inconsistent bytes/unicode data for string table data in python3 about pyfits

[embray]

The attribute-based access is not intended to be used with PyFITS tables. It's a misfeature of Numpy.

[juliantaylor]

maybe, but are you sure its no bug that you get a completely different type back depending on which you use?
regular numpy recarrays do not exhibit this behavior.

[embray]

Yes--this is because the string columns are read in natively as byte arrays (essentially their native storage format in FITS files) but calling FITS_rec.field or FITS_rec.__setitem__ returns a converted array that contains unicode strings (for the sake of always presenting the user with strings instead of bytes). This is the same mechanism by which other data types are converted, such as 'T'/'F' to True/False, and how BSCALE/BZERO are handled. Going through the attribute access subverts returning the converted arrays.