[fitsbits] TDIMn

[Eric Greisen]

I share much of Preben's uncertainties about adopting the appendices.

The TDIMn (B.2) is absolutely needed and is or soon will be used.  It
is also a convention that can be ignored by general readers that do
not need to know the internal details of a table column's data.  It
uses a previously non-standard keyword and so has been ignored by
general readers.

The substring convention (B.3) is also relatively harmless in that a
general routine may ignore all the substring notation.  However, that
notation appears as part of the data value of a standard keyword and
so may (should??) confuse a general reader that does not support the
extension.  Has this been used by anyone?

The heap convention however does worry me.  It changes the data
structure in the FITS file and I expect that general FITS readers will
not normally support the convention internally.  In fact, I have
always regarded this convention with the worry that it encourages a
lack of discipline in designing the data tables in the first place.
Thus an over-worked designer may decide to dump all sorts of data into
a single table piling most of the "table" data into an unstructured
heap, rather than building two or more cleanly structured tables
separating unlike data into separate columns/tables.

I do not doubt that the heap may be an excellent choice for an
internal format in a data system geared to handle it.  But as a
communication-of-data choice I expect that it will not be implemented
generally and hence stand in the way of communication.

Eric Greisen