Inconsistent rotations for Camera3D

[rabipelais]

The panRad function pre-multiplies the new rotation to the old orientation, while tiltRad and rollRad post-multiply it. This leads, for example, to the following unexpected behaviour:

• roll 90 degrees to the right
• now tilting up and panning right give the exact same behaviour, which is rotating upwards (with respect to the rotated camera). This makes sense for tilting, not so much for panning.

Is this behaviour desired?

[acowley]

No, that doesn't sound good. I'll merge a PR.

[acowley]

I suspect I made things inconsistent because I had a camera control scenario where I wanted panning relative to the global up axis. I think making things consistent as you've done is the right thing, but I'm tempted to add an alternative pan that does things the old way. One argument for offering an option unique to panning is that it is rotation with respect to a commonly implicit axis induced by gravity. Do you think it would make sense to offer alternatives to other camera rotations to offer developer-friendly camera controls?

[rabipelais]

I had the use case where the program read gyroscopic data, where roll, pitch, and yaw were always in global coordinates. I changed tilt and roll to match pan, and wrote ...Local and ...LocalRad versions. Since local rotations were the common case for me in the rest of the program, I swapped them and had ...Global... versions. So I suppose it is not a very uncommon use case. They are, however, simple functions to write. I don't know what you prefer with regards to having a larger, but possibly more complete, API. (I also wrote a reorientate :: Quaternion a -> Camera a -> Camera a function. Didn't like the name, though.)

[acowley]

I guess we should have global versions, too.

I don't love the expansion, but this seems like a pretty common distinction. I agree they're simple functions, but part of the point of the linear package was just to provide common definitions of simple things. I guess I'd call reorientate reorient just by virtue of the latter being slightly shorter.

My vote is to add them all in.

[rabipelais]

That makes sense. Then I also think it is a good idea to add them. Quick question, would you expect that reorient substitutes the orientation of the camera completely, or multiplies it to the correct orientation, i.e. reorient q c = c { orientation = orientation c * q } (then again the question about local vs. global axis)?

[rabipelais]

(The PR only adds the ...Global and ...GlobalRad versions, not the reorient function. I think better names are due.)

[acowley]

Yeah, I was thinking about that multiply vs. substitute question. The ambiguity demonstrates the problem with the name. I'd expect reorient to replace the orientation completely, while something like rotateCamera to do the multiplication, but I don't know if my reading of those names is representative. The former could be something like resetOrientation which is perhaps more clear.