Introduction to the gfortran array descriptor
With the approval of Fortran 90, its array capabilities were largely improved. While still far from languages like APL, the extra functionality required a rethinking of the concept array in Fortran. This led to the need for array descriptors in the language.
Many programming languages have a concept of array, as a very essential kind of aggregated data type where all the elements of the aggregation are of the same type and can be accessed by some indexing mechanism usually an integer expression. Beyond that basic functionality, languages differ.
In pre-Fortran 90 arrays were pretty simple and their size could only be a compile-time constant or an expression of the parameters of the functions. This is, there were not dynamic arrays of any kind in Fortran 77. This changed in Fortran 90 with the introduction of 4 kinds of arrays:
As you can see both assumed-shape and deferred-shape raise the bar of complexity of arrays. The reason is that we have to be able to compute their sizes at runtime. Fortran 90 introduces LBOUND, UBOUND and SIZE intrinsics that allow us to query the boundaries and number of elements of an array, or one of its dimensions. For explicit-shape either these values are constant, so they can be computed by the compiler, or must depend on parameters so we still can relatively easy compute those in runtime without too much effort. But for assumed-shape and deferred-shape, this property becomes so general that we need to have the boundaries of the array somewhere.
