[fitswcs] CAR projections

[David Berry]

I've come across several images which use a CAR projection to represent
the galactic plane as in the following example:

NAXIS1  =                 2961  
NAXIS2  =                  561  
CTYPE1  = 'GLON-CAR'            
CRVAL1  =         1.850000e+02  
CDELT1  =        -1.250000e-01  
CRPIX1  =                  1.0  
CTYPE2  = 'GLAT-CAR'            
CRVAL2  =        -3.500000e+01  
CDELT2  =         1.250000e-01  
CRPIX2  =                  1.0  
      
The owners of these images say that the intention of these headers is to
place the galactic centre at the centre of the image, with  GLON and GLAT
parallel to X and Y. This is on the assumption that the correct
interpretation of the above headers is:

      GLON = CDELT1*(X-CRPIX1)+CRVAL1
      GLAT = CDELT2*(Y-CRPIX2)+CRVAL2

Thus, they expect the WCS grid to be a simple rectangular cartesian
system. This is what they get with some image display tools, but
not with others (mine included) which produce a non-linear swirly
pattern of grid lines which definitely does not put the galactic centre 
at the centre of the image.

Now, are these headers just plain wrong, or is this an accepted 
interpretation of the CAR projection? My understanding is that in a CAR
projection, the axes in the projection plane are linearly related to the 
axes of the *native* spherical coordinate system, not the *celestial*
spherical coord system. So, unless the two systems are identical (i.e. 
CRVAL1 and CRVAL2 are both zero - the reference point for a cylindrical
projection is at the native origin), the mapping from X/Y to GLON/GLAT
will involve a non-linear component caused by the spherical rotation from
native to celestial coordinates.

So are the creators of these headers in error for not placing the
reference point at the galactic centre, or am I in error for
mis-interpreting the headers?

David