) are stable for the
+/// lifetime of the probe, so per-material bind groups built once at
+/// `set_material_reflection_probe` time stay valid frame after frame
+/// even as the engine repaints the texture each frame.
+pub struct PlanarReflectionProbe {
+    /// World-space y of the reflective plane (only used to build `d`
+    /// when normal == +Y; otherwise carried for diagnostics).
+    pub plane_y: f32,
+    /// Unit normal of the plane in world space. The mirror matrix
+    /// reflects across `n · p = d`, where `d = n · (0, plane_y, 0)`.
+    pub normal:  [f32; 3],
+    /// Texture extent — width = height for a square probe; both
+    /// dimensions get rounded to ≥ 16 px in `new` so a tiny RT can't
+    /// crash the renderer.
+    pub resolution: u32,
+
+    pub color_rt:    wgpu::Texture,
+    pub color_view:  wgpu::TextureView,
+    pub depth_rt:    wgpu::Texture,
+    pub depth_view:  wgpu::TextureView,
+
+    /// Dummy G-buffer attachments for the user-material probe pass.
+    /// Opaque-profile material pipelines target the full 4-attachment
+    /// opaque layout (hdr + material + velocity + albedo), so the
+    /// probe's material pass must present the same four attachments —
+    /// wgpu validates pipeline targets against pass attachments
+    /// exactly. Only the hdr result is kept; these three are cleared
+    /// each frame and their stores discarded.
+    pub aux_material_rt:   wgpu::Texture,
+    pub aux_material_view: wgpu::TextureView,
+    pub aux_velocity_rt:   wgpu::Texture,
+    pub aux_velocity_view: wgpu::TextureView,
+    pub aux_albedo_rt:     wgpu::Texture,
+    pub aux_albedo_view:   wgpu::TextureView,
+}
+
+impl PlanarReflectionProbe {
+    /// Allocate the colour + depth textures sized to `resolution²`.
+    /// Format choices match the engine's HDR pipeline so the probe
+    /// texture interoperates cleanly with the rest of `material_abi`:
+    ///
+    ///   - colour: `Rgba16Float` (`HDR_FORMAT`) so emissive geometry
+    ///     reflected into the probe doesn't clamp to LDR
+    ///   - depth: `Depth32Float` so the mirrored draws can z-test
+    ///     against each other without a separate downsample
+    pub fn new(
+        device: &wgpu::Device,
+        plane_y: f32,
+        normal:  [f32; 3],
+        resolution: u32,
+    ) -> Self {
+        // Clamp absurd inputs — a 0-px texture is illegal in wgpu
+        // and a 16k-px probe would cost more than the rest of the
+        // frame combined. 16..=4096 covers every realistic case.
+        let res = resolution.clamp(16, 4096);
+
+        let color_rt = device.create_texture(&wgpu::TextureDescriptor {
+            label: Some("planar_reflection_color"),
+            size: wgpu::Extent3d { width: res, height: res, depth_or_array_layers: 1 },
+            mip_level_count: 1,
+            sample_count: 1,
+            dimension: wgpu::TextureDimension::D2,
+            format: super::formats::HDR_FORMAT,
+            usage: wgpu::TextureUsages::RENDER_ATTACHMENT
+                 | wgpu::TextureUsages::TEXTURE_BINDING,
+            view_formats: &[],
+        });
+        let color_view = color_rt.create_view(&Default::default());
+
+        let depth_rt = device.create_texture(&wgpu::TextureDescriptor {
+            label: Some("planar_reflection_depth"),
+            size: wgpu::Extent3d { width: res, height: res, depth_or_array_layers: 1 },
+            mip_level_count: 1,
+            sample_count: 1,
+            dimension: wgpu::TextureDimension::D2,
+            format: super::formats::DEPTH_FORMAT,
+            usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
+            view_formats: &[],
+        });
+        let depth_view = depth_rt.create_view(&Default::default());
+
+        // Aux G-buffer dummies — see the struct field comment. Formats
+        // must byte-match what `Renderer::compile_material` passes to
+        // the pipeline descriptors or the probe pass fails validation.
+        let make_aux = |label: &str, format: wgpu::TextureFormat| {
+            let tex = device.create_texture(&wgpu::TextureDescriptor {
+                label: Some(label),
+                size: wgpu::Extent3d { width: res, height: res, depth_or_array_layers: 1 },
+                mip_level_count: 1,
+                sample_count: 1,
+                dimension: wgpu::TextureDimension::D2,
+                format,
+                usage: wgpu::TextureUsages::RENDER_ATTACHMENT,
+                view_formats: &[],
+            });
+            let view = tex.create_view(&Default::default());
+            (tex, view)
+        };
+        let (aux_material_rt, aux_material_view) =
+            make_aux("planar_reflection_aux_material", super::formats::MATERIAL_FORMAT);
+        let (aux_velocity_rt, aux_velocity_view) =
+            make_aux("planar_reflection_aux_velocity", super::formats::VELOCITY_FORMAT);
+        let (aux_albedo_rt, aux_albedo_view) =
+            make_aux("planar_reflection_aux_albedo", wgpu::TextureFormat::Rgba8Unorm);
+
+        // Normalise the supplied normal — caller may pass a non-unit
+        // vector; downstream math (specifically the reflection
+        // matrix below) assumes |n| == 1.
+        let n = normalise(normal);
+
+        Self {
+            plane_y, normal: n, resolution: res,
+            color_rt, color_view, depth_rt, depth_view,
+            aux_material_rt, aux_material_view,
+            aux_velocity_rt, aux_velocity_view,
+            aux_albedo_rt, aux_albedo_view,
+        }
+    }
+}
+
+/// Build the world-space reflection matrix for plane (n, plane_y).
+///
+/// Reflects a world-space point `p` across the plane `n · p = d`
+/// where `d = n · (0, plane_y, 0) = n.y * plane_y`. The returned 4×4
+/// matrix R has `R · p = p - 2 (n·p - d) n`, with R applied
+/// post-multiply on column vectors (Bloom convention).
+///
+/// Plug this into the view chain via `mirror_view = view * R` —
+/// post-multiplying R into the view matrix means: world → mirror
+/// (R) → camera (view).
+pub fn reflection_matrix(plane_y: f32, normal: [f32; 3]) -> [[f32; 4]; 4] {
+    let n = normalise(normal);
+    let d = n[1] * plane_y;
+    // Standard Householder-style reflection across the plane.
+    // Column-major; multiplies p as `R * p` (post-mul).
+    let nx = n[0]; let ny = n[1]; let nz = n[2];
+    [
+        [1.0 - 2.0 * nx * nx, -2.0 * nx * ny,       -2.0 * nx * nz,       0.0],
+        [-2.0 * ny * nx,        1.0 - 2.0 * ny * ny, -2.0 * ny * nz,       0.0],
+        [-2.0 * nz * nx,       -2.0 * nz * ny,        1.0 - 2.0 * nz * nz, 0.0],
+        [2.0 * nx * d,          2.0 * ny * d,         2.0 * nz * d,         1.0],
+    ]
+}
+
+/// Compose a mirrored view matrix from the camera's current view
+/// matrix and the reflection plane.
+///
+/// In Bloom's column-major / post-mul convention, `view` transforms
+/// world points into camera space (`p_cam = view * p_world`). To
+/// render the mirror image we want `p_cam = view * R * p_world` —
+/// reflect first, then apply the existing view. The returned matrix
+/// is `view * R`.
+///
+/// **Caller MUST flip front-face cull mode** when using this view —
+/// reflection inverts triangle winding. The renderer handles that
+/// by binding pipelines compiled with `cull_mode = None` for the
+/// mirrored pass; we accept the small fragment-shader cost for V1.
+pub fn mirrored_view(view: [[f32; 4]; 4], plane_y: f32, normal: [f32; 3]) -> [[f32; 4]; 4] {
+    let r = reflection_matrix(plane_y, normal);
+    mat4_multiply(view, r)
+}
+
+/// Reflect the camera's world position across the plane. Used for
+/// `PerView.camera_pos` in the mirrored UBO so view-dependent shading
+/// (Fresnel, specular, parallax) sees the mirror camera, not the
+/// real one.
+pub fn mirrored_camera_pos(pos: [f32; 3], plane_y: f32, normal: [f32; 3]) -> [f32; 3] {
+    let n = normalise(normal);
+    let d = n[1] * plane_y;
+    let dist = n[0] * pos[0] + n[1] * pos[1] + n[2] * pos[2] - d;
+    [
+        pos[0] - 2.0 * dist * n[0],
+        pos[1] - 2.0 * dist * n[1],
+        pos[2] - 2.0 * dist * n[2],
+    ]
+}
+
+/// Recompute `inv_proj` from a possibly-modified projection. After the
+/// EN-011 V2 `oblique_proj` rewrite, the reflection pass uses a near-
+/// plane-clipped projection that differs from the main camera's, so
+/// the inverse needs to be recomputed for any view-space reconstruction
+/// downstream (SSR, deferred unprojection).
+pub fn inv_proj_for(proj: [[f32; 4]; 4]) -> [[f32; 4]; 4] {
+    mat4_invert(proj)
+}
+
+/// Transform a world-space plane `(Nx, Ny, Nz, d)` (such that
+/// `N · p + d = 0` for points on the plane) into eye / view space
+/// using the view matrix. Plane equations transform by the inverse-
+/// transpose of the matrix; for a rigid view matrix this is the same
+/// as the matrix itself, but we do it the general way to stay safe
+/// for skewed views.
+///
+/// For Bloom's column-major / post-mul convention, applying the
+/// inverse-transpose of `view` to a plane `(N, d)` equates to the
+/// standard direct-3D / OpenGL formulation.
+pub fn world_plane_to_eye_space(
+    view: [[f32; 4]; 4],
+    plane_world: [f32; 4],
+) -> [f32; 4] {
+    let inv = mat4_invert(view);
+    // Inverse-transpose, column-major. Multiplying the *transpose* of
+    // `inv` by the plane vector is the same as multiplying `inv`
+    // post-multiplied by the row vector — but we have a column-vec
+    // mul helper so we transpose `inv` first by reading rows as cols.
+    let inv_t: [[f32; 4]; 4] = [
+        [inv[0][0], inv[1][0], inv[2][0], inv[3][0]],
+        [inv[0][1], inv[1][1], inv[2][1], inv[3][1]],
+        [inv[0][2], inv[1][2], inv[2][2], inv[3][2]],
+        [inv[0][3], inv[1][3], inv[2][3], inv[3][3]],
+    ];
+    mat4_mul_vec4(&inv_t, &plane_world)
+}
+
+/// EN-011 V2 — modify a projection matrix so its near plane is clipped
+/// at the given eye-space plane. Used for planar reflection to prevent
+/// geometry below the water from polluting the reflection along the
+/// shoreline edge.
+///
+/// Reference: Eric Lengyel, "Oblique View Frustum Depth Projection
+/// and Clipping" (Journal of Game Development, 2005). The technique
+/// shifts the projection's near plane to coincide with the supplied
+/// plane (typically the water plane), then any geometry on the wrong
+/// side gets clipped at the rasterizer.
+///
+/// `proj` is the original column-major projection matrix.
+/// `plane_eye_space` is `(Nx, Ny, Nz, d)` such that points on the
+/// plane satisfy `N · p_eye + d = 0`. Use `world_plane_to_eye_space`
+/// to convert from a world-space plane.
+pub fn oblique_proj(
+    proj: [[f32; 4]; 4],
+    plane_eye_space: [f32; 4],
+) -> [[f32; 4]; 4] {
+    let c = plane_eye_space;
+
+    // Far-plane corner in clip space is in the direction of (sgn(c.x),
+    // sgn(c.y), 1, 1) — Lengyel §2. Pulled back into eye space by
+    // multiplying with `inv(proj)`.
+    let sx = if c[0] >= 0.0 { 1.0 } else { -1.0 };
+    let sy = if c[1] >= 0.0 { 1.0 } else { -1.0 };
+    let q_clip = [sx, sy, 1.0, 1.0];
+    let inv_p = mat4_invert(proj);
+    let q = mat4_mul_vec4(&inv_p, &q_clip);
+
+    // Scale `c` so the near-plane crosses through the supplied plane:
+    //   M = (2 / dot(c, q)) · c
+    // Then the new third row of P (which controls the depth output) is
+    //   P_row2 = M - P_row3
+    // (P_row3 is the standard perspective w-row, all stays the same.)
+    let denom = c[0] * q[0] + c[1] * q[1] + c[2] * q[2] + c[3] * q[3];
+    if denom.abs() < 1e-10 {
+        // Degenerate plane orientation w.r.t. the frustum — leave
+        // projection unchanged rather than divide-by-zero. The
+        // reflection still renders, just without near-plane clipping.
+        return proj;
+    }
+    let scale = 2.0 / denom;
+    let m = [c[0] * scale, c[1] * scale, c[2] * scale, c[3] * scale];
+
+    // proj is column-major: proj[col][row]. The "third row" we want
+    // to replace is at row index 2 across all four columns. The
+    // "fourth row" is at row index 3. New row-2 = M - row3 — but
+    // since wgpu's clip-space z range is [0, 1] (not [-1, 1] like
+    // OpenGL), the depth-rescaling pre-step is `M - P_row3` exactly
+    // as in Lengyel's original derivation; the [-1, 1] vs [0, 1]
+    // difference is absorbed by the scale.
+    let mut out = proj;
+    out[0][2] = m[0] - proj[0][3];
+    out[1][2] = m[1] - proj[1][3];
+    out[2][2] = m[2] - proj[2][3];
+    out[3][2] = m[3] - proj[3][3];
+    out
+}
+
+fn normalise(v: [f32; 3]) -> [f32; 3] {
+    let len_sq = v[0] * v[0] + v[1] * v[1] + v[2] * v[2];
+    if len_sq < 1e-10 {
+        // Default to +Y — a horizontal mirror like a calm lake.
+        return [0.0, 1.0, 0.0];
+    }
+    let inv = 1.0 / len_sq.sqrt();
+    [v[0] * inv, v[1] * inv, v[2] * inv]
+}
+
+// =====================================================================
+// Tests
+// =====================================================================
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::renderer::util::mat4_mul_vec4;
+
+    /// A point above a horizontal y=0 plane reflects to the mirror
+    /// position below it — y-coordinate flips sign, x and z preserved.
+    #[test]
+    fn reflection_matrix_mirrors_above_to_below() {
+        let r = reflection_matrix(0.0, [0.0, 1.0, 0.0]);
+        let p = [3.0_f32, 5.0, -7.0, 1.0];
+        let out = mat4_mul_vec4(&r, &p);
+        assert!((out[0] - 3.0).abs() < 1e-5, "x preserved (got {})", out[0]);
+        assert!((out[1] + 5.0).abs() < 1e-5, "y flipped (got {})", out[1]);
+        assert!((out[2] + 7.0).abs() < 1e-5, "z preserved (got {})", out[2]);
+        assert!((out[3] - 1.0).abs() < 1e-5, "w preserved");
+    }
+
+    /// A non-zero `plane_y` shifts the mirror — point at y=10 across
+    /// plane y=2 lands at y = 2 - (10 - 2) = -6.
+    #[test]
+    fn reflection_matrix_offset_plane() {
+        let r = reflection_matrix(2.0, [0.0, 1.0, 0.0]);
+        let p = [0.0_f32, 10.0, 0.0, 1.0];
+        let out = mat4_mul_vec4(&r, &p);
+        assert!((out[1] - (-6.0)).abs() < 1e-4, "y reflected across y=2 (got {})", out[1]);
+    }
+
+    /// Camera position helper agrees with applying the matrix to a
+    /// (cam_pos, 1) vec4 — sanity check.
+    #[test]
+    fn mirrored_camera_pos_matches_matrix() {
+        let plane_y = 0.5;
+        let n = [0.0, 1.0, 0.0];
+        let cam = [1.5_f32, 4.0, -2.0];
+
+        let helper_out = mirrored_camera_pos(cam, plane_y, n);
+        let r = reflection_matrix(plane_y, n);
+        let mat_out = mat4_mul_vec4(&r, &[cam[0], cam[1], cam[2], 1.0]);
+
+        assert!((helper_out[0] - mat_out[0]).abs() < 1e-5);
+        assert!((helper_out[1] - mat_out[1]).abs() < 1e-5);
+        assert!((helper_out[2] - mat_out[2]).abs() < 1e-5);
+    }
+
+    /// Reflection is its own inverse — applying it twice returns
+    /// the original point.
+    #[test]
+    fn reflection_is_involution() {
+        let r = reflection_matrix(0.5, [0.0, 1.0, 0.0]);
+        let p = [3.0_f32, 5.0, -7.0, 1.0];
+        let once = mat4_mul_vec4(&r, &p);
+        let twice = mat4_mul_vec4(&r, &once);
+        for i in 0..4 {
+            assert!((twice[i] - p[i]).abs() < 1e-4, "comp {} mismatch", i);
+        }
+    }
+
+    /// Headless wgpu device. Mirrors the EN-006 pattern from
+    /// `transient.rs` / `impulse_field.rs`. Returns `None` when no GPU
+    /// is available so the test skips gracefully on bare CI.
+    fn try_create_device() -> Option<(wgpu::Device, wgpu::Queue)> {
+        let instance = wgpu::Instance::new(wgpu::InstanceDescriptor {
+            backends: wgpu::Backends::all(),
+            ..wgpu::InstanceDescriptor::new_without_display_handle()
+        });
+        let adapter = pollster::block_on(instance.request_adapter(&wgpu::RequestAdapterOptions {
+            power_preference: wgpu::PowerPreference::LowPower,
+            compatible_surface: None,
+            force_fallback_adapter: true,
+        })).ok()?;
+        let (device, queue) = pollster::block_on(adapter.request_device(
+            &wgpu::DeviceDescriptor {
+                label: Some("planar-reflection-test-device"),
+                required_features: wgpu::Features::empty(),
+                required_limits: wgpu::Limits::downlevel_defaults(),
+                ..Default::default()
+            },
+        )).ok()?;
+        Some((device, queue))
+    }
+
+    /// EN-011 — verify `PlanarReflectionProbe::new` allocates HDR
+    /// colour + Depth32 attachments at the requested resolution.
+    /// Pure construction test — render-graph integration lives in
+    /// `Renderer::dispatch_planar_reflections`.
+    #[test]
+    fn probe_creation_allocates_hdr_and_depth_attachments() {
+        let Some((device, _queue)) = try_create_device() else { return; };
+        let probe = PlanarReflectionProbe::new(&device, 0.5, [0.0, 1.0, 0.0], 256);
+        assert_eq!(probe.resolution, 256, "resolution clamped to caller value");
+        assert_eq!(probe.normal, [0.0, 1.0, 0.0], "normalised +Y stays +Y");
+        let color_size = probe.color_rt.size();
+        assert_eq!(color_size.width, 256);
+        assert_eq!(color_size.height, 256);
+        assert_eq!(probe.color_rt.format(), super::super::formats::HDR_FORMAT);
+        assert_eq!(probe.depth_rt.format(), super::super::formats::DEPTH_FORMAT);
+    }
+
+    /// Tiny resolutions are clamped up to 16 px to keep wgpu happy.
+    #[test]
+    fn probe_resolution_clamps_minimum() {
+        let Some((device, _queue)) = try_create_device() else { return; };
+        let probe = PlanarReflectionProbe::new(&device, 0.0, [0.0, 1.0, 0.0], 4);
+        assert_eq!(probe.resolution, 16, "clamps to 16 px floor");
+    }
+
+    /// EN-011 V2 — oblique projection clips a point on the wrong side
+    /// of the supplied plane. Point ABOVE the eye-space plane should
+    /// project inside the [-w, w] z range (clip-space-z divides to
+    /// [0, 1] in wgpu); point BELOW the plane projects with z > w
+    /// (i.e. ndc.z > 1 in wgpu) so the rasterizer clips it.
+    ///
+    /// Setup: identity view (eye-space == world-space), a horizontal
+    /// plane y = 0 (so eye-space plane = (0, 1, 0, 0)), and a perspective
+    /// projection. We then check a point ABOVE (y = +5) renders, and a
+    /// point BELOW (y = -5) gets clipped.
+    #[test]
+    fn oblique_proj_clips_below_plane() {
+        use crate::renderer::util::mat4_perspective;
+        // Standard perspective matching mat4_perspective conventions.
+        let proj = mat4_perspective(70.0_f32.to_radians(), 16.0/9.0, 0.1, 100.0);
+        // Horizontal plane y = 0 in eye-space, normal +Y, points above
+        // satisfy y > 0 → N·p + d > 0 (kept). The plane equation
+        // (Nx, Ny, Nz, d) for "y >= 0 is above" is (0, 1, 0, 0).
+        let plane_eye = [0.0_f32, 1.0, 0.0, 0.0];
+        let oblique = oblique_proj(proj, plane_eye);
+
+        // A point in eye space ABOVE the plane, in front of camera:
+        // y = +5 (above), z = -10 (in front, right-handed view).
+        let p_above = [3.0_f32, 5.0, -10.0, 1.0];
+        let clip_above = mat4_mul_vec4(&oblique, &p_above);
+        // wgpu / D3D / Metal clip space: visible when 0 <= z <= w.
+        // For a point above the plane, z/w should be inside [0, 1].
+        assert!(clip_above[3] > 0.0, "point above plane has positive w (got {})", clip_above[3]);
+        let ndc_z_above = clip_above[2] / clip_above[3];
+        assert!(ndc_z_above >= 0.0 && ndc_z_above <= 1.0,
+            "above-plane point ndc.z within [0,1] (got {})", ndc_z_above);
+
+        // A point BELOW the plane: y = -5 (below), z = -10 (in front).
+        // Oblique projection moves the near plane to coincide with
+        // y = 0, so this point is on the wrong side and should clip.
+        // wgpu's clip space requires 0 <= z <= w to be visible — any
+        // ndc.z outside [0, 1] (or w <= 0) means the rasterizer drops
+        // the fragment.
+        let p_below = [3.0_f32, -5.0, -10.0, 1.0];
+        let clip_below = mat4_mul_vec4(&oblique, &p_below);
+        let visible = clip_below[3] > 0.0
+            && clip_below[2] >= 0.0
+            && clip_below[2] <= clip_below[3];
+        assert!(!visible,
+            "below-plane point should be clipped; got clip = ({}, {}, {}, {})",
+            clip_below[0], clip_below[1], clip_below[2], clip_below[3]);
+    }
+
+    /// `world_plane_to_eye_space` round-trip: a horizontal world plane
+    /// y = 0 (normal +Y) under an identity view stays as (0, 1, 0, 0)
+    /// in eye space — sanity check the inverse-transpose plumbing.
+    #[test]
+    fn world_plane_to_eye_space_identity_view() {
+        use crate::renderer::util::IDENTITY_MAT4;
+        let plane_world = [0.0_f32, 1.0, 0.0, 0.0]; // y = 0 plane
+        let plane_eye = world_plane_to_eye_space(IDENTITY_MAT4, plane_world);
+        for i in 0..4 {
+            assert!((plane_eye[i] - plane_world[i]).abs() < 1e-5,
+                "identity view leaves plane unchanged (comp {}: {} vs {})",
+                i, plane_eye[i], plane_world[i]);
+        }
+    }
+}