Progress and Issues

This wiki is meant to be read as a temporary reference as to my recent progress and the issues I'm facing. Thus, I'd like to keep it short and precise. A detailed description will be updated on the Blog https://aaryapatil.wordpress.com/ soon.

Progress and Issues:

1. I have spent a significant amount of time to understand the workflow of the io.fits sub-package. I thus have the following chain of control flow in mind:

astropy.io.connect The connect file basically connects all the I/O FITS operations that the user specifies in the io.fits namespace. Our focus will be on read_table_fits() and write_table_fits(), for the time being, since writing is our first priority and a comparatively easier task, write_table_fits() will be the main focus. From write_table_fits(), the control flows to table_to_hdu() from .convenience.

astropy.io.convenience The convenience file provides all the convenience functions to read and write FITS. We are particularly interested in the table_to_hdu. Here I am doing the following:

if table.has_mixin_columns:
    unsupported_cols = table.columns.not_isinstance((BaseColumn, Quantity, Time))
time_cols = table.columns.isinstance(Time)
    if time_cols:
        newtable = Table(table)
        table = get_time_table(newtable)

The get_time_table will replace the time columns by embedded columns (dictionary of columns) of dimensions (N,2). The issue being that it should be a TFORMn='2D' FITS Binary Table column. I am not sure if that happens explicitly once I replace the columns as a dictionary of columns.

Also, I haven't yet figured out how to define them as a dictionary of columns.

diction = dict(jd1 = Column([1,2,3]), jd2 = Column([12,4,5]))
diction = Column(dict(jd1 = [1,2,3], jd2 = [12,4,5]))

if I make a table column as table['col'] = diction using both strategies, it fails and provides each table element as a dictionary of columns which is broadcasted in the first case ::

<Column name='c' dtype='object' length=3>
{'jd1': <Column dtype='int64' length=3>
1
2
3, 'jd2': <Column dtype='int64' length=3>
12
 4
 5}
{'jd1': <Column dtype='int64' length=3>
1
2
3, 'jd2': <Column dtype='int64' length=3>
12
 4
 5}
{'jd1': <Column dtype='int64' length=3>
1
2
3, 'jd2': <Column dtype='int64' length=3>
12
 4
 5}

and a dictionary of arrays broadcasted in the second ::

<Column name='c' dtype='object' length=3>
{'jd1': [1, 2, 3], 'jd2': [12, 4, 5]}
{'jd1': [1, 2, 3], 'jd2': [12, 4, 5]}
{'jd1': [1, 2, 3], 'jd2': [12, 4, 5]}

Thus, the summary of Issues include:

1. How to define the dictionary of columns for jd1, jd2.
2. Would the high level (python level) embedded Table columns be enough to have an embedded Binary Table column in FITS
3. Where to place the logic, as a separate module or as a function in the Time Class.

I have read the FITS primer to get a quick glance of FITS, read about the various keywords supported, the Time documentation and metadata, the SOFA definitions to get a grasp of the basic concepts and the FITS Time journal to start the mapping of keywords relevant to Time.