getWorldScale and getWorldQuaternion return incorrect values when object scale has three negative components

Describe the bug

mesh.getWorldScale and mesh.getWorldQuaternion return incorrect values for a mesh with negative x, y, and z scale components.

If mesh.scale is (-1, -1, -1), getWorldScale returns (-1, 1, 1), losing two minus signs. If mesh.quaternion is (0, 0, 0, 1), getWorldQuaternion returns (1, 0, 0, 0).

To Reproduce

Steps to reproduce the behavior:

1. Go to https://codepen.io/bennlich/pen/NWXyLKm
2. Click anywhere on the canvas
3. Notice how the image texture flips, and look at the console to see how the values of mesh.scale differ before and after.
4. Go to https://codepen.io/bennlich/pen/MWrQPWE to see the analogous bug with getWorldQuaternion

Expected behavior

In the live example above, I would not expect the mesh to flip. I would expect getWorldScale to return the same value as the mesh’s current local scale. Same with getWorldQuaternion.

Platform:

• Device: Desktop
• OS: Linux
• Browser: Chrome
• Three.js version: 0.139.2

[WestLangley]

Scaling by ( - 1, - 1, 1 ), for example, is equivalent to a rotation-only around the z-axis.

quaternion and scale are coupled in such cases and you can’t consider them individually.

[donmccurdy]

Related:

…it is possible to compute a choice of translation 𝑇, rotation 𝑅, and scale 𝑆 with the same combined result as [matrix] 𝑀. But there is no guarantee that these will be the same values originally used to create 𝑀, as we’ll see below…

https://gamedev.stackexchange.com/questions/117528/calculate-matrix-transformation-components-separately

Matrix decomposition (or higher-level methods like getWorldScale) cannot be guaranteed to return the same translation/rotation/scale (TRS) components as the originals. Repeated round-trip conversions between matrixes and TRS components should be avoided.