Producer Consumer Comparison 

The C++ source code for this example is taken from the blog post here by Andrew Wei. A detailed description of the C++ software is given in the blog post.

This solution is shown in parallel with a corresponding Ada solution.

Both C++ and Ada separate interface specifications from implementation. C++ uses the header file, in this case the file named Buffer.hpp, to provide the interface specification for the buffer used in this example. C++ is not very strict about what goes into a header file and what goes into a .cpp file. The Ada version creates an Ada package. The Ada package specification defines the task types named producer_Int and consumer_Int. The buffer shared by all instances of producer_int and consumer_int is defined within the Ada package body file.

Interface Specification Files

package pc_tasks is
   task type produce_Int (Id : Natural);
   task type consume_Int (Id : Natural);
end pc_tasks;

//
//  Buffer.hpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/31/21.
//

#ifndef Buffer_hpp
#define Buffer_hpp

#include <mutex>
#include <condition_variable>
#include <stdio.h>
 
#define BUFFER_CAPACITY 10

class Buffer {
    // Buffer fields
    int buffer [BUFFER_CAPACITY];
    int buffer_size;
    int left; // index where variables are put inside of buffer (produced)
    int right; // index where variables are removed from buffer (consumed)
    
    // Fields for concurrency
    std::mutex mtx;
    std::condition_variable not_empty;
    std::condition_variable not_full;
    
public:
    // Place integer inside of buffer
    void produce(int thread_id, int num);
    
    // Remove integer from buffer
    int consume(int thread_id);
    
    Buffer();
};

#endif /* Buffer_hpp */

This comparison shows the interface for Ada task types while the C++ interface (.hpp) file shows the interface for the Buffer class. The C++ interface definition defines the interfaces, both public and private, to the Buffer class.

Shared Buffer Implementations

The Ada package body contains both the definition and implementation of the shared buffer object and the implementation of the task types.

with Ada.Text_IO; use Ada.Text_IO;
with Ada.Numerics.Discrete_Random;

package body pc_tasks is
   type Index_T is mod 10; -- Modular type for 10 values
   type Circular_Array is array (Index_T) of Integer;

   protected buffer is
      entry produce (Item : in Integer);
      entry consume (Item : out Integer);
   private
      Buf      : Circular_Array;
      P_Index  : Index_T := Index_T'First;
      C_Index  : Index_T := Index_T'First;
      Buf_Size : Natural := 0;
   end buffer;

   protected body buffer is
      entry produce (Item : in Integer) when Buf_Size < Index_T'Modulus is
      begin
         Buf (P_Index) := Item;
         P_Index       := P_Index + 1;
         Buf_Size      := Buf_Size + 1;
      end produce;

      entry consume (Item : out Integer) when Buf_Size > 0 is
      begin
         Item     := Buf (C_Index);
         C_Index  := C_Index + 1;
         Buf_Size := Buf_Size - 1;
      end consume;
   end buffer;

   task body produce_Int is
      subtype decimal is Integer range 1 .. 10;
      package rand_int is new Ada.Numerics.Discrete_Random (decimal);
      use rand_int;
      value : decimal;
      seed  : Generator;
   begin
      Reset (seed);
      for I in 1 .. 4 loop
         value := Random (seed);
         buffer.produce (value);
         Put_Line ("Task" & Id'Image & " produced" & value'Image);
         delay 0.1;
      end loop;
   end produce_Int;

   task body consume_Int is
      Num : Integer;
   begin
      for I in 1 .. 6 loop
         buffer.consume (Num);
         Put_Line ("Task" & Id'Image & " consumed" & Num'Image);
         delay 0.1;
      end loop;
   end consume_Int;

end pc_tasks;

//
//  Buffer.cpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/31/21.
//

#include <iostream>
#include "Buffer.hpp"

Buffer::Buffer() {
    buffer_size = 0;
    left = 0;
    right = 0;
}

void Buffer::produce(int thread_id, int num) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    std::cout << "thread " << thread_id << " produced " << num << "\n";
    
    // Wait if the buffer is full
    not_full.wait(unique_lock, [this]() {
        return buffer_size != BUFFER_CAPACITY;
    });
    
    // Add input to buffer
    buffer[right] = num;
    
    // Update appropriate fields
    right = (right + 1) % BUFFER_CAPACITY;
    buffer_size++;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that buffer isn't empty
    not_empty.notify_one();
}

int Buffer::consume(int thread_id) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    // Wait if buffer is empty
    not_empty.wait(unique_lock, [this]() {
        return buffer_size != 0;
    });
    
    // Getvalue from position to remove in buffer
    int result = buffer[left];
    
    std::cout << "thread " << thread_id << " consumed " << result << "\n";
    
    // Update appropriate fields
    left = (left + 1) % BUFFER_CAPACITY;
    buffer_size--;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that the buffer isn't full
    not_full.notify_one();
    
    // Return result
    return result;
}

The Ada part of the package implementing the shared buffer is isolated below, along with a repeat of the C++ Buffer class implementation.

   type Index_T is mod 10; -- Modular type for 10 values
   type Circular_Array is array (Index_T) of Integer;

   protected buffer is
      entry produce (Item : in Integer);
      entry consume (Item : out Integer);
   private
      Buf      : Circular_Array;
      P_Index  : Index_T := Index_T'First;
      C_Index  : Index_T := Index_T'First;
      Buf_Size : Natural := 0;
   end buffer;

   protected body buffer is
      entry produce (Item : in Integer) when Buf_Size < Index_T'Modulus is
      begin
         Buf (P_Index) := Item;
         P_Index       := P_Index + 1;
         Buf_Size      := Buf_Size + 1;
      end produce;

      entry consume (Item : out Integer) when Buf_Size > 0 is
      begin
         Item     := Buf (C_Index);
         C_Index  := C_Index + 1;
         Buf_Size := Buf_Size - 1;
      end consume;
   end buffer;

//
//  Buffer.cpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/31/21.
//

#include <iostream>
#include "Buffer.hpp"

Buffer::Buffer() {
    buffer_size = 0;
    left = 0;
    right = 0;
}

void Buffer::produce(int thread_id, int num) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    std::cout << "thread " << thread_id << " produced " << num << "\n";
    
    // Wait if the buffer is full
    not_full.wait(unique_lock, [this]() {
        return buffer_size != BUFFER_CAPACITY;
    });
    
    // Add input to buffer
    buffer[right] = num;
    
    // Update appropriate fields
    right = (right + 1) % BUFFER_CAPACITY;
    buffer_size++;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that buffer isn't empty
    not_empty.notify_one();
}

int Buffer::consume(int thread_id) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    // Wait if buffer is empty
    not_empty.wait(unique_lock, [this]() {
        return buffer_size != 0;
    });
    
    // Getvalue from position to remove in buffer
    int result = buffer[left];
    
    std::cout << "thread " << thread_id << " consumed " << result << "\n";
    
    // Update appropriate fields
    left = (left + 1) % BUFFER_CAPACITY;
    buffer_size--;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that the buffer isn't full
    not_full.notify_one();
    
    // Return result
    return result;
}

 The Ada buffer definition begins by defining the array type to be used in the shared buffer. Ada allows arrays to be indexed by any discrete type. In this example an Ada modular type is declared and named Index_T. Type Index_T is declared to be "mod 10", which specifies that all its values are in the range of 0 through 9 and all the arithmetic operators return a value within this range. The only arithmetic operator used in this example is "+". Addition on a modular type is modular. Thus, in this example, 9 + 1 yields 0, which is exactly as needed for a circular buffer.

Type Circular_Array is an array indexed by Index_T. Every element of Circular_Array is an Integer.

Ada protected objects are protected against race conditions. They are specifically used for data shared by Ada tasks. Ada tasks are commonly implemented as operating system threads, but may also be used on a bare-bones system using only a compiler-generated Ada runtime.

The protected object is named buffer. It is separated into a specification and an implementation, but both parts in this example are contained in the package body. The protected specification declares the name of the protected object. There are three kinds of methods that may be used inside a protected object: procedures, entries and functions. Procedures have exclusive unconditional read-write access to the data in the protected object. Entries have conditional read-write access to the data in a shared object. Functions have shared read-only access to the data in the protected object. This example only uses two entries. The private portion of the protected object contains the definition of the data members in the protected object. Buf is an instance of Circular_Array. P_Index is an instance of the Index_T type and is used by the producer to add new data to the protected object. C_Index is an instance of Index_T type and is used by the consumer to index data read from the protected object. Buf_Size is an instance of the Natural subtype of Integer. Natural is a pre-defined subtype of Integer with a minimum value of 0. Buf_Size is initialized with the value 0.

The protected body implements the two entries. Each entry has an associated condition which must evaluate to True for the entry to execute. All tasks calling an entry while its controlling condition evaluates to False are implicitly placed in an entry queue for the called entry using a queuing policy of First-In-First-Out (FIFO). The next entry call in the queue is serviced as soon as the condition evaluates to TRUE. Tasks suspended in the entry queue are given access to the protected entry before any new tasks, thus maintaining the temporal condition of the calls.

The produce entry can only write to the protected object when Buf_Size is less than Index_T'Modulus, which in this case evaluates to 10. The consumer entry can only read from the protected object when Buf_Size is greater than 0.

Each entry implicitly handles all locking and unlocking of the protected object. The protected object is locked just before the "begin" reserved word in each entry and is unlocked just before the "end" reserved word in each entry. The modular nature of the index manipulations as well as the implicit lock manipulations explain the relatively simple Ada code compared with the more verbose corresponding C++ code.

Thread Implementations

The Ada task implementations are also contained in the package body while the C++ thread implementations are contained in the main.cpp file. Those corresponding source code sections are compared below.

   task body produce_Int is
      subtype decimal is Integer range 1 .. 10;
      package rand_int is new Ada.Numerics.Discrete_Random (decimal);
      use rand_int;
      value : decimal;
      seed  : Generator;
   begin
      Reset (seed);
      for I in 1 .. 4 loop
         value := Random (seed);
         buffer.produce (value);
         Put_Line ("Task" & Id'Image & " produced" & value'Image);
         delay 0.1;
      end loop;
   end produce_Int;

   task body consume_Int is
      Num : Integer;
   begin
      for I in 1 .. 6 loop
         buffer.consume (Num);
         Put_Line ("Task" & Id'Image & " consumed" & Num'Image);
         delay 0.1;
      end loop;
   end consume_Int;

// Takes in reference to a buffer and adds a random integer
void produceInt(Buffer &buffer) {
    for (int i = 0; i < 4; i++) {
        // Generate random number between 1 and 10
        int new_int = rand() % 10 + 1;
        buffer.produce(i, new_int);
        std::this_thread::sleep_for(std::chrono::milliseconds(100));
    }
}

// Takes in reference to a buffer and returns the latest int added
// in the buffer
void consumeInt(Buffer &buffer) {
    for (int i = 0; i < 6; i++) {
        buffer.consume(i);
        std::this_thread::sleep_for(std::chrono::milliseconds(100));
    }
}

The Ada tasks have visibility to the buffer protected object because the task bodies and the protected object are both defined within the same package body scope.

The produce_Int task defines an Integer subtype named decimal with valid values in the range of 1 through 10. That type is passed to the generic package Ada.Numerics.Discrete_Random so that random numbers in the range of 1 through 10 will be generated for this program. The random number seed is reset based upon the system clock using the Reset(seed) command. The for loop iterates through the values 1 through 4. Each iteration generates a new random value, writes the value to buffer.produce, outputs the value to standard output and then delays 0.1 seconds (100 milliseconds).

The consumer_int task defines a local integer variable named Num. The for loop iterates through the numbers 1 through 6. Each iteration calls buffer.consume, assigning the entry out parameter to Num, outputs the consumed value and delays 0.1 seconds.

Program Entry Points

While the program entry point for a C++ program is named "main", the program entry point for an Ada program may have any name chosen by the programmer. Note that the Ada entry point is a procedure, meaning it does not return any value and that procedure has no parameters.

with pc_tasks;    use pc_tasks;
with Ada.Text_IO; use Ada.Text_IO;

procedure Main is
begin
   Put_Line("Executing code in main...");
   declare
      Produce_Task0 : produce_Int (0);
      Consume_Task0 : consume_Int (1);
      Produce_Task1 : produce_Int (2);
      Consume_Task1 : consume_Int (3);
      Produce_Task2 : produce_Int (4);
   begin
      null;
   end;
   Put_Line ("Done!");
end Main;

int main(int argc, const char * argv[]) {
    std::cout << "Executing code in main...\n";
    
    // Initialize random seed
    srand (time(NULL));
    
    // Create Buffer
    Buffer buffer;
    
    // Create a thread to produce
    std::thread produceThread0(produceInt, std::ref(buffer));
    
    std::thread consumeThread0(consumeInt, std::ref(buffer));
    
    std::thread produceThread1(produceInt, std::ref(buffer));
    
    std::thread consumeThread1(consumeInt, std::ref(buffer));
    
    std::thread produceThread2(produceInt, std::ref(buffer));
    
    produceThread0.join();
    produceThread1.join();
    produceThread2.join();
    consumeThread0.join();
    consumeThread1.join();
    
    std::cout << "Done!\n";
    return 0;
}

Ada does not provide a "join()" method. Instead, the code block in which a task or set of task is declared cannot complete until all the tasks within that code block complete. The idiom shown above declared the 5 task type instances within an inner code block, which does not complete until all five of the tasks have terminated. Upon completion of that inner block the message "Done!" is output to standard output.

Complete code listings for both languages

package pc_tasks is
   task type produce_Int (Id : Natural);
   task type consume_Int (Id : Natural);
end pc_tasks;

//
//  Buffer.hpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/31/21.
//

#ifndef Buffer_hpp
#define Buffer_hpp

#include <mutex>
#include <condition_variable>
#include <stdio.h>
 
#define BUFFER_CAPACITY 10

class Buffer {
    // Buffer fields
    int buffer [BUFFER_CAPACITY];
    int buffer_size;
    int left; // index where variables are put inside of buffer (produced)
    int right; // index where variables are removed from buffer (consumed)
    
    // Fields for concurrency
    std::mutex mtx;
    std::condition_variable not_empty;
    std::condition_variable not_full;
    
public:
    // Place integer inside of buffer
    void produce(int thread_id, int num);
    
    // Remove integer from buffer
    int consume(int thread_id);
    
    Buffer();
};

#endif /* Buffer_hpp */

with Ada.Text_IO; use Ada.Text_IO;
with Ada.Numerics.Discrete_Random;

package body pc_tasks is
   type Index_T is mod 10; -- Modular type for 10 values
   type Circular_Array is array (Index_T) of Integer;

   protected buffer is
      entry produce (Item : in Integer);
      entry consume (Item : out Integer);
   private
      Buf      : Circular_Array;
      P_Index  : Index_T := Index_T'First;
      C_Index  : Index_T := Index_T'First;
      Buf_Size : Natural := 0;
   end buffer;

   protected body buffer is
      entry produce (Item : in Integer) when Buf_Size < Index_T'Modulus is
      begin
         Buf (P_Index) := Item;
         P_Index       := P_Index + 1;
         Buf_Size      := Buf_Size + 1;
      end produce;

      entry consume (Item : out Integer) when Buf_Size > 0 is
      begin
         Item     := Buf (C_Index);
         C_Index  := C_Index + 1;
         Buf_Size := Buf_Size - 1;
      end consume;
   end buffer;

   task body produce_Int is
      subtype decimal is Integer range 1 .. 10;
      package rand_int is new Ada.Numerics.Discrete_Random (decimal);
      use rand_int;
      value : decimal;
      seed  : Generator;
   begin
      Reset (seed);
      for I in 1 .. 4 loop
         value := Random (seed);
         buffer.produce (value);
         Put_Line ("Task" & Id'Image & " produced" & value'Image);
         delay 0.1;
      end loop;
   end produce_Int;

   task body consume_Int is
      Num : Integer;
   begin
      for I in 1 .. 6 loop
         buffer.consume (Num);
         Put_Line ("Task" & Id'Image & " consumed" & Num'Image);
         delay 0.1;
      end loop;
   end consume_Int;

end pc_tasks;

//
//  Buffer.cpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/31/21.
//

#include <iostream>
#include "Buffer.hpp"

Buffer::Buffer() {
    buffer_size = 0;
    left = 0;
    right = 0;
}

void Buffer::produce(int thread_id, int num) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    std::cout << "thread " << thread_id << " produced " << num << "\n";
    
    // Wait if the buffer is full
    not_full.wait(unique_lock, [this]() {
        return buffer_size != BUFFER_CAPACITY;
    });
    
    // Add input to buffer
    buffer[right] = num;
    
    // Update appropriate fields
    right = (right + 1) % BUFFER_CAPACITY;
    buffer_size++;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that buffer isn't empty
    not_empty.notify_one();
}

int Buffer::consume(int thread_id) {
    // Acquire a unique lock on the mutex
    std::unique_lock<std::mutex> unique_lock(mtx);
    
    // Wait if buffer is empty
    not_empty.wait(unique_lock, [this]() {
        return buffer_size != 0;
    });
    
    // Getvalue from position to remove in buffer
    int result = buffer[left];
    
    std::cout << "thread " << thread_id << " consumed " << result << "\n";
    
    // Update appropriate fields
    left = (left + 1) % BUFFER_CAPACITY;
    buffer_size--;
    
    // Unlock unique lock
    unique_lock.unlock();
    
    // Notify a single thread that the buffer isn't full
    not_full.notify_one();
    
    // Return result
    return result;
}

with pc_tasks;    use pc_tasks;
with Ada.Text_IO; use Ada.Text_IO;

procedure Main is
begin
   Put_Line("Executing code in main...");
   declare
      Produce_Task0 : produce_Int (0);
      Consume_Task0 : consume_Int (1);
      Produce_Task1 : produce_Int (2);
      Consume_Task1 : consume_Int (3);
      Produce_Task2 : produce_Int (4);
   begin
      null;
   end;
   Put_Line ("Done!");
end Main;

//
//  main.cpp
//  ProducerConsumer
//
//  Created by Andrew Wei on 5/30/21.
//

#include <thread>
#include <iostream>
#include "Buffer.hpp"
#include <stdlib.h>

// Takes in reference to a buffer and adds a random integer
void produceInt(Buffer &buffer) {
    for (int i = 0; i < 4; i++) {
        // Generate random number between 1 and 10
        int new_int = rand() % 10 + 1;
        buffer.produce(i, new_int);
        std::this_thread::sleep_for(std::chrono::milliseconds(100));
    }
}

// Takes in reference to a buffer and returns the latest int added
// in the buffer
void consumeInt(Buffer &buffer) {
    for (int i = 0; i < 6; i++) {
        buffer.consume(i);
        std::this_thread::sleep_for(std::chrono::milliseconds(100));
    }
}

int main(int argc, const char * argv[]) {
    std::cout << "Executing code in main...\n";
    
    // Initialize random seed
    srand (time(NULL));
    
    // Create Buffer
    Buffer buffer;
    
    // Create a thread to produce
    std::thread produceThread0(produceInt, std::ref(buffer));
    
    std::thread consumeThread0(consumeInt, std::ref(buffer));
    
    std::thread produceThread1(produceInt, std::ref(buffer));
    
    std::thread consumeThread1(consumeInt, std::ref(buffer));
    
    std::thread produceThread2(produceInt, std::ref(buffer));
    
    produceThread0.join();
    produceThread1.join();
    produceThread2.join();
    consumeThread0.join();
    consumeThread1.join();
    
    std::cout << "Done!\n";
    return 0;
}

 Summary

The same producer-consumer problem can be solved using both Ada and C++. The C++ approach to the producer-consumer pattern requires far more attention to low level details than does the Ada approach to the producer-consumer pattern. The greatest difference in the two approaches is the requirement for the C++ programmer to explicitly manipulate mutex locking/unlocking and suspended thread wait and notification commanding.

A recent question on StackOverflow asked about whether or not a mutex is locked twice by the same thread without unlocking the mutex. See https://stackoverflow.com/questions/76890110/c-the-thread-repeatedly-locks-the-same-mutex-multiple-times?noredirect=1#comment135552146_76890110

The C++ source code presented in the question is:

1 #include <exception>
2 #include <memory>
3 #include <mutex>
4 #include <stack>
5
6 struct empty_stack : std::exception
7 {
8    const char* what() const throw();
9 };
10
11 template<typename T>
12 class threadsafe_stack
13 {
14 private:
15    std::stack<T> data;
16    mutable std::mutex m;
17 public:
18    threadsafe_stack(){}
19    threadsafe_stack(const threadsafe_stack& other) {
20       std::lock_guard<std::mutex> lock(other.m);
21       data = other.data;
22 }
23    threadsafe_stack& operator=(const threadsafe_stack& other) = delete;
24    void push(T new_value)
25    {
26       std::lock_guard<std::mutex> lk(m);
27       data.push(std::move(new_value));
28    }
29    std::shared_ptr<T> pop()
30    {
31       std::lock_guard<std::mutex> lock(m);
32       if (data.empty()) throw empty_stack();
33       std::shared_ptr<T> const res(std::make_shared<T>(data.top()));
34       data.pop();
35       return res;
36    }
37    void pop(T& value)
38    {
39       std::lock_guard<std::mutex> lock(m);
40       if (data.empty()) throw empty_stack();
41       *value = data.top();
42       data.pop();
43    }
44    bool empty() const
45    {
46       std::lock_guard<std::mutex> lock(m);
47       return data.empty();
48    }
49 };
 

One issue with this program design, which has nothing to do with the question asked, is throwing an exception when pop is called on an empty stack. An empty stack is not an exceptional situation. In fact the stack starts off as an empty stack when it is created.

Exception handling should not be used to deal with non-exceptional program state.

The following Ada stack implementation demonstrates how an empty stack should be handled in a thread-safe manner.

The Ada program defines a generic thread package with the same operations shown in the C++ example above.

The Ada program creates a generic stack package and a main procedure. Ada packages are separated into two parts, the package specification and the package body or implementation. The package specification defines the package API.

1 with Ada.Containers.Doubly_Linked_Lists;
2 generic
3    type element_type is private;
4 package generic_stack is
5    package List_Pack is new
6       Ada.Containers.Doubly_Linked_Lists (element_type);
7 use List_Pack;
8
9 protected type Stack is
10    procedure Push (Item : element_type);
11    entry Pop (Item : out element_type);
12    function Is_Empty return Boolean;
13 private
14    Buffer : List := Empty_List;
15 end Stack;
16
17 end generic_stack;
 

The package creates an Ada protected type. A protected type or protected object is a passive unit of concurrency which is implicitly protected from race conditions. There are three kinds of methods available to be defined within a protected type. Protected procedures implicitly acquire an unconditional exclusive read-write lock on the data in the protected type. Protected entries implicitly acquire a conditional exclusive read-write lock on the data in the protected type. A task suspends while the entry boundary condition evaluates to false. The suspended tasks are placed in an entry queue so that they can be activated when the boundary condition evaluates to true. The third kind of method available for a protected object is a function. Protected functions are limited to read-only access to the data in the protected type. Protected functions acquire a shared read lock on the data in the protected object.

The implementation of the package is in a separate file containing the package body.

1  package body generic_stack is
2
3     -----------
4     -- Stack --
5     -----------
6
7     protected body Stack is
8
9        ----------
10       -- Push --
11       ----------
12
13       procedure Push (Item : element_type) is
14       begin
15          Buffer.Append (Item);
16       end Push;
17
18       ---------
19       -- Pop --
20       ---------
21
22       entry Pop (Item : out element_type) when
23                  not Buffer.Is_Empty is
24       begin
25          Item := Buffer.Last_Element;
26          Buffer.Delete_Last;
27       end Pop;
28
29       --------------
30       -- Is_Empty --
31       --------------
32
33       function Is_Empty return Boolean is
34       begin
35          return Buffer.Is_Empty;
36       end Is_Empty;
37
38    end Stack;
39
40 end generic_stack;

The push procedure implicitly locks the Buffer when it begins executing and unlocks the Buffer as the procedure completes.

The pop entry only executes when Buffer.Is_Empty evaluates to false. When that condition is met any immediate call, or any task waiting in the entry queue is allowed to execute the entry. The default queuing policy of the entry queue is FIFO and all tasks suspended in the entry queue will be executed before any new call can be executed. The shared read-write lock is applied when execution of the entry starts and is release upon completion of the entry.

The Is_Empty function can only execute when no read-write lock is applied to the Buffer data element. Upon starting execution the function applies a shared read lock, allowing any number of tasks to read simultaneously and preventing any procedure or function to execute until the function completes and releases its shared read lock.

The main procedure, which is the program entry point for this example, follows.

1  with Ada.Text_IO; use Ada.Text_IO;
2  with generic_stack;
3
4  procedure Main is
5     package int_stack is new generic_stack (Integer);
6     use int_stack;
7
8     The_Stack : Stack;
9
10    task producer is
11       entry Start;
12    end producer;
13
14    task body producer is
15    begin
16       accept Start;
17       Put_Line ("Producer started.");
18       for I in 1 .. 20 loop
19          The_Stack.Push (I);
20          Put_Line ("Pushed" & I'Image);
21          delay 0.001;
22       end loop;
23    end producer;
24
25    task consumer is
26       entry Start;
27    end consumer;
28
29    task body consumer is
30       Num : Integer;
31    begin
32       accept Start;
33       Put_Line ("Consumer started.");
34       for I in 1 .. 20 loop
35          The_Stack.Pop (Num);
36          Put_Line ("Popped" & Num'Image);
37       end loop;
38    end consumer;
39
40 begin
41    consumer.Start;
42    delay 0.01;
43    producer.Start;
44 end Main;
 

Line 5 creates an instance of the generic_stack package passing the type Integer as the generic parameter. This results in a stack of integer values.

Line 8 declares an instance of the Stack type named The_Stack.

Line 10 declares the interface for the producer task. This task has one entry named Start.

Line 14 declares the implementation of the producer task.

Line 16 accepts the Start entry. The producer task will suspend until another task calls its Start entry. Task entries implement a Rendezvous logic allowing synchronous coordination of tasks. After accepting Start the producer executes a for loop 20 times, each time pushing the current value of the loop control variable onto the stack, displaying a message to standard output and then delaying (sleeping) for 0.001 seconds.

The consumer task declaration begins at line 25. The consumer also has a Start entry.

The consumer task accepts its start entry at line 32 and then pops 20 values off of the stack. The consumer task has no delay statement. The consumer tasks will then complete its iteration before the producer task pushes another value onto the stack. Each call to the pop entry will therefore encounter an empty stack until the producer pushes another value onto the stack.

Both the producer task and the consumer tasks begin execution immediately, but both tasks suspend until their Start entry is called.

Line 40 begins execution of the main procedure which is also the top-level task for the program. Line 41 calls consumer.Start. Line 42 causes the main procedure to delay for 0.01 seconds before line 43 calls producer.start.

This delay in starting the tasks ensures that the consumer must wait at least 0.01 seconds before the first data will be pushed onto the stack. In other words, the stack will be empty for at least 0.01 seconds.

The output of this program is:

Consumer started.
Producer started.
Pushed 1
Popped 1
Pushed 2
Popped 2
Pushed 3
Popped 3
Pushed 4
Popped 4
Pushed 5
Popped 5
Pushed 6
Popped 6
Pushed 7
Popped 7
Pushed 8
Popped 8
Pushed 9
Popped 9
Pushed 10
Popped 10
Pushed 11
Popped 11
Pushed 12
Popped 12
Pushed 13
Popped 13
Pushed 14
Popped 14
Pushed 15
Popped 15
Pushed 16
Popped 16
Pushed 17
Popped 17
Pushed 18
Popped 18
Pushed 19
Popped 19
Pushed 20
Popped 20
 

Clearly the program did not encounter any exceptions, yet the consumer could only pop values after they were pushed onto the stack.

Arrays in both C and Ada are a compound type with the elements arranged sequentially in memory. All the elements in an array are of a single type. For instance an array may contain integer elements or it may contain floating point elements, or it may contain strings of characters, or it may even contain other arrays.

One might therefore assume that arrays in C and Ada are fundamentally the same. This article explores that assumption and finds some clear differences between C arrays and Ada arrays.

The impact of these differences is demonstrated by reviewing the Merge-Sort algorithm implemented in both languages.

The first example is a C implementation of the Merge-Sort algorithm.

1 /*     
2 * C Program to Perform Merge Sort using Recursion and Functions
3 */
4
5 #include <stdio.h>
6 #include <stdlib.h>
7
8 // merge function
9 void Merge(int arr[], int left, int mid, int right)
10 {
11    int i, j, k;
12    int size1 = mid - left + 1;
13    int size2 = right - mid;
14
15    // created temporary array
16    int Left[size1], Right[size2];
17
18    // copying the data from arr to temporary array
19    for (i = 0; i < size1; i++)
20        Left[i] = arr[left + i];
21
22    for (j = 0; j < size2; j++)
23       Right[j] = arr[mid + 1 + j];
24
25    // merging of the array
26    i = 0; // intital index of first subarray
27    j = 0; // inital index of second subarray
28    k = left; // initial index of parent array
29    while (i < size1 && j < size2)
30    {
31        if (Left[i] <= Right[j])
32        {
33            arr[k] = Left[i];
34            i++;
35        }
36        else
37        {
38            arr[k] = Right[j];
39            j++;
40        }
41        k++;
42    }
43
44    // copying the elements from Left[], if any
45    while (i < size1)
46    {
47        arr[k] = Left[i];
48        i++;
49        k++;
50    }
51
52    // copying the elements from Right[], if any
53    while (j < size2)
54    {
55        arr[k] = Right[j];
56        j++;
57        k++;
58    }
59 }
60
61 //merge sort function
62 void Merge_Sort(int arr[], int left, int right)
63 {
64    if (left < right)
65    {
66
67        int mid = left + (right - left) / 2;
68
69        // recursive calling of merge_sort
70        Merge_Sort(arr, left, mid);
71        Merge_Sort(arr, mid + 1, right);
72
73        Merge(arr, left, mid, right);
74    }
75 }
76
77 // driver code
78 int main()
79 {
80     int size;
81     printf("Enter the size: ");
82     scanf("%d", &size);
83
84     int arr[size];
85     printf("Enter the elements of array: ");
86     for (int i = 0; i < size; i++)
87     {
88         scanf("%d", &arr[i]);
89     }
90
91     Merge_Sort(arr, 0, size - 1);
92
93     printf("The sorted array is: ");
94     for (int i = 0; i < size; i++)
95     {
96         printf("%d ", arr[i]);
97     }
98     printf("\n");
99     return 0;
100 } 

Both the Merge_Sort function and the Merge function take several parameters.

The first parameter in each function specifies an array of int elements. The remaining parameters specify various int values. These extra parameters are needed in C because all array indices must begin at 0, the size of the array parameter is not passed with the array, and C does not provide array slicing.

The C programmer must provide all this information as additional function parameters.

The C language does not provide assignment of one array to another. The programmer must explicitly create a loop and assign each element one at a time:

18    // copying the data from arr to temporary array
19    for (i = 0; i < size1; i++)
20        Left[i] = arr[left + i];
21
22    for (j = 0; j < size2; j++)
23       Right[j] = arr[mid + 1 + j];
 

Each of these examples is a demonstration of the low-level terseness of C. Each of these examples shows how much more writing must be done in C to achieve what Ada does very simply.

The Ada version of the Merge-Sort algorithm is implemented in a generic Ada package so that the sort procedure declared within the package specification can be used to sort any mutable data type.

The Ada solution is done using three filles.

1 generic
2    type element_type is private;
3    type array_type is array (Integer range <>) of element_type;
4    with function "<" (left, right : element_type) return Boolean is <>;
5 package generic_merge_sort is
6    procedure sort (Item : in out array_type);
7 end generic_merge_sort;

The generic package specification requires three parameters when an instance of the package is declared. The first parameter is the name of the element type for the array which will be sorted. The second parameter is the name of the array type to be sorted. This parameter is restricted to an unconstrained array type indexed by Integer values. Each element of the array type must be the type passed to the first parameter. The third parameter is a possible overloading of the “<” function. This is an optional parameter. If the element_type already has a “<” function defined that function will be used by default.

The procedure sort is defined to take one parameter with mode “in out”, meaning the parameter value(s) will be used and possibly modified within the procedure. All modifications will be visible to the calling scope.

The package body contains the logic used to implement the merge-sort algorithm.

1 package body generic_merge_sort is
2    procedure merge (Item : in out array_type) is
3       mid   : Integer    := Item'First + (Item'Last - Item'First) / 2;
4       Left  : array_type := Item (Item'First .. mid);
5       Right : array_type := Item (mid + 1 .. Item'Last);
6       I     : Integer    := Left'First;
7       J     : Integer    := Right'First;
8       K     : Integer    := Item'First;
9    begin
10      while I <= Left'Last and then J <= Right'Last loop
11         if Left (I) < Right (J) then
12            Item (K) := Left (I);
13            I        := I + 1;
14         else
15            Item (K) := Right (J);
16            J        := J + 1;
17         end if;
18         K := K + 1;
19      end loop;
20      -- copying unused items from Left
21      while I <= Left'Last loop
22         Item (K) := Left (I);
23         I        := I + 1;
24         K        := K + 1;
25      end loop;
26      -- copying unused items from right
27      while J <= Right'Last loop
28         Item (K) := Right (J);
29         J        := J + 1;
30         K        := K + 1;
31      end loop;
32   end merge;
33
34   ----------
35   -- sort --
36   ----------
37
38   procedure sort (Item : in out array_type) is
39      mid : Integer;
40   begin
41      if Item'Length > 1 then
42         mid := Item'First + (Item'Last - Item'First)/2;
43         sort(Item(Item'First .. Mid));
44         sort(Item(Mid + 1 .. Item'Last));
45         merge(Item);
46      end if;
47   end sort;
48
49 end generic_merge_sort;

This example make extensive use of Ada array attributes. The value of mid in both the merge procedure and the sort procedure is calculated using the ‘First and ‘Last array attributes, simplifying the procedure signature for both sort and merge. Both procedures only need an instance or slice of an array passed to them.

In the merge procedure the Left and Right instances of array_type are created and initialized with slices of the array Item passed to the procedure. Ada’s ability to assign one array or slice to another array or slice of the same type eliminates the need to explicitly calculate the sizes needed for the Left and Right arrays. It also eliminates the need to explicitly copy the values iteratively.

4       Left  : array_type := Item (Item'First .. mid);
5       Right : array_type := Item (mid + 1 .. Item'Last);

For reference, here is the corresponding C code:

12    int size1 = mid - left + 1;
13    int size2 = right - mid;
14
15    // created temporary array
16    int Left[size1], Right[size2];
17
18    // copying the data from arr to temporary array
19    for (i = 0; i < size1; i++)
20        Left[i] = arr[left + i];
21
22    for (j = 0; j < size2; j++)
23       Right[j] = arr[mid + 1 + j];

A comparison of the C merge_sort function and the Ada sort procedure provides more insight to the differences between C arrays and Ada arrays.

C merge_sort:

62 void Merge_Sort(int arr[], int left, int right)
63 {
64    if (left < right)
65    {
66
67        int mid = left + (right - left) / 2;
68
69        // recursive calling of merge_sort
70        Merge_Sort(arr, left, mid);
71        Merge_Sort(arr, mid + 1, right);
72
73        Merge(arr, left, mid, right);
74    }
75 }

Ada sort:

38   procedure sort (Item : in out array_type) is
39      mid : Integer;
40   begin
41      if Item'Length > 1 then
42         mid := Item'First + (Item'Last - Item'First)/2;
43         sort(Item(Item'First .. Mid));
44         sort(Item(Mid + 1 .. Item'Last));
45         merge(Item);
46      end if;
47   end sort;

The condition in the “if” statement, while achieving the same result, is quite different. The C function relies upon the correctness of the left and right parameters passed as the second and third arguments to the function. The Ada procedure only relies upon the ‘Length attribute of the array. The Ada syntax is much less error-prone and a more direct statement to a human reader about the intention of the conditional statement. In order to get a complete understanding of the purposes of the C left and right parameters one must read the context of the main procedure, while the meaning of the Ada conditional is completely clear in the local context. Since the sort parameter type is an unconstrained array type every slice of the array is also an instance of the unconstrained array type.

The C Merge_Sort function recursively calls itself passing the beginning of the arr parameter each time but with different values for left and right.

The Ada sort function recursively calls itself passing slices of the parameter Item to each recursive call. In this manner each recursion of Sort only sees the slice passed to it and all the other elements of the array initially passed to the sort procedure from main are not available to the recursive call.

The C Merge procedure passes the full array, but also passes three parameters containing the left, mid and right index positions in the array. The Ada merge procedure only array slice passed to it by the sort procedure.

The main function in C is:

78 int main()
79 {
80     int size;
81     printf("Enter the size: ");
82     scanf("%d", &size);
83
84     int arr[size];
85     printf("Enter the elements of array: ");
86     for (int i = 0; i < size; i++)
87     {
88         scanf("%d", &arr[i]);
89     }
90
91     Merge_Sort(arr, 0, size - 1);
92
93     printf("The sorted array is: ");
94     for (int i = 0; i < size; i++)
95     {
96         printf("%d ", arr[i]);
97     }
98     printf("\n");
99     return 0;
100 }

The main procedure in Ada is:

1 with Ada.Text_IO;         use Ada.Text_IO;
2 with Ada.Integer_Text_IO; use Ada.Integer_Text_IO;
3 with generic_merge_sort;
4
5 procedure Main is
6    type int_arr is array (Integer range <>) of Integer;
7    package int_sort is new generic_merge_sort
8           (element_type => Integer, array_type => int_arr);
9    use int_sort;
10   Num_Elements : Positive;
11 begin
12   Put ("Enter the size of the array: ");
13   Get (Num_Elements);
14   declare
15      the_array : int_arr (1 .. Num_Elements);
16   begin
17      Put_Line ("Enter the array values:");
18      for value of the_array loop
19         Get (value);
20      end loop;
21      sort (the_array);
22      Put_Line ("The sorted array is:");
23      for value of the_array loop
24         Put (value'Image & " ");
25      end loop;
26      New_Line;
27   end;
28 end Main;

Lines 6 through 9 of the Ada program are needed to create an instance of the generic_merge_sort package.

6    type int_arr is array (Integer range <>) of Integer;
7    package int_sort is new generic_merge_sort
8           (element_type => Integer, array_type => int_arr);
9    use int_sort;

An inner block is declared so that an array of the size entered by the user can be created. The rest of the program is performed in this inner block.

14   declare
15      the_array : int_arr (1 .. Num_Elements);
16   begin
17      Put_Line ("Enter the array values:");
18      for value of the_array loop
19         Get (value);
20      end loop;
21      sort (the_array);
22      Put_Line ("The sorted array is:");
23      for value of the_array loop
24         Put (value'Image & " ");
25      end loop;
26      New_Line;
27   end;

The for loop used to read all the values into the array starting at line 18 is an Ada iterator loop. The loop parameter “value” becomes an alias for each array element starting at the first element and ending at the last element. Thus, each value entered by the user is placed directly into the array without explicit index notation.

The array is sorted.

21      sort (the_array);

Finally, the sorted array is output to standard output and the program ends.

Conclusion:

The C and Ada examples implement the same merge-sort algorithm to sort an array of integer values. C arrays are more primitive abstractions than are Ada arrays. The more primitive abstractions require more lines of C source code than does the Ada implementation. The C implementation also requires a more complex parameter list for both its Merge function and its Merge_Sort function. The Ada implementation concentrates on merely sorting an array without additional parameters.

The built-in characteristics of Ada arrays really do simplify the manipulation of arrays, even when those arrays are passed to an Ada function or procedure.

I learned long ago that complexity is the enemy of correctness. In these examples we see that lower complexity also provides fewer opportunities for programmer error. We also see that a programming language with a low level syntax is not necessarily simpler than a programming language with a higher level syntax. Terse is not always simpler. Terse does not always result in less typing for the programmer.

Many programming languages support concurrent behavior through the creation of threads and the explicit manipulation of semaphores, locks or mutexes. The C language seems to have initiated this approach to handling concurrency.

C example

The following example of using a mutex with thread comes from C Program to Show Thread Interface and Memory Consistency Errors – GeeksforGeeks

The C threading model uses the pthread library to create threads. The behavior of each created thread is controlled by the function passed to the thread as part of the pthread_create function parameter list. First an array of pthread_t structures is created:

Next, the Id number of each thread is created as the function named thread_function is passed to each element of the thread as a parameter to the pthread_create function:

The third parameter to the pthread_create command is the function passed to the newly created thread. The fourth argument is the parameter passed to the function passed to the thread. Not only is the thread function parameter passed as the fourth argument to the pthread_create function, it is also cast to a pointer to void. This bit of syntax can be confusing because the value being passed is not a pointer, but rather an int.

Let's now look inside the function named thread_function, but before that, let's look at the creation of the pthread_mutex used to control access to the shared counter.

Yes, the pthread_mutex_t instance used to control access to the shared counter is itself a shared instance of a type. You might also see that the shared_counter_mutex is only loosely connected to the shared_counter integer variable. That loose connection is a voluntary connection not enforced by any syntax in the C pthreads library.

The thread_function is defined as:

The thread function must explicitly lock the shared_counter mutex, then increment the shared_counter, then output the current value of the shared_counter and finally unlock the shared_counter_mutex.

A major source of confusion

The order of the locking modifying and unlocking the mutex is critical. The shared_counter knows nothing of these locks. The pthread_mutex_t has no syntactical connection to the shared_counter. On the other hand, failing to lock, perform operations and then unlock the mutex will result in semantic failures which prevent the mutex from properly locking the shared_counter and then unlocking the shared counter after the operations are completed. Even worse, there is no syntax in the C pthread library prohibiting a thread from simply accessing the shared_counter while completely ignoring the mutex lock and unlock behaviors.

Further program issues

The remainder of the program calls pthread_join, forcing the main thread to wait until all the pthreads have completed before continuing to execute its sequence of instructions.

It is important to understand that the pthread_join function called for each thread can fail and return failure status.

Once the pthreads have completed the program frees all the pointers in the array of pointers to pthread_t that were created to create the array of threads. This is done because it is always correct to free dynamically allocated memory when that memory is no longer needed in the program.

While this action is not particularly confusing, it is yet another detail that should be explicitly implemented.

Ada Example

The Ada programming language has tasking, which is roughly equivalent to threads, built into the core language since the first Ada language standard in 1983. In the 1995 standard protected types were added to the core language, which allow asynchronous communication between tasks. There are no special Ada libraries to implement tasking or protected types.

A protected type or protected object is protected against inappropriate simultaneous access to a shared data structure. A protected type or object is built very much in the philosophy of Object Oriented Programming inasmuch as the protected object has programmer-defined behaviors, however the locking and unlocking of the protected object is performed implicitly by the object itself and not through explicit lock and unlock methods called by the task accessing the protected object.

There are three kinds of protected methods.

• Protected procedures – Protected procedures control an unconditional exclusive read-write lock on the protected object. The task calling the protected procedure can access the protected object whenever the object is unlocked.

• Protected entries – Protected entries control a conditional exclusive read-write lock on the protected object. The calling task can only execute the entry when its boundary condition evaluates to True and only then can the task implicitly acquire the exclusive read-write lock.

• Protected functions – Protected functions control an unconditional shared read lock on the protected object. This allows multiple tasks to simultaneously read from the protected object at the same time, but prevents any task to execute any procedure or entry while tasks are calling protected functions.

The following Ada example creates a protected object implementing the shared counter. The protected object in this example implements one procedure and one function. The source code for this program is:

Ada allows functions, procedures, protected objects and task types to be declared within any function, procedure or task type. This means all items declared in this program are completely local to this program and not “static” as are multiple functions created in the same file in C.

Ada variables are declared by stating the name of the variable, followed by a colon, followed by the type of the variable. This is opposite the order of declaring variables in C where the type is listed first followed by the variable name.

The variable name is Num_Tasks. The type is Positive, which is a pre-defined subtype of the Ada Integer type. Integer is equivalent to the C int type. Positive is an Integer with a minimum possible value of 1 and a maximum possible value of Integer'Last, which is the same as MAX_INT in C. This variable is used to contain the number of tasks the user specifies during the execution of the program.

The protected object named counter is declared in two parts. Protected objects are always declared in two parts. The protected specification defines the public view of the protected methods along with a private view of the data elements in the protected object. The protected body contains the definition of the protected methods, and is not visible to any task calling the protected object.

The protected specification for the counter protected object is:

There are two methods for this protected object. The procedure named update has one parameter which passes a value of type Natural out to the calling task. Natural is a predefined subtype of Integer with a minimum value of 0. The function named get_value returns a value of the subtype Natural.

In the private section of the protected specification one data element is declared. It is a variable named count. The type of the variable is Natural and it is initialized to 0.

The protected body for the counter protected object is:

The update procedure increments count and passes its current value out through the parameter named The_Count.

The function get_value simply returns the value of count. The Ada compiler will issue an error message if the programmer attempts to modify the count data member of the protected object within the execution of the get_value function. Functions are read-only methods in a protected object.

The procedure update implicitly implements a read-write lock and the function get_value implicitly implements a read lock.

The task type named thread is defined in two parts. The first part, the task specification, defines that name of the task type and the direct interfaces to the task type. In this case one task entry is defined. That entry is used to set the task ID.

The task entry implements a direct synchronous communication channel to each instance of the thread type. The entry synchronization scheme is called a Rendezvous. The word "rendezvous" is a French word meaning "a meeting at an agreed time and place". A task calling another task's entry is will suspend until the task declaring the entry accepts the entry call. Similarly, a task accepting an entry will suspend until another task calls that entry. Once both the task calling the entry and the task accepting the entry are in this state for an overlapping time period any data specified in the entry specification is passed between the calling task and the called task. Once the entry has completed both tasks continue to execute concurrently.

The behavior of the task type is defined in the task body.

The task body declared two local variables. Each instance of the task type has unique instances of these two variables.

The first action taken by the thread task is to accept the set_id entry, passing the Id value from a calling task to this task. The accept statement assigns the value of Id to the task's local variable named Me. Task entries implement a Rendezvous behavior. The thread task will wait at the accept statement until its entry is called by another task. Once the value is passed to the thread task both tasks will the resume concurrent behavior.

The thread task calls the protected object's update procedure, using its local variable My_Count to receive the current value of the counter protected object. Finally, the thread task simply outputs the value it received from the counter.update procedure.

The next “begin” begins the execution of the Main task.

The Main task prompts the user for the number of tasks to create and reads the number entered by the user.

An inner block is created starting at the “declare” reserved word. That inner block has a declarative section in which an array of thread tasks, equal in number to the value entered by the user, is created. The tasks in the array named pool begin executing immediately. Their first responsibility is to accept the set_id entry, so each task waits until its set_id entry is called.

The “for” loop iterates through each element in the pool array assigning the element's index value as the ID for the task.

Immediately after accepting its ID number each task executes counter.update and then outputs the value of the counter passed back through the counter.update procedure.

The inner block will only complete when all the tasks created within the block complete. This produces the same effect as the “join” command in the C example.

After the inner block completes the Main task calls the counter.get_value function and displays the final value of the shared counter. Completion of the inner block automatically frees the array of task objects which were created on the stack at the start of the inner block. No explicit loop to free the task elements is needed or even possible.

Conclusion

While the two programs above achieve the same behavior dealing with allowing multiple threads or tasks to update a shared counter, the C solution contains more opportunities for programmer confusion and error. It also requires a lot more coding by the programmer than the Ada solution.

 Overview

It is often asserted that all general purpose programming languages can solve the same set of problems. This means that no one programming language has functional advantages over any other programming language.

That assertion is only mostly true, which means it is false.

For example, weakly typed languages can perform and make use of implicit type conversions while strongly typed languages cannot. Strongly typed languages must employ explicit conversions to achieve a similar effect.

The purpose of this article is to begin discussing some of the things that are difficult in one commonly used language and relatively easy in another.

Scalar Ranges

Programming languages derived from Pascal syntax allow scalar types and subtypes to be defined by the programmer, while programming languages derived from C syntax do not allow the programmer to define scalar types or subtypes.

In C++, for example, a class must be declared encapsulating the behavior of a scalar type with a programmer specified range of values. Making a subtype of that class then requires the creation of an inherited class expressing the restrictions distinguishing the subtype.

In C++ enums are encapsulated in a class as illustrated by the following Stack Overflow issue:

How can I implicitly convert an enum to its subset and vice versa in C++?

More precisely, the feature I want is like implicitly convert an enum to its subset enum and vice versa.

The code I wish it working:

But enum cannot do the conversion. Now I have two options:

// 1. static_assert with template parameter

template <Human H>

void female_func() {

    static_assert(H == Human::B);

    // ...

}

// 2. manually convert it

#define _ENUM_TO_ENUM(e1, e2) \

    static_cast<e2>(static_cast<std::underlying_type_t<decltype(e1)>>(e1))

void female_func(_ENUM_TO_ENUM(SOMEONE, Female)) {

    // But this way the compiler does not check if the value is valid.

    // I can put anything in.

    // ...

}

As is shown above, the concept of a scalar range and subrange is complicated by the need in C++ to express such a type as a class.

One answer provided to this question is

enum class Gender {MALE, FEMALE};

struct Human

{

    Gender m_gender;

    Human(Gender g) : m_gender{g}

    {}

    virtual ~Human() = default;

};

struct Man : public Human

{

    Man() : Human{Gender::MALE}

    {}

};

struct Woman : public Human

{

    Woman() : Human(Gender::FEMALE)

    {}

};

void human_func(const Human & h)

{

    //...

}

void man_func(const Man & m)

{

    //...

}

void woman_func(const Woman & w)

{

    //...

}

It is clear that this approach may work for an enum with 2 values, but becomes unusable with an enum containing tens or hundreds of values.

The Ada programming language, on the other hand, uses the concept of scalar ranges and subtypes extensively.

The Character type in Ada is an enumeration type with the range of values expressed as nul .. 'ÿ'. The ASCII characters are a subset of the Character type with value in the range of nul .. del. Within the ASCII characters the upper case characters are the range ‘A’ .. ‘Z’ and the lower characters are the range ‘a’ .. ‘z’.

If the programmer wants to pass only upper case characters as a parameter to a procedure the procedure can be defined as

This procedure will only accept characters in the range specified by the subtype Upper.

A function that counts all the upper case characters in a string can be defined as

The Ada samples above exhibit the behavior and usage requested by the person using C++ in the Stack Overflow question above.

The Ada program is not encumbered with the heavy syntax and rules associated with C++ classes.

I found the following programming example in Bit Array in C - Sanfoundry

The program creates a bit array containing 58 elements. The program works as expected and manages to illustrate very arcane features of the C programming language.

I challenge anybody not familiar with C bit shifting to completely understand how the two void functions named get_bit and toggle_bit actually work.

These functions are examples of the “simplicity” of C programming. While the syntax is compact the semantics of these functions are not. Describing what these two functions do would take several paragraphs of text.

As an avid Ada programmer I decided to implement a bit array in Ada and perform the same behaviors on the bit array.

The Ada program is slightly longer, sacrificing compactness for clarity.

The Ada programming language measures the size of data types in units of bits while the C program must convert bytes to bits in determining its size.

Compare the corresponding Ada and C code sections:

C:

Ada:

The Ada code defines an integral type named bit with 0 and 1 as the only valid values. The Ada code then defines an index range to be used in the array type. Knowing the index range prevents buffer overflows in other parts of the program.

The array type bit_array is declared. C provides no means to define array types, only array instances.