We'd welcome anybody making a pull-request with these changes! Ideally we'd have some unit tests as well ;).

Just a note, the original code makes use of mSinCos which at least on x86 uses the fsincos instruction which in theory is faster than calling mSin & mCos separately.

If I'm reading it right, translating it to stock would give:

QuatF& QuatF::set( const EulerF & e )
{
F32 cx, sx;
F32 cy, sy;
F32 cz, sz;
mSinCos( e.x * 0.5f, sx, cx ); //flip
mSinCos( e.y * 0.5f, sy, cy ); //flip
mSinCos( e.z * 0.5f, sz, cz ); //flip

// Qyaw(z) = [ (0, 0, sin z/2), cos z/2 ]
// Qpitch(x) = [ (sin x/2, 0, 0), cos x/2 ]
// Qroll(y) = [ (0, sin y/2, 0), cos y/2 ]
// this = Qresult = Qyaw_Qpitch_Qroll ZXY
//
// The code that folows is a simplification of:
// roll_=pitch;
// roll_=yaw;
// _this = roll;
F32 cycz, sysz, sycz, cysz;
cycz = cy_cz;
sysz = sy_sz;
sycz = sy_cz;
cysz = cy_sz;
w = cycz_cx - sysz_sx; //flip
x = cycz_sx + sysz_cx;
y = sycz_cx + cysz_sx; //flip
z = cysz_cx - sycz*sx;

return *this;
}

Bit beyond my personal comfort level though for throwing that out as a PR, since it's not math I'm heavily familiar with.

I'm gonna work on this, and add some unit tests in my gtest-tests branch.

Okay, I'm writing unit tests for this using the following approach. I create an EulerF(x, 0, 0) and an AngAxisF(Point3F(1, 0, 0), x), which should represent identical rotations, right? Then I set two QuatFs from each representation. Then I check angleBetween the two quaternions.

Currently it works if I expect the angle to be less than 1e-3. Is that too large a tolerance? I mean, that's 1/1000th of a radian...

This fix is apparently good. Will need to fix existing logic that relies on this conversion (looking at you, TurretShape).