GILDAS has two slightly different data types : Images and Tables. Images are data sets of up to 4 dimensions, with a header specifying the World coordinate system, the type of projection used, spectroscopic information, etc... Tables are essentially like 2-D images (a 2-D image can be treated as a Table in fact), but the only relevant information in this case is the number of lines and columns in the table. The header is described in more details in the programmer's guide.
GILDAS was originally based on the mapping memory concept, in which an image, resident on the disk, is considered as part of the virtual memory space of the user. Memory mapping has the great advantage of separating the Algorithms from the Input/Output system. The portable Unix version, though no longer using memory mapping, still preserves a complete separation between I/O statements and Algorithms. Algorithms are just standard subroutines operating on arrays storing the images.