Nope, that actually seems wrong if the object is rotated. Looking into it further.

Nope, that actually seems wrong if the object is rotated. Looking into it further.

Seems like this happens with the original code as well: the health bar rotates with the object. I thought it would act more like a billboard.

I thought it would act more like a billboard.

yeah it's supposed to, good find

We may have to rethink the offset. With the changes in this PR, dual bars would overlay each other when viewed top down. In the multi-bar scenario, the bars' positions shouldn't change relative to each other even when the camera is rotated/tilted

Yes, I've noticed that too. If viewed directly top down, they will overlap. But if there is a small angle (which is typical even in something like an RTS), and the bars have different offsets (heights), then they are both seen fairly clearly:

what do you think about having the offset supplied as a vec3 instead of f32, where the components are applied as follows:
if the camera view vector is parallel to an axis (such as top-down for the y-axis), the corresponding component in the offset vector is applied 0% (not at all). If it's orthogonal, it is applied a 100% (i.e. front view for the y-axis). You could then specify 0 for the z component of the offset vector and in your case and in the top down view the bar would be centered.

(every angle in between orthogonal/parallel would be a percentage of the rotation, i.e. 45 degrees would be 50%)

It's hard for me to say without playing around with it :)