Rationale for Ada 2005

John Barnes
Contents   Index   References   Search   Previous   Next 

1.3 Overview of changes

It would be tedious to give a section by section review of the changes as seen by the Reference Manual language lawyer. Instead, the changes will be presented by areas as seen by the user. There can be considered to be six areas:
Improvements to the OO model. These include a more traditional notation for invoking an operation of an object without needing to know precisely where the operation is declared (the Obj.Op(...) or prefixed style), Java-like multiple inheritance using the concept of interfaces, the introduction of null procedures as a category of operation rather like an abstract operation, and the ability to do type extension at a more nested level than that of the parent type. There are also explicit features for overcoming nasty bugs that arise from confusion between overloading and overriding.
More flexible access types. Ada 95 access types have a hair-shirt flavour compared with other languages because of the general need for explicit conversions with named access types. This is alleviated by permitting anonymous access types in more contexts. It is also possible to indicate whether an access type is an access to a constant and whether a null value is permitted. Anonymous access-to-subprogram types are also introduced thus permitting so-called downward closures.
Enhanced structure and visibility control. The most important change here is the introduction of limited with clauses which allow types in two packages to refer to each other (the mutual dependence problem referred to in the WG9 guidelines). This is done by extending the concept of incomplete types (and introducing tagged incomplete types). There are also private with clauses just providing access from a private part. And there are significant changes to limited types to make them more useful; these include initialization using limited aggregates and composition using a new form of return statement.
Tasking and real-time improvements. Almost all of the changes are in the Real-Time Systems annex. They include the introduction of the Ravenscar profile (as explicitly mentioned in the WG9 guidelines) and a number of new scheduling and dispatching policies. There are also new predefined packages for controlling execution time clocks and execution time budgets and for the notification of task termination and similar matters. A change related to the OO model is the introduction of protected and task interfaces thereby drawing the OO and tasking aspects of the language closer together.
Improvements to exceptions, numerics, generics etc. There are some minor improvements in the exception area, namely, neater ways of testing for null occurrence and raising an exception with a message. Two small but vital numeric changes are a Mod attribute to solve problems of mixing signed and unsigned integers and a fix to the fixed-fixed multiplication problem (which has kept some users locked into Ada 83). There are also a number of new pragmas such as: Unsuppress to complement the Suppress pragma, Assert which was already offered by most vendors, Preelaborable_Initialization which works with the existing pragma Preelaborate, No_Return which indicates that a procedure never returns normally, and Unchecked_Union to ease interfacing to unchecked unions in C. There is also the ability to have more control of partial parameters of generic formal packages to improve package composition.
Extensions to the standard library. New packages include a comprehensive Container library, mechanisms for directory operations and access to environment variables, further operations on times and dates, the vectors and matrices material from ISO/IEC 13813 (as directed in the WG9 guidelines) plus commonly required simple linear algebra algorithms. There are also wide-wide character types and operations for 32-bit characters, the ability to use more characters in identifiers, and improvements and extensions to the existing string packages. 
Of course, the areas mentioned above interact greatly and much of 2 and 3 could be classified as improvements to the OO model. There are also a number of changes not mentioned which will mostly be of interest to experts in various areas. These cover topics such as streams, object factory functions, subtle aspects of the overload resolution rules, and the categorization of packages with pragmas Pure and Preelaborate.
The reader might feel that the changes are quite extensive but each has an important role to play in making Ada more useful. Indeed many other changes were rejected as really unnecessary. These include old chestnuts such as in out and out parameters for functions (ugh), extensible enumeration types (a slippery slope), defaults for all generic parameters (would lead one astray), and user-defined operator symbols (a nightmare).
Before looking at the six areas in a little more detail it is perhaps worth saying a few words about compatibility with Ada 95. The guidelines gave the ARG freedom to be sensible in this area. Of course, the worst incompatibilities are those where a valid program in Ada 95 continues to be valid in Ada 2005 but does something different. It is believed that serious incompatibilities of this nature will never arise. There are however, a very few minor and benign such incompatibilities concerning the raising of exceptions such as that with access parameters discussed in Section 1.3.2.
However, incompatibilities whereby a valid Ada 95 program fails to compile in Ada 2005 are tolerable provided they are infrequent. A few such incompatibilities are possible. The most obvious cause is the introduction of three more reserved words: interface, overriding, and synchronized. Thus if an existing Ada 95 program uses any of these as an identifier then it will need modification. The introduction of a new category of unreserved keywords was considered for these so that incompatibilities would not arise. However, it was felt that this was ugly, confusing, and prone to introducing nasty errors. In any event the identifiers Overriding and Synchronized are likely to be rare and although Interface is clearly a likely identifier nevertheless to have it both as an identifier and as a keyword in the same program would be nasty. Note also that the pragma Interface which many compilers still support from Ada 83 (although not mentioned by Ada 95 at all) is being put into Annex J for obsolescent features.

Contents   Index   References   Search   Previous   Next 
© 2005, 2006, 2007 John Barnes Informatics.
Sponsored in part by:
The Ada Resource Association and its member companies: ARA Members AdaCore Polyspace Technologies Praxis Critical Systems IBM Rational Sofcheck and   Ada-Europe: