The final International Standard for Ada 95 incorporates a number of changes to the Committee Draft [CD 93] of September 1993 (version 4.0 of RM9X) and the Draft International Standard [DIS 94] of June 1994 (version 5.0). These were made in response to formal comments made by the ISO members as part of the ballots on these drafts and to informal comments made by individual reviewers.
Although many of the changes are of an editorial nature several are of significance to the normal user. The more important changes are outlined in this appendix for the convenience of readers familiar with the drafts. Unless otherwise mentioned changes are with respect to the CD; if a change has occurred since the DIS then this is explicitly mentioned. A reference to that section of the rationale containing further discussion of the topic is given where appropriate.
The organization of the standard has been rearranged into a more logical order. The most significant to users familiar with Ada 83 is that chapter 14 on input-output has been moved into the annex on the predefined environment where it logically belongs. The annexes have themselves been reordered so that the mandatory annexes on the predefined environment and interfacing to other languages come first, followed by the six specialized needs annexes, the obsolescent features and then finally the non-normative annexes summarizing attributes and so on.
Trailing underlines are not allowed in identifiers whereas they were in the Committee Draft. Reversion to the Ada 83 rule was deemed appropriate because allowing just trailing underlines did not achieve the flexibility desired for wide compatibility with other languages such as C. Permitting leading underlines and multiple embedded underlines would have given greater compatibility but was considered unacceptable given the strength of concern for readability of program text. (2.1)
Modular types are no longer described in terms of principal values and secondary values; they just have a value. A consequence is that conversion to and from integer types always preserves the numerical value or raises Constraint_Error. Wraparound on conversion no longer occurs. (3.3.2)
The ancestor part can now be a subtype name as an alternative to an expression. This enables an extension aggregate to be written even when the ancestor is abstract such as in the case of controlled types. (3.6.1, 7.4)
The package Ada.Finalization is restructured. The types Controlled and Limited_Controlled are no longer derived from types in the package System.Implementation (which no longer exists) but are simply abstract private types. The previous problem with writing aggregates for types derived from abstract types is now overcome by the new form of extension aggregate mentioned above.
The procedure Split is now called Adjust. The procedures Adjust and Finalize are no longer abstract but have null bodies like Initialize. (7.4)
The new pragma Storage_Size permits setting the storage size for individual tasks of a task type. This pragma is placed in the task specification and could thus depend on the value of a task discriminant. It replaces the use of an attribute definition clause for setting Storage_Size which gave the same attribute value to all tasks of the type. (9.6)
It is no longer necessary for a child of a generic unit to be instantiated as a child of an instantiation of its parent. This requirement of the CD and DIS caused problems for many applications and a child can now be instantiated anywhere provided the generic child is visible. (10.1.3)
The package Ada.Exceptions is significantly restructured. The generic child Ada.Exceptions.Messages has been deleted. The ability to attach a user message to the raising of an exception can now be done more flexibly using the procedure Raise_Occurrence and the new attribute Identity. A major advantage is that exceptions so raised do not all have the same name Exception_With_Message.
The type Exception_Occurrence is now limited so that occurrences cannot be directly assigned. Exceptions can now be saved by a procedure and function Save_Occurrence. This approach overcomes implementation problems associated with the size of saved information. (11.2)
The Access attribute can now be applied to objects in generic bodies when the access type is external. The associated accessibility check is dynamic and raises Program_Error if it fails. This gives greater flexibility in the use of generics. Note that the Access attribute still cannot be applied to subprograms in generic bodies when the access type is external. (12.3)
The rules regarding these attributes have been somewhat changed. They can now only be applied to first subtypes (and objects) and not to all subtypes. Furthermore the Address must be a multiple of the Alignment. (13.1)
This has been slightly restructured. The classification and conversion functions are now in a child package Characters.Handling and the package Characters is itself empty (other than the pragma Pure). The reason for this change is so that the child Characters.Latin_1 can be used without introducing any unnecessary executable code from its parent. A related change is that the package Standard.ASCII is now obsolescent; programmers are expected to use Characters.Latin_1 instead. (A.1)
The pragmas Import and Export now have a fourth parameter. The third parameter now gives the name of the entity in the other language and the fourth parameter gives the link name. (B.1)
A number of improvements have been made to Text_IO.
The concept of an error output stream has been added in line with facilities in many operating systems. Subprograms enable the error stream to be manipulated in a manner similar to the default output stream.
The functions Current_Input, Current_Output and Current_Error are overloaded with versions returning an access value. This enables the current stream to be preserved for later use in a more flexible manner.
The procedure Get_Immediate provides immediate non-buffered and non- blocking input; this is useful for interactive applications.
The procedure Look_Ahead returns the next character without removing it; this enables the user to write procedures which behave in a similar manner to the predefined procedures Get for integer and real types.
The procedure Get for real types will now accept a literal in more liberal formats; leading and trailing digits around the radix point are not required and indeed the point itself may be omitted. This enables data produced by programs written in languages such as Fortran to be processed directly.
The procedure Flush is added; this outputs the contents of the current buffer.
Nongeneric packages equivalent to instantiations of Integer_IO and Float_IO with the predefined types have been added since the DIS. These will be found of considerable benefit for teaching Ada since simple input-output of numbers can now be performed without the introduction of genericity. (A.4)
The package Ada.Command_Line enables a program to access the commands and parameters of the command line interpreter if any. It also enables a program to set its result status. (A.5)
The package Ada.Numerics.Random_Numbers has been considerably restructured and renamed as Ada.Numerics.Float_Random. The additional generic package Ada.Numerics.Discrete_Random produces streams of random discrete values. (A.3)
The package Ada.Text_IO.Pictures is now called Ada.Text_IO.Editing. The description has been recast to avoid dependence on the COBOL standard. (F.2)