# Session 2 Summary

C++ Programming Course, Summer Term 2018, 20. April 2018

Code discussed in this session

# Of Types, States and Concepts

The C-style representation of strings is notoriously error-prone and causes security incidents on a daily basis. The fundamental problem with C-strings is their lack of a well-defined type specification.

Briefly, yet formally put: when a function receives a value of C-string type, it cannot rely on any semantics that restrict the value to a valid state.

If this sounds confusing, consider this example:

char * stringish; // <- used as string type by convention (not by definition!)

The variable stringish could denote a pointer to a single character or the start address of a dynamically allocated memory range.

Functions like strlen in the C standard library expect a string by convention. That is to say: there is an informal aggreement between the implementer of the function and the programmer calling it that only strings will be passed to this function. Should the programmer pass a pointer to a single character, neither the compiler nor the runtime can prevent the execution.

This code would compile flawlessly and run perfectly fine until the unevitable segmentation / access violation fault:

char * not_a_string    = (char *)(123);        // Oh boy, the C-style cast ...
int i_will_never_exist = strlen(not_a_string); // *sigh*

So at this point, stringish has been declared but not defined yet, as no value has been assigned. A valid state of a C-string variable must contain an address of the start address of an memory range allocated for a string:

stringish = (char *)malloc(nchars);

Now, the C-string referenced by stringish is allocated, but not initialized. Passing it to strlen would still fail, as a valid state of a C-string is null-terminated: a string of length n must occupy a memory range of n+1 characters as a '\0' must be assigned after the last character in the string to denote the end of the string (see man 3 strlen).

stringish[nchars-1] = '\0';

Only now is stringish a valid state of a C-string. It still contains nonsensical values no character sequence has been written into allocated memory, but it can be safely used in any operation expecting a string.

Exactly this is the definition of a valid state.

The state of an object is valid if all operations defined for its type can be
executed on the object without breaking the operation semantics.

In conclusion, there are three steps to achieve a valid state of an object:

1. Declaration
2. Allocation
3. Definition (implies Initialization)

In C, these steps are not coupled and it is in the programmer’s the responsibility to specify them correctly. Compilers got better at detecting “code smells”, but compiler warnings - if created on a Turing machine - will never help in non-trivial cases for well-known reasons.

# The Tao of Concepts

A practicable solution to this is the definition of types as a unseparable coupling of operations and states. In C++, a category of types that satisfy common semantics is called concept, an essential term with specific connotation in C++. A concept definition consists of:

Valid Expressions
A set of syntactically defined interactions that have common behaviour for any type of this concept, for example (a+b) for numerals
Semantics
Pre-conditions, invariants and post-conditions of these expressions with respect to objects (= states) of the type
Dependent Types
Types provided by a base type that are visible to the user; A list container, for example, must provide a dependent type like element_type as the programmer needs a way to determine the type of values returned by operations like head(list)

Note that there is no mention of “interface” (in the Java sense), “methods” or “class members”. Concepts are not about function signatures as you know them from API documentations. Signatures are a technical necessity, not a formal description.

When reasoning about C++ concepts, you should tune your mind set to the fundamentals of category theory. This sounds like obscenely acamedic formalese, but have a second look at the first paragraph in this section.

For example, the concrete types list and array are equivalent in several aspects but also exhibit peculiar differences.

## Semantics

Originating from linguistics, semantics is the study of meaning. Formal means to represent semantics is a necessity as intuitive reasoning is obviously biased and may drastically differ between individuals.

In the context of this course, and actually computer science in general, formal semantics describe the effect of operations and their interoperability and interdependency with respect to the scope of their invocation.

We usually define semantics of an operation by specifiying:

• Preconditions
• Invariants
• Postconditions

## Concept

Similar the Ideon in Platon’s terminology, a concept (lat. conceptum - sth. perceived) is an abstract idea representing fundamental characteristics.

To illustrate Platon’s definition of an abstract object, let’s assume two strangers, maybe foreigners, talk about dogs. The concrete visual rendition in the mental model of both will diverge. But they there is no need to negotiate definitions to have a conversation about dogs because the characteristics associated with the general concept “dog” do not conflict.

Similarily, concepts are used in C++ to define an ontology of types. Specifying algorithm variants for every single type is impossible but the the unlimited set of types can be reduced to a small set of common characteristics.

In the STL, a prominent concept is the Iterator.

The iterator concept (and its subcategories) serves as a minimal abstraction to communicate access to data elements between algorithms and containers.

# A String Class in C

Enough with the formal boilerplate: what does coupling states and operations in type definitions actually mean in code?

Considering the string example above, we learned that the three steps towards a valid state of a string should be combined in a single operation.

In C++, such “unified” operations are constructors. They make it already syntactically impossible to create an object with an invalid state.

We can’t achieve this ideal in C because its type system is too weak, so we still have to resort to convention. But we can specify types that satisfy robust concepts.

In the listing below, the type String is defined as

• A set of properties that specify the string’s state, represented as a struct; in the object-oriented sense, an object of this struct type are string instances

• Functions that accept instances of String and modify its state, better known as methods in the object-oriented model

So, what makes a function a method? The term method refers to the following crucial criterion:

A method is a function that implements an operation on an object of a specific
type that transfers the object from one valid state to another.

This definition describes the following conditions:

Pre-Condition
Before invoking the method, the object is in a valid state
Post-Condition
After executing the method’s operation, the object is in a valid state

This also implies the following:

Invariant
If an object is in a valid state, invoking any method defined for the object’s type results in a semantically correct execution of the method.

Now, finally, let’s see how the Rule of Three for a String class could be implemented in C:

#include <string.h>
#include <stdlib.h>

typedef struct {
char  * data;
char    size;
} String;

String * string_new() {
String * target = (String *)(malloc(sizeof(String));
target->data = NULL;
target->size = 0;
return target;
}

String * string_assign(String * lhs, String * rhs) {
if (lhs->size != rhs->size) {
lhs->size = rhs->size;
lhs->data = realloc(lhs->data, lhs->size);
}
strcpy(target->data, str);
return lhs;
}

String * string_copy(const String * other) {
// see why conceptually sound methods are useful?
String * target = string_new();
string_assign(target, other);
return target;
}

void     string_delete(String * str) {
free(str->data);
free(str);
}

# Type Qualifiers, References vs. Pointers

This documented code listing contains the findings discussed in the lab session. This section gives a brief summary.

Anatomy of an assignment by example:

  //     left-hand side       right-hand side
//            |                    |
//  .---------'---------.     .----'---.
const Type * lvalue   =   & rvalue
//   |     |    | |            | |
//   |     |    | '- named     | '- named or unnamed
//   |     |    |              |
//   |     |    '- type        '- operator
//   |     |
//   |     '- referenced type
//   |
//   '- cvr

CVR is short for the qualifier category “const, volatile, restrict”.

  int value = 100;

This is the definition of variable p_value with its declaration on the left hand side.
On the left hand side or in a declaration, * refers to a type:

  int * p_value;

On the right hand side, & is an operator:

  p_value = &value;

On the left hand side of an assignment * is the dereference operator:

  *p_value = 222;

You actually can’t mix this up (more important, neither can the compiler) as * is not a complete type:

  int * inst = NULL; // declaration/definition, int * is complete type
* inst = NULL; // assignment of value referenced by pointer

Pointer values have no restriction whatsoever. Therefore, they cannot be trusted:

  p_value = NULL;
p_value = (int*)(123); // <-- may I introduce you to the
//     horrible, horrible C-style cast?

On the right hand size, * is an operator …

  int  deref   = *p_value;

… except when used within type specifiers like casts, of course:

  long deref_l = *((long *)p_value);
//             ^       ^
//             |       |
//             |       '- type specifier, casting to "pointer-to-long*
//             '- dereference operator

On the left hand side, & denotes a reference type. References must be initialized (= defined) with their declaration. Think of references as ‘variable aliases’. They are always named and can only refer to another named value.

  int & ref = my_variable;

For a mental model, you can consider references as quantum-entangled.

  ref = 123;
assert(value == 123);

int & one = 1; // <-- nope, 1 is not named
one = 2;       // <-- you see the problem?

Variables ref and value are identical. In C/C++, two variables are identical if they represent the same address:

  assert(&ref == &value);

Equality refers to the values of variables. Obviously, identity implies equality.

  int a, b;
a = 12;
b = 12; // a equals b

In C++ lingo, named values are usually called “lvalues”, because they can be used at the left hand side of an assignment. Unnamed values are called “rvalues” because they can only be used at the right hand side of an assignment.

For example, 123 is an rvalue because a statement like:

123 = variable;

… would be nonsensical.