Nonformal introduction to ADA-95

From EDM2
Jump to: navigation, search

By Leonid Dulman

Since ADA-95 may become the base programming language in 21 century, the question "How to teach ADA-95?" has top-priority.

What principles must we follow teaching that programming language?

  1. The language must be studied from "tabula rasa", not referring to other programming languages (including ADA-83);
  2. We have to avoid referring to the notions that will be introduced later;
  3. We have to use natural examples and clear analogies.

ADA-95 is a strong and logically complete language. We need to understand all its base constructions. Thus, its study "online" is impossible. We must come back to the earlier introduced notions permanently expanding their sphere.

We study the language by a spiral, the new language constructions are continuations of the earlier introduced ones.

The main question for a student (in a study process) is how to solve his problems by means of the language implementations?

For a teacher it is very important to show what problems can be solved by means of certain language constructions.

You must remember that some problems can not be solved by means of ADA-95 clauses.There are procedures with variable number of parameters, problems that have a single realization (such as explicit transfer of control), and different realization problems (such as leaving a loop by statements GOTO or EXIT).

The top important goals of training are to connect the aims of a teacher and a student, and to get accurate answers for all questions. This work is an attempt to give common ADA-95 descriptions using the above principles.

Let us introduce the term "PROGRAM" as some entire language construction which transforms the several input data to require output results.

                         ADA-95  program
                      ┌───────────────────┐
                     ^│     work data     │.
                    . │   --------------- │ .
                   . ^│      process 1    │  .
                  . . └───────────────────┘   .
┌───────────────┐   . ^                        .
│  common data  │ .  .┌───────────────────┐     .  ┌───────────────┐
│protected data │....>│     work data     │      . │   results     │
└───────────────┘ .. .│   --------------- │------->│   (output     │
┌───────────────┐ ... │      process 2    │      . │    data    )  │
│     input     │.. . └───────────────────┘     .  └───────────────┘
│     data      │.   .                         .
└───────────────┘ .   V                       .
                   .  ┌───────────────────┐  .
                    . │    work data      │ .
                     V│   --------------- │.
                      │      process n    │
                      └───────────────────┘

Input data are transformed into results by means of one or more parallel execution processes.

Thus we manipulate objects (data) that must be transformed and processes which execute these transformations.

The objects used for the different means have different qualitative characteristics. The qualitative characteristic of the objects are named TYPE.

The object's type answers the question "What is it?"

What types of objects would we like to have in the language?

  1. Types for engineering and business calculations such as temperature or money (FLOAT and FIXED types);
  2. Types of objects for numbering other objects, for instance quantity of books. There are INTEGER and MOD types;
  3. The types of objects for working with words are CHARACTER and WIDE_CHARACTER;
  4. The types of TRUE/FALSE objects are BOOLEAN.

The types described above are named PREDEFINED in the language. They provide a base for new types or specifications named SUBTYPE.

In ADA-95 the type is subtype for itself.

But if you a composer and want to write the notes, what do you do?

The language creates objects such as notes or colors by means of enumeration types.

TYPE note is (C,D,E,F,G,A,H);

The common types INTEGER, MOD, BOOLEAN, CHARACTER, WIDE_CHARACTER, are enumeration types.

For example, TYPE BOOLEAN is (FALSE,TRUE).

A objects can be created by means of the above types:

  VELOCITY : FLOAT;
  MONEY : FIXED;
  TYPE BYTE IS MOD 16#100#; -- one byte.
  NUMBER : BYTE;

We have the magic word "NEW" to create a new type having the same "qualities" and with the same internal structure as an already defined type.

  TYPE MEAT IS NEW FLOAT;
  TYPE BEEF IS NEW MEAT;

But how can we describe a group of homogeneous elements, for instance text line?

There is an array construction in the language for this purpose.

TYPE STRING IS ARRAY(POSITIVE RANGE <>) OF CHARACTER;
TYPE WIDE_STRING IS ARRAY(POSITIVE RANGE <>) OF WIDE_CHARACTER;

These types are predefined and represent unbounded arrays. A variable which determines element location is named index.

An index has a first and a last value. Its range is defined using the first and last values as range 3 .. 13 for instance.

TYPE TABL IS ARRAY(1..10, -1..11) OF FLOAT; is an example of a two dimensional array of float numbers with bounds in the first direction of (1, 10) and in the second direction of (-1,11).

The index type is commonly INTEGER or MOD but we have the question "can we have an array with week-days as the index?". For example, the array might store the usual working time for each day.

The index may be of any enumeration type in ADA-95.

TYPE WEEK IS (SN,MN,TU,WE,TH,FR,ST);
TYPE WORK_TIME IS ARRAY(WEEK) OF INTEGER;
TM : WORK_TIME := (0,8,8,8,8,6,0);

Then TM(ST) for instance is equal to 0.

Let us assume we want to collect information about a student in one object.

We need name and growth. How is this object described? The compound type RECORD may be used somewhat analogous to STRUCTURE in other languages.

The record includes of one or more element types.

       TYPE STUDENT IS RECORD
               NAME   : WIDE_STRING(1..24);
               GROWTH : FLOAT;
       END RECORD;

Notice that to separate the record from other statements we write "END RECORD" after the record components.

Let us remark that a type contains is the quality characteristics of an object and it answer the question "what is it ?". But often we have to answer the question "where is it?". How can we refer to the object? This may be done directly by name such as

LENGTH, WIDTH : INTEGER;   and then
WIDTH := 3;
LENGTH := width * 5;

Alternatively, we often desire to refer to an object via its storage location or address. The POINTER terminology commonly used in other languages is replaced in ADA by the term "ACCESS TYPE" for indirect location of or reference to objects.

This access type object is the address for the object connected with it.

TYPE POINTER IS ACCESS STUDENT;

Remark the important properties of the access types:

  1. An access type points to objects connected with it;
  2. The access type object can create recursive data (a list or tree for instance). It can be a component of a record (node);
  3. If a type description contains the key word ALIASED, all objects of this type may be called by reference.

A simple example of an object and its pointer is follow:

TYPE OBJECT_TYPE IS .........
TYPE POINT_TO_OBJECT_TYPE IS ACCESS ALL OBJECT_TYPE;
OBJECT          : ALIASED OBJECT_TYPE;
POINT_TO_OBJECT : POINT_TO_OBJECT_TYPE := OBJECT'ACCESS;

Now we shall describe a transforming "process".

The process is a set of the language statements which may be classified into groups:

  1. Assignment_statement is ":=";
  2. Control statements are LOOP, IF, CASE, GOTO, SELECT;
  3. Transform statements are subprogram calls.
  4. The statement of an emergency situation is EXCEPTION.

There are two forms of subprogram: procedures and functions. A procedure call is a statement; a function call is an expression and returns a value.

  • Parameters are the objects that take part in transformations.
  • Procedures and functions may be used in recursive processes.

Inside the procedure or the function we can use others procedures and functions.

Notice that subprogram parameters may be only objects and they must have strongly defined types.

If a function has the parameters and the result of the same type, it is named a primitive operation. The primitive operations are introduced to finish the recursive process. They are predefined for the predefined types.

For example functions "+","-","*","/","**", MOD, REM are predefined for the INTEGER type.

The notation I := "+"(J,K) means that the integer variable i is a sum of the integers j and k.

It guarantees the checking of parameter types for the given subprogram. We can identify subprogram not only by the name, but by the types or the numbers of parameters. This property is named operation compatibility.

Let us have

       I   : INTEGER;
       S,R : CHARACTER; ,

then before the function R:="+"(S,I) can be used it must be defined since the normal "+" function expects its two parameters to be of the same type.

The function "+" with two parameters must be defined. The first is the character and the second is the integer. And it returns the character type. This function has nothing common with the function "+" in statement I:="+"(J,K) where J and K are of the same type and where "+" is predefined;

Two questions arise regarding the subprogram conception:

  1. Can we call or access a subprogram by way of its object address?;
  2. What can we do if a subprogram is the parameter of a subprogram?

For example, we want to calculate an integral of various functions.

This problem may be solved by means of using access type objects as parameters for subprograms.

  TYPE FUN_PTR IS ACCESS FUNCTION(X:FLOAT) RETURN FLOAT;
  FUN, FUNSIN, FUNCOS : FUN_PTR;
          @@...........
  FUNSIN := SIN'ACCESS; FUNCOS := COS'ACCESS; and so on.

But if FUN is an object, it may be used as a subprogram parameter for the integral calculation.

Thus a subprogram has parameters, but have can we generalize this very flexible mechanism to types? Can a type depends on parameters?

An array index is the analogy of parameter, but what do we do for an compound objects made up of several types? We use RECORDS.

If a parameter controls the internal numeric characteristics of a record, it's called a discriminant.

For example

            TYPE VARYING(MAXIMUM_LENGTH:INTEGER) IS RECORD
            LEN : INTEGER RANGE 1..MAXIMUM_LENGTH;
            CNT : STRING(1..MAXIMUM_LENGTH);
            END RECORD;

This record description allows creation of varying length strings depending on the length given at string object creation.

   S1 : VARYING(100); S2:VARYING(255);

But the parameters can also be used for the record structure control. The parameter is called a variant parameter.

For example,

 TYPE MOTOR IS (CAR,LORRY);
 TYPE AUTO(MASH : MOTOR) IS RECORD
  SPEED : FLOAT;
  CASE MASH IS
   WHEN CAR    => MEN : INTEGER;
   WHEN LORRY  => WEIGHT : FLOAT;
  END CASE;
  END RECORD;

Thus we may create either a car with a specific number of MEN (passengers) or a lorry with a specific WEIGHT. It is a very comfortable mechanism. You can create objects that have many common properties, but with some difference in their structure.

However it is a static mechanism and you have to examine the all possible variants. But if you design a large program, it's very difficult to see all possible requirements at once and these types may be very cumbersome.

The language allows creation of the new types dynamically. These types inherit the structure of an earlier declared type and add new necessary record components.

They are called TAGGED records.

       TYPE FIGURE IS TAGGED RECORD
               NAME   : STRING(1..8);
               SQUARE : FLOAT;
       END RECORD;

Now we can create additional new types from this parent type as needed by adding the necessary components. Let us note that all derived types keep the properties of the tagged record. These new types are able to be the prototype for other tagged types allowing a "tree" to be created. This tree joins tagged types into the single term "CLASS".

Since the class is introduced dynamically, its subprograms must be called in dynamically too. This property of the tagged types is named dynamic dispatch. In this case the type controls the choice of the subprograms for operation compatibility.

      TYPE TRIANGLE IS NEW FIGURE WITH RECORD
           L1,L2,L3:FLOAT;
           END RECORD;
      TYPE RECTANGLE IS NEW TRIANGLE WITH RECORD
           L4      :FLOAT;
           END RECORD;

The record unites the different object characteristics in one structure utilizing the full logic characteristic including all the parents. But how can we unite the different subprograms in one logic structure?

For example, assume we need a group of I/O subprograms?

The term PACKAGE is introduced into the language to create the logic unit. It joins objects, types and subprograms.

The packages have specifications in which types, objects, subprograms and exceptions are described. If you change the package specification you must recompile all program units which utilize, with, or reference the package.

If the package specification has subprogram(s), the package has a BODY with the code commonly in a different file from the SPECIFICATION. The body includes Ada source code for the execution of the subprogram(s). If the package body is changed and is in a separate file, only the BODY file needs to be recompiled. This provides large development time savings for large programs with many separate files.

The package type-specifications may be declared as PRIVATE. It's very important that all transformations of objects of private data types be done by means of the package subprograms only. If a type is designated as LIMITED PRIVATE we can't compare two objects of the type except in the package-defined procedures.

       PACKAGE PASSWORD IS
               TYPE KEY IS LIMITED PRIVATE;
               FUNCTION CANCEL(KL:KEY) RETURN BOOLEAN;
               PRIVATE
                       TYPE KEY IS STRING(1..8);
       END PASSWORD;

Packages have may contain the logical full system, but there are two difficulties:

  1. We must recompile the entire system If we want to make even a minor modification;
  2. A package may be very large but you may need only a small subset.

It's the package SSP,for example.

How are these problems solved? We added new record components for the tagged types to create the related new child types. By analogy we have the parents-children mechanism for the packages. The types and the primitive operations are introduced in root level, and new subprograms are described in the children packages.

PACKAGE PASSWORD.CHANGE IS
PROCEDURE CHG(KL:KEY);
END PASSWORD.CHANGE;

This mechanism allows creation of packages with smaller size and more logical complication. Now we have solved almost the all problems of one process program design. But we must mention the one "little" problem.

If we use many types for input/output data, we need many packages (subprograms). But all types derived from one predefined type, use equivalent operations. For instance, in the FIGURE type above, we may need a display function for each parent and child, SQUARE, TRIANGLE, and RECTANGLE.

In ADA-95 we have the subprograms with parameters, the records with parameters, and it's naturally to ask the question "can a package have parameters?"

The language has the type parametric apparatus. They are the GENERIC procedures and packages.

For example

GENERIC TYPE REAL IS DIGITS <>;
PACKAGE IO IS
    PROCEDURE READ (X : OUT REAL);
    PROCEDURE WRITE(X : REAL);
END IO;

When the IO specification and body of the package are compiled we write

 WITH IO;
  ...............................
 PACKAGE MEAT_IO IS NEW IO(MEAT);
 PACKAGE VEAL_IO IS NEW IO(VEAL);
 Z: MEAT;   Y: VEAL;

Now we have the i/o procedures for the MEAT meat.read(z), meat.write(z) and the procedures veal.read(y), veal.write(y) for the veal. It's needed to remark that the bodies of generic packages and subprograms would "with" TEXT_IO and we have to use the primitive operations therein.

But what to do if a system is been designing? We approximately know what to do with data but don't know the data internal structure and formation. We want to connect the process of planning and design with the process of programming.

This method is named object-oriented programming (OOP) and is realized in ADA-95 by means of packages with ABSTRACT data types. The peculiarity of these packages is the absence of body subprograms because the realization does not yet permit body definition during this stage. We want to introduce complex numbers but have not yet decided on their presentation (Cartesian, polar or mix) at this stage of the system design.

       PACKAGE ROOT IS
               TYPE DATA IS ABSTRACT TAGGED NULL RECORD;
               FUNCTION "+"(X,Y:DATA) RETURN DATA IS ABSTRACT;
               FUNCTION "-"(X,Y:DATA) RETURN DATA IS ABSTRACT;
               FUNCTION "*"(X,Y:DATA) RETURN DATA IS ABSTRACT;
               FUNCTION "/"(X,Y:DATA) RETURN DATA IS ABSTRACT;
       END ROOT;

This package has introduced abstract type DATA and declared the primitive operations. Now we can create child package root.complex and declare type complex_type.

PACKAGE ROOT.COMPLEX_CARTESIAN IS
        TYPE COMPLEX_TYPE IS NEW DATA WITH PRIVATE;
        FUNCTION "+"(X,Y:COMPLEX_TYPE) RETURN COMPLEX_TYPE;
        FUNCTION "-"(X,Y:COMPLEX_TYPE) RETURN COMPLEX_TYPE;
        FUNCTION "*"(X,Y:COMPLEX_TYPE) RETURN COMPLEX_TYPE;
        FUNCTION "/"(X,Y:COMPLEX_TYPE) RETURN COMPLEX_TYPE;

PRIVATE
        TYPE COMPLEX_TYPE IS NEW DATA WITH RECORD
        RE :FLOAT;
        IM :FLOAT;
        END RECORD;

END ROOT.COMPLEX_CARTESIAN;

We can write the primitive operations "+","-","*","/" and describe the package body therefore.

Let us finally remark that ADA-95 gives users the unrestricted capability to write complicated and safe programs.

Now we can study inter process communications. What variants of these communications we are interested in?

  1. There are fully independent processes. They start at the same time and don't interact. For example, there are postmen. They go to work in the morning, collect the correspondence and go away to deliver it.
  2. There are independent processes which may be aborted by time limit or special signals, for example, there is a blitz-chess.
  3. There are processes which synchronize their acts in fixed clock time. For example, the employees were invited to the conference at 11 o'clock. Everybody does his work before 11, they gather at the meeting place at 11 o'clock, and everybody does his work again after conference.
  4. There are processes which are time independent, but they work with common (shared) objects and depend on their accessibility or inaccessibility. For example, there is work with a data base in the correction mode.
  5. There are the processes with time and shared objects dependency. For example, the permission to turn on the fire extinguish system if the protection system has worked for 5 seconds.

TASKs are the processes called from the main process and executed in parallel with it. The task is an object of the "process" type. We use the term "TASK TYPE" for their declarations.

The task has two execution methods. These methods repeat consistently:

  1. There may be an independent asynchronous part;
  2. There may be a non-independent part requiring synchronization of time or data with others tasks.

If the synchronization part is needn't (post's example) the second part is absent. If the synchronization is required, the synchronously part is named ENTRY and marked in the group

 ACCEPT <entry_name>[(parameter list)] DO
  ..........
 END  [<entry_name>];

internal of the task body.

The language has an asymmetric task mechanism if ones tasks require rendezvous with others for data exchange and synchronization their actions.

Thus, the requiring-rendezvous task executes independently until the entry-calling point. The called on rendezvous task executes independently until its ACCEPT_statement. Whatever tasks has arrived first, must wait for the partner and only after rendezvous may it continue its independent execution.

If one task object with Entry point called from many others tasks then to ACCEPT_statement will be QUEUE and requiring-rendezvous tasks will be served in FIFO mode.

Thus the task type, similarly for the package description, requires the specification part to declare entry points rendezvous, and the body to contain source code to accomplish the needed action. As a rule, some device (disk or elevator) is connected with task, but its control is connected with entry.

       TASK TYPE READCARD IS
               ENTRY BUFFER(C:OUT CHARACTER) ;
       END READCARD;

Entry description is similar to the procedure ones. Let us remark, that tasks can be declared with discriminant (record analog), and from task types we can use both the task objects and the task arrays (family).

For example, READERS:ARRAY(1..8) OF READCARD;

If you want to connect device interruption with Entry point you may do it. For example task semaphore:

      TASK LIGHT is
        entry STOP ; -- button to stop traffic
        for STOP'address use 16#100# ;  -- address interruption
     end LIGHT ;

In this example describe semaphore controlled task LIGHT with entry STOP for stopping traffic. In main program or others tasks needn't to write LIGHT.STOP for calling entry because calling modeled by means of turning on button on semaphore and takes interruption at address 16#100#.

The task starts at once after the object declaration (after BEGIN).

But how can we control its start and finish ?

The task finish may be natural (end of task body), normal (execution TERMINATE_statement) and abnormal (execution ABORT_statement from other tasks).

The task start is a more complex problem. The direct start mechanism is absent but the indirect mechanism is provided via ACCESS types. The task type (similarly to any other) can be connected with the access type. The task object is dynamically created by the generator NEW.

The task execution is suspended until partners reach the rendezvous-point and waiting may be unlimited in a time.

How can we limit the rendezvous wait time or refuse the rendezvous with the task A and require the rendezvous with the task B?

The language has a SELECT_statement for this goal.

An example follows. The students have arrived at the lecture, but the professor is absent. How long they must wait for a rendezvous? Assume that they have decided to wait 15 seconds. Then we must write

SELECT
DELAY 15.0;
GOHOME;
THEN ABORT
Student.LECTURE;
END SELECT;

But students may arrive early at the lecture. More correctly we should write:

 SELECT
   DELAY UNTIL Time_encounter+15.0;
   GOHOME;
   THEN ABORT
   Student.LECTURE;
 END SELECT; -- where Time_encounter is the time set for the lecture.

We can guarantee safe work with shared data by means of the SELECT_mechanism, as in the calling task or in the called task, but the ADA-95 has more elegant solution. This is the PROTECTED types.

The protected type is a little "kingdom" in which is declared the objects, the procedures, the functions and the entries.

The shared data, used by different tasks, must be protected from use during changes.

All data manipulations can be done only by means of the protected functions for reading only, the protected procedures for monopolizing changing and the protected entry for the synchronizing. The boolean object-semaphore allows or prohibits shared data reformation in entry calls. It keeps waiting for the its TRUE value.

Consider a simple and frequent example, the system timer.

       WITH ADA.CALENDAR;
---------------------
       TYPE SYSTEM_TIME IS NEW ADA.CALENDAR.TIME;
       PROTECTED TYPE SYSTEM_CLOCK IS
               FUNCTION READ_TIME(TM:SYSTEM_TIME) RETURN STRING;
               ENTRY CAN_CHANGE(NT:STRING;TM:IN OUT SYSTEM_TIME);
               PRIVATE
               ENTRY FREE_TIME;
               FLAG:BOOLEAN:=FALSE;
               END SYSTEM_CLOCK;
       PROTECTED BODY SYSTEM_CLOCK IS
               FUNCTION READ_TIME(TM:SYSTEM_TIME) RETURN STRING IS
               BEGIN
                       @@..........................
               END READ_TIME;
       ENTRY CAN_CHANGE(NT:STRING;TM:IN OUT SYSTEM_TIME) WHEN NOT FLAG IS
               BEGIN  FLAG:=TRUE;
               TM:=.... ;-- implementation define;
               REQUEUE  FREE; --
       END CAN_CHANGE;
       ENTRY FREE WHEN TRUE IS
               BEGIN FLAG:=FALSE;
       END FREE;
       END SYSTEM_CLOCK;
       ST_CLOCK:SYSTEM_CLOCK; -- protected object
       TM:SYSTEM_TIME;
       We can read time from any tasks in any moment.
       TMM:=CLOCK.READ_TIME(TM);

But the actual system time change can be accomplished by the entry CAN_CHANGE and only by the condition FLAG=FALSE. It's the main condition of execution.

The statement REQUEUE for private entry FREE has done TMM access to change from others tasks. The requeue statement is designed to handle two main situations:

  • after an accept statement or entry body begins execution, it may be determined that the request cannot be satisfied immediately. Instead, there is a need to requeue the caller until the request can be handled.
  • alternatively, part of the request may be handled immediately, but there may be additional steps in the process that need to be performed at a later point.

I would be like to do two examples used REQUEUE statement.

First: You go to hospital first and you need dentist but present only surgeon and pediatrician.

  1. You may wait dentist and nobody in queue can go to doctors (perhaps they need surgeon or pediatrician). There is calling entry with barrier.
  2. You may go to tail of queue and others may go to doctors. There is REQUEUE statement.

Second: You are invited to boss but he is busy.

  1. You may wait boss free (entry with barrier)
  2. You may come back to work (REQUEUE statement with ABORT).

Or next example:

Protected type behaves as the king, nominating audience, and a tasks causing ENTRY point as the ambassadors, desiring to receive it. Because in audience time the king communicates only with one ambassador,there is the queue. And there where the present queue there is some priorities system. In elementary case when everyone are equal - it is FIFO (who earlier has come, that earlier have served) In more complex case in queue there is the moving according to priority. Some very proud ambassadors can not wish to stand in queue (REQUEUE with ABORT), and some "very petty" can simply deliver in tail (REQUEUE).

The task types and protected types may be declared only in subprograms or packages. They are the PRIVATE LIMITED types and can be used as parameters in the subprograms directly or by access.

It is impossible to give a full description of Ada-95 in this little paper, but I hope that now you can read ISO standard and understand it.