The ADI Design Goals

The following summarises the ADI design discussed below.

• The number of ADI root routines (ie. ignoring those which differ only in the type of key arguments) should be kept as small as possible.

• The interface should allow advantage to be taken of particular gains in efficiency in accessing data in particular ways, eg. sequentially in FITS, or chunking of HDS arrays.

• The system should have as few static limits as possible (eg. on the number of datasets). By doing everything dynamically upper limits are removed, and the saving in memory in the far more usual case of operation near the lower limit is well worthwhile.

• The system should allow inheritance in the sense that if, for example, a new feature is added to the processing of HDS arrays, then that feature would be available to manipulate the data, variance and other array components of a binned dataset.