AI_Assistant/cad-editor
Watch
1
0
Fork
You've already forked cad-editor
 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

fix: correct CPU projection axes to match GPU look_at_rh convention

Browse source 
viewport_content_wires() computed view_right via world_z × view_direction
(cross-product). This gives the OPPOSITE sign to the right vector that
look_at_rh produces for views with a Y component (FRONT/BACK views):

  GPU look_at_rh:  screen_right = f × up = (−vd) × (rot*Y) = rot*X
  Old CPU formula: screen_right = Z × vd  ← opposite sign for FRONT/BACK

The mismatch caused the paper-space projection to appear mirror-flipped
after panning/rotating in MSPACE.

Fix: use camera_for_viewport() to obtain the exact same rotation
quaternion the GPU uses, then derive both axes from it:
  view_right = rotation * X
  view_up    = rotation * Y

Also replaces the perspective depth dot-product (mp · vd) with
mp · (rotation*Z) so the depth direction is consistent too.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
Hakan Seven 2026-04-23 09:28:54 +03:00
commit 1c0520e4e4
1 changed files with 14 additions and 18 deletions
Download patch file Download diff file Expand all files Collapse all files

32
src/scene/mod.rs
View file

@ -702,22 +702,19 @@ impl Scene {
// ── Per-viewport frozen layer set ───────────────────────────── // ── Per-viewport frozen layer set ─────────────────────────────
let frozen: HSet<Handle> = vp.frozen_layers.iter().cloned().collect(); let frozen: HSet<Handle> = vp.frozen_layers.iter().cloned().collect();
// ── View direction coordinate frame ─────────────────────────── // ── View coordinate frame ─────────────────────────────────────
let vd = glam::Vec3::new( // Use camera_for_viewport so the axes match the GPU renderer exactly.
vp.view_direction.x as f32, // GPU uses look_at_rh(eye, target, rotation*Y), which gives:
vp.view_direction.y as f32, // screen_right = normalize((target-eye) × up) = rotation * X
vp.view_direction.z as f32, // screen_up = rotation * Y
).normalize_or(glam::Vec3::Z); // The old Z×vd cross-product gives the opposite sign for views
// that have a Y component in view_direction (front/back), causing mirroring.
// Compute view_right and view_up from the direction vector. let cam_frame = match self.camera_for_viewport(vp.common.handle) {
let world_z = glam::Vec3::Z; Some(c) => c,
let view_right = if (vd.dot(world_z)).abs() > 0.99 { None => continue,
// Looking straight up/down: use X as right.
glam::Vec3::X
} else {
world_z.cross(vd).normalize()
}; };
let view_up = vd.cross(view_right).normalize(); let view_right = cam_frame.rotation * glam::Vec3::X;
let view_up = cam_frame.rotation * glam::Vec3::Y;
// ── Scale & viewport parameters ─────────────────────────────── // ── Scale & viewport parameters ───────────────────────────────
let scale = if vp.custom_scale.abs() > 1e-9 { let scale = if vp.custom_scale.abs() > 1e-9 {
@ -795,9 +792,8 @@ impl Scene {
let u = mp.dot(view_right); let u = mp.dot(view_right);
let v = mp.dot(view_up); let v = mp.dot(view_up);
if use_perspective { if use_perspective {
// vd points from target toward camera, so depth along vd is the // Eye direction (target → camera) = rotation * Z.
// distance from the target plane toward the camera. let d_vd = mp.dot(cam_frame.rotation * glam::Vec3::Z);
let d_vd = mp.dot(vd);
// Forward distance from camera (positive = in front of camera). // Forward distance from camera (positive = in front of camera).
let fwd = camera_dist - d_vd; let fwd = camera_dist - d_vd;
if fwd <= 0.001 { if fwd <= 0.001 {