[AstroPy] Writing unsigned int with pyfits

[Erik Bray]

On 04/30/2012 09:45 AM, Martin Raue wrote:
> Dear all,
>
> I am trying to create a FITS file with a bintable extension containing unsigned integers with pyfits but could not get it to work (example below). Does maybe someone know what I am doing wrong here?
>
> Best wishes,
> Martin
>
> import numpy as np
> import pyfits as pf
>
> phdu = pf.PrimaryHDU(uint=True)
>
> a_uint = np.ones(1, dtype=np.uint32)
>
> events = pf.new_table(
>      pf.ColDefs([
>          pf.Column(name='EVENT_ID', array=a_uint, format='J', bscale=1, bzero=2**31)
>          pf.Column(name='OBS_ID', array=a_uint, format='J', bscale=1, bzero=2**31)
> #       ... a few more columns
>          ])
>      )
>
> hdulist = pf.HDUList([phdu, events])
>
> hdulist.writeto('tmp.fits')
>
> hdulist = None
>
> f = pf.open('tmp.fits')
>
> In [2]: f[1].data
> Out[2]:
> FITS_rec([(1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0)],
>        dtype=[('EVENT_ID', '>i4'), ('OBS_ID', '>i4'), ('TIME', '>f4'), ('RA', '>f4'), ('DEC', '>f4'), ('DETX', '>f4'), ('DETY', '>f4'), ('ENERGY', '>f4'), ('HIL_MSW', '>f4'), ('HIL_MSL', '>f4')])
>
> In [3]: f[1].data.field('EVENT_ID')[0].dtype
> Out[3]: dtype('float64')

I believe that you are doing this correctly.  However, when a column has 
non-trivial BSCALE and/or BZERO values, the values in that column are 
converted to floats on the fly.  That's because the TSCALn and TZEROn 
values themselves are actually treated as floats by the FITS standard.

So while the column format shows up as ints, the actual values given to 
the user are floats with the bscale+bzero applied.  If you save changes 
to the file the values will be converted back to ints.

I think the problem here is that PyFITS supports unsigned ints in image 
data by using the `uint=True` argument when opening a FITS file.  Using 
that argument causes PyFITS to recognize that you want pseudo-unsigned 
ints to be returned as actual unsigned ints, and not floats.  However, 
it seems like it only works for images, and not for table columns. 
That, perhaps, should be fixed.

Erik