Showing what is selected

SelectionCage draws eight corner brackets round an object's extent, in the object's own axes, standing off it by a width in render pixels. It is handed a BoundingBox, a world matrix, a camera and a colour, and it writes twenty-four segments into a GizmoDraw — the same line list the grid, the entity markers and every contributed gizmo go into.

SceneLines calls it once per selected entity, and that is the whole of the wiring: whatever SceneLines emits appears in both of the editor's viewport paths.

A viewport has to say which object you just clicked, and colour cannot always say it.

The editor draws a pane one of two ways. ScenePresenter shades the editor's own block-out shapes with the tool renderer, and there a selected surface is simply tinted amber — SceneMeshes.SelectedColour. FramePresenter draws the pane through a GraphicsCompositor instead: ForwardPlus over the same ECS-extracted objects a game would draw, with the editor's tools recorded over the top in one Tools pass.

Nothing the tool renderer does to a surface exists in that second pane. The amber tint is not there, and neither was the inverted-hull rim that preceded it — that rim lived in the editor's own instanced mesh shader, which a compositor frame never binds. So a composed pane drew a selected object pixel-for-pixel like an unselected one, while the transform gizmo sat on it insisting it was selected. Measured: with the gizmo switched off, selecting a crate changed eleven pixels out of half a million.

In that pane an overlay is not a compromise, it is the only correct answer. A composed pane exists to show the frame a game would draw. Painting the selected object amber is precisely the edit that destroys what the pane is for — the material you selected the object to look at is the thing you can no longer see. A tint is right in a diagram; a picture wants its annotation drawn over it.

You do not want SelectionCage for saying an object has an extent. SceneShow.Bounds is that, and the two are deliberately different drawings — see below.

It takes geometry, not a scene, so a contributed gizmo can use it for whatever it considers selected.

Brackets rather than a box, because a box already exists. SceneShow.Bounds draws twelve continuous edges round every shaped entity as a diagnostic about extent. A selection drawn as a second wire box, in the same place, is two drawings a person cannot tell apart — and rendering them together proves it: they read as one thick doubled box. SelectionCage.Corner is a quarter, so each bracket takes a quarter of its edge and the middle half of every edge is empty. A cage is exactly 2 × Corner of a wire box's total length, and it is a wire box at Corner = 0.5.

The bounds box no longer turns amber. It used to, from a build in which nothing else round a selected object was. Now the two questions get one answer each: the box says what extent an object has, in the neutral colour extent is drawn in; the cage says which object is selected.

The standoff is in pixels, and it is divided by the object's scale on the way in. A cage exactly on the extent z-fights the silhouette it is annotating, and a gap in world units is invisible on a building and swallows a bolt. SelectionCage.Standoff is four render pixels, resolved through EditorCamera.WorldPerPixel at the object — the inverted hull's own rule, kept. Because the gap is a world distance and the extent is in the object's own space, each axis divides by that axis's scale: without it a crate scaled fourfold on X carries four times the gap on X, and a bracket further from one face than from the next reads as a bug in the bracket rather than as a scale on the object.

It is depth-tested, and it is not behind a show flag. The near corners show and the far ones are hidden by the object's own surface, which is what makes it read as a cage in the scene rather than a decal on the glass. And every other thing SceneLines draws asks SceneViewport.Show first; this does not, because a flag names a class of thing the scene has whether or not anybody asked to see it, and whether the click just made landed is not something a viewport may be configured to stop reporting.

An entity with no geometry gets no cage. A light, a camera and an empty have no extent, and a box round one would be a box round an arbitrary constant. They are already answered: SceneLines draws their marker cross in the selection's colour and at 1.6 times the size, and that reaches a composed pane the same way this does.

An entity's extent comes from whichever of three places it has geometry, and the order matches what the viewport actually draws: an EditMesh being edited wins over a mesh asset, and a mesh asset wins over a PrimitiveShape. The asset's size is resolved through SceneViewport.Meshes, the same IMeshSource the scene's mesh components already read:

That is on the viewport rather than on SceneLines for a reason worth repeating, because it is the same reason this page exists. Two presenters draw a pane and each holds a SceneLines of its own, so anything hung on one of those appears in one pane's worth of the editor and silently not in the other.

A mesh that has not loaded yet is not caged at a fallback size. IMeshSource is ask-don't-wait, so the miss is the ask and the cage appears on the frame the geometry does — which is the frame the object appears on. A guessed extent would draw a box of the wrong size round nothing at all for as long as the disk took, and then jump.

• The frame panel — what a GraphicsCompositor document is, and what a pane driven by one draws.
• Editor modes — what else the viewport is drawing while a mode is active.
• Editor/Vixen.Editor.SceneView/README.md — the viewport's own notes, including the inverted-hull outline this replaced in the composed pane and did not replace in the tool pane.