As well as normal binary and ASCII FITS tables, STIL supports
FITS files which contain tabular data stored in column-oriented format.
This means that the table is stored in a BINTABLE extension HDU,
but that BINTABLE has a single row, with each cell of that row
holding a whole column's worth of data. The final (slowest-varying)
dimension of each of these cells (declared via the TDIMn headers)
is the same for every column, namely,
the number of rows in the table that is represented.
The point of this is that all the cells for each column are stored
contiguously, which for very large, and especially very wide tables
means that certain access patterns (basically, ones which access
only a small proportion of the columns in a table) can be much more
efficient since they require less I/O overhead in reading data blocks.
Such tables are perfectly legal FITS files, but general-purpose FITS software may not recognise them as multi-row tables in the usual way. This format is mostly intended for the case where you have a large table in some other format (possibly the result of an SQL query) and you wish to cache it in a way which can be read efficiently by a STIL-based application.
For performance reasons, it is advisable to access colfits files uncompressed on disk. Reading them from a remote URL, or in gzipped form, may be rather slow (in earlier versions it was not supported at all).
This format can be automatically identified by its content so you do not need to specify the format explicitly when reading colfits-basic tables, regardless of the filename.
Like the normal (row-oriented) FITS handler,
two variants are supported:
with (colfits-plus) or without (colfits-basic)
metadata stored as a VOTable byte array in the primary HDU.
For details of the FITS-plus convention, see Section 4.1.3.1.