In this paper, Hodgson and Gardner describe the error mechanisms incorporated into the Ada NAG library. These mechanisms are flexible, yet very simple to use when default conditions are employed. The NAG approach is based on data structures that record error diagnostic details, and procedures to perform recovery operations when errors are detected.

Errors are classified according to severity and recorded with enough detail to allow them to be diagnosed accurately. The NAG library designers have been thorough, taking into account such special Ada issues as concurrent tasks and exceptions. A user may provide application-specific error-handling procedures or adopt the conventions of the default procedures included in the library.

The authors have written this paper to provide an example of one workable method of handling errors in large libraries. This is an important topic since Ada programming encourages the use of reusable libraries, and there are many possible error-handling strategies. The authors have succeeded in making their approach clear, and have given enough detail to allow the reader to design a similar system.

This paper is interesting because it describes some carefully designed error-handling conventions. However, there are a few points that most Ada users will not agree with. The authors refer to some of the characteristics of Ada exceptions as “shortcomings” without providing any rationale for this judgment. Many of these “shortcomings” are in fact the best possible choices given the Ada requirement of real-time efficiency. Some of the NAG error description types appear to have components of arbitrary length, which will cause a few Ada compilers to allocate excessive amounts of memory. Finally, and most seriously, the use of generics with error-handling procedures supplied as generic parameters will result in unbearable recompilation costs if the user wants to vary the application error-handling procedures frequently. This appears to be the Achilles’ heel of the NAG approach.

This paper will be of interest to those designing Ada libraries. The reader must be quite familiar with Ada to understand the concise descriptions and somewhat complex examples easily. The paper is well written, logically organized, and worthwhile reading for this audience.