Skip to content

Tables with 2D columns of strings fail round trip to FITS format under py3 #5287

Closed

Description

@sbailey

Example using astropy 1.2.1 and python 3.5:

In [1]: #- Tables with 1D columns of strings can round-trip to FITS just fine, including
   ...: #- the encoding/decoding to get form unicode to bytes and back
   ...: t1 = Table()
   ...: t1['x'] = ['aa', 'bb', 'cc']
   ...: t1.write('t1.fits', overwrite=True)
   ...: t2 = Table.read('t1.fits')
   ...: t1.dtype, t2.dtype
   ...: 
Out[1]: (dtype([('x', '<U2')]), dtype([('x', '<U2')]))

In [2]: t1['x'] == t2['x']
   ...: 
Out[2]: array([ True,  True,  True], dtype=bool)

In [3]: #- Add a 2D column of strings and try again:
   ...: t2['y'] = [['x', 'xx'], ['y', 'yy'], ['z', 'zz']]
   ...: t2.write('t2.fits', overwrite=True)
   ...: t3 = Table.read('t2.fits')
   ...: t2.dtype, t3.dtype
   ...: 
Out[3]: 
(dtype([('x', '<U2'), ('y', '<U2', (2,))]),
 dtype([('x', '<U2'), ('y', '<U8', (2,))]))

Note that the dtype for 'y' column has changed from '<U2' to '<U8'.

This works for python 2.7, where the dtype remains 'S2'.

Note that this problem is covered by the unit tests reported in #5280, but the discussion there has focused on the underlying io.fits library and how to create a BinTableHDU, so I'm separating this out as a separate Table-focused issue since that isolates the user from the underlying I/O library.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions