Do you know that? $44-$49 average hourly rates for freelancers with less than five years of experience next

See right interview answers on 30 common job interview questions


Collapse | Expand

1. Are there any problems with performing mathematical operations on different variable types?

C has three categories of built-in data types: pointer types, integral types, and floating-point types.
Pointer types are the most restrictive in terms of the operations that can be performed on them. They are
limited to

- subtraction of two pointers, valid only when both pointers point to elements in the same array. The result
is the same as subtracting the integer subscripts corresponding to the two pointers.

+ addition of a pointer and an integral type. The result is a pointer that points to the element which would
be selected by that integer.

Floating-point types consist of the built-in types float, double, and long double. Integral types consist of
char, unsigned  char,  short, unsigned short,  int,  unsigned int,  long, and  unsigned long. All of these types
can have the following arithmetic operations performed on them:

+ Addition

- Subtraction

* Multiplication

/ Division

Integral types also can have those four operations performed on them, as well as the following operations:
% Modulo or remainder of division

<< Shift left

>> Shift right

& Bitwise AND operation

| Bitwise OR operation

^ Bitwise exclusive OR operation

! Logical negative operation

~ Bitwise "one's complement" operation

Although C permits "mixed mode" expressions (an arithmetic expression involving different types), it
actually converts the types to be the same type before performing the operations (except for the case of pointer
arithmetic described previously). The process of automatic type conversion is called "operator promotion."

2. Can a variable be both const and volatile?

Yes. The  const modifier means that this code cannot change the value of the variable, but that does not mean that the value cannot be changed by means outside this code. For instance, the timer structure was accessed through a volatile const pointer. The function itself did not change the value of the timer, so it was declared const. However, the value was changed by hardware on the computer, so it
was declared volatile. If a variable is both const and volatile, the two modifiers can appear in either order.

3. Can static variables be declared in a header file?

You can't declare a static variable without defining it as well (this is because the storage class modifiers static and extern are mutually exclusive). A static variable can be defined in a header file, but this would cause each source file that included the header file to have its own private copy of the variable, which is probably not what was intended.

4. Do variables need to be initialized?

No. All variables should be given a value before they are used, and a good compiler will help you find variables
that are used before they are set to a value. Variables need not be initialized, however. Variables defined
outside a function or defined inside a function with the static keyword are already initialized to 0 for you if you do not explicitly initialize them.

Automatic variables are variables defined inside a function or block of code without the static keyword.
These variables have undefined values if you don't explicitly initialize them. If you don't initialize an
automatic variable, you must make sure you assign to it before using the value.

Space on the heap allocated by calling malloc() contains undefined data as well and must be set to a known
value before being used. Space allocated by calling calloc() is set to 0 for you when it is allocated.

5. How can you determine the maximum value that a numeric variable can hold?

The easiest way to find out how large or small a number that a particular type can hold is to use the values
defined in the ANSI standard header file limits.h. This file contains many useful constants defining the values
that can be held by various types, including these:

 Value Description
 CHAR_BIT-Number of bits in a char
 CHAR_MAX-Maximum decimal integer value of a char
 CHAR_MIN-Minimum decimal integer value of a char
 MB_LEN_MAX-Maximum number of bytes in a multibyte character
 INT_MAX-Maximum decimal value of an int
 INT_MIN-Minimum decimal value of an int
 LONG_MAX-Maximum decimal value of a long
 LONG_MIN-Minimum decimal value of a long
 SCHAR_MAX-Maximum decimal integer value of a signed char
 SCHAR_MIN-Minimum decimal integer value of a signed char
 SHRT_MAX-Maximum decimal value of a short
 SHRT_MIN-Minimum decimal value of a short
 UCHAR_MAX-Maximum decimal integer value of unsigned char
 UINT_MAX-Maximum decimal value of an unsigned integer
 ULONG_MAX-Maximum decimal value of an unsigned long int
 USHRT_MAX-Maximum decimal value of an unsigned short int

For integral types, on a machine that uses two's complement arithmetic (which is just about any machine
you're likely to use), a signed type can hold numbers from -2(number of bits - 1) to +2(number of bits - 1)
 - 1. 

An unsigned type can hold values from 0 to +2(number of bits)- 1. For instance, a 16-bit signed integer can hold numbers from -215(-32768) to +215 - 1 (32767).

6. How reliable are floating-point comparisons?

Floating-point numbers are the "black art" of computer programming. One reason why this is so is that there
is no optimal way to represent an arbitrary number. The Institute of Electrical and Electronic Engineers
(IEEE) has developed a standard for the representation of floating-point numbers, but you cannot guarantee
that every machine you use will conform to the standard.

Even if your machine does conform to the standard, there are deeper issues. It can be shown mathematically
that there are an infinite number of "real" numbers between any two numbers. For the computer to
distinguish between two numbers, the bits that represent them must differ. To represent an infinite number
of different bit patterns would take an infinite number of bits. Because the computer must represent a large
range of numbers in a small number of bits (usually 32 to 64 bits), it has to make approximate representations
of most numbers.

Because floating-point numbers are so tricky to deal with, it's generally bad practice to compare a floating-
point number for equality with anything. Inequalities are much safer. If, for instance, you want to step
through a range of numbers in small increments, you might write this:

#include <stdio.h>
const float first = 0.0;
const float last = 70.0;
const float small = 0.007;
main()
{
        float   f;
        for (f = first; f != last && f < last + 1.0; f += small)
                ;
        printf("f is now %g\n", f);
}

However, rounding errors and small differences in the representation of the variable small might cause f to
never be equal to last (it might go from being just under it to being just over it). Thus, the loop would go
past the value last. The inequality f < last + 1.0 has been added to prevent the program from running
on for a very long time if this happens. If you run this program and the value printed for f is 71 or more,
this is what has happened.

A safer way to write this loop is to use the inequality f < last to test for the loop ending, as in this example:

float   f;
for (f = first; f < last; f += small)
        ;

You could even precompute the number of times the loop should be executed and use an integer to count
iterations of the loop, as in this example:

float   f;
int     count = (last - first) / small;
for (f = first; count-- > 0; f += small)

7. Is it acceptable to declare/define a variable in a C header?

A global variable that must be accessed from more than one file can and should be declared in a header file.
In addition, such a variable must be defined in one source file. Variables should not be defined in header files,
because the header file can be included in multiple source files, which would cause multiple definitions of
the variable.

 

The ANSI C standard will allow multiple external definitions, provided that there is only one
initialization. But because there's really no advantage to using this feature, it's probably best to avoid it and maintain a higher level of portability.

"Global" variables that do not have to be accessed from more than one file should be declared static and should not appear in a header file.

8. What is a const pointer?

The access modifier keyword const is a promise the programmer makes to the compiler that the value of a
variable will not be changed after it is initialized. The compiler will enforce that promise as best it can by not
enabling the programmer to write code which modifies a variable that has been declared const.

A "const pointer," or more correctly, a "pointer to const," is a pointer which points to data that is const
(constant, or unchanging). A pointer to const is declared by putting the word const at the beginning of the
pointer declaration. This declares a pointer which points to data that can't be modified. The pointer itself
can be modified. The following example illustrates some legal and illegal uses of a const pointer:

const char  *str = "hello";
char  c = *str    /* legal */
str++;            /* legal */
*str = 'a';       /* illegal */
str[1] = 'b';     /* illegal */

The first two statements here are legal because they do not modify the data that str points to. The next two
statements are illegal because they modify the data pointed to by str.

Pointers to const are most often used in declaring function parameters. For instance, a function that counted
the number of characters in a string would not need to change the contents of the string, and it might be written this way:

my_strlen(const char *str)
{
        int count = 0;
        while (*str++)
        {
            count++;
        }
        return count;
}

Note that non-const pointers are implicitly converted to const pointers when needed, but const pointers
are not converted to non-const pointers. This means that my_strlen() could be called with either a const
or a non-const character pointer.

9. What is operator promotion?

If an operation is specified with operands of two different types, they are converted to the smallest type that
can hold both values. The result has the same type as the two operands wind up having. To interpret the rules,
read the following table from the top down, and stop at the first rule that applies.

 If Either Operand Is And the Other Is Change Them To
 long double-any other type-long double
 double-any smaller type-double
 float-any smaller type-float
 unsigned long-any integral type-unsigned long
 long-unsigned > LONG_MAX-long
 long-any smaller type-long
 unsigned-any signed type-unsigned

The following example code illustrates some cases of operator promotion. The variable f1 is set to 3/4.
Because both 3 and 4 are integers, integer division is performed, and the result is the integer 0. The variable
f2 is set to 3/4.0. Because 4.0 is a float, the number 3 is converted to a float as well, and the result is
the float 0.75.

#include <stdio.h>
main()
{
    float f1 = 3 / 4;
    float f2 = 3 / 4.0;
    printf("3 / 4 == %g or %g depending on the type used.\n", f1, f2);
}

10. What is page thrashing?

Some operating systems (such as UNIX or Windows in enhanced mode) use virtual memory. Virtual
memory is a technique for making a machine behave as if it had more memory than it really has, by using
disk space to simulate RAM (random-access memory). In the 80386 and higher Intel CPU chips, and in most
other modern microprocessors (such as the Motorola 68030, Sparc, and Power PC), exists a piece of
hardware called the Memory Management Unit, or MMU.

The MMU treats memory as if it were composed of a series of "pages." A page of memory is a block of
contiguous bytes of a certain size, usually 4096 or 8192 bytes. The operating system sets up and maintains
a table for each running program called the Process Memory Map, or PMM. This is a table of all the pages
of memory that program can access and where each is really located.

Every time your program accesses any portion of memory, the address (called a "virtual address") is processed
by the MMU. The MMU looks in the PMM to find out where the memory is really located (called the
"physical address"). The physical address can be any location in memory or on disk that the operating system
has assigned for it. If the location the program wants to access is on disk, the page containing it must be read
from disk into memory, and the PMM must be updated to reflect this action (this is called a "page fault").
Hope you're still with me, because here's the tricky part. Because accessing the disk is so much slower than
accessing RAM, the operating system tries to keep as much of the virtual memory as possible in RAM. If
you're running a large enough program (or several small programs at once), there might not be enough RAM
to hold all the memory used by the programs, so some of it must be moved out of RAM and onto disk (this
action is called "paging out").

The operating system tries to guess which areas of memory aren't likely to be used for a while (usually based
on how the memory has been used in the past). If it guesses wrong, or if your programs are accessing lots of
memory in lots of places, many page faults will occur in order to read in the pages that were paged out. Because
all of RAM is being used, for each page read in to be accessed, another page must be paged out. This can lead
to more page faults, because now a different page of memory has been moved to disk. The problem of many
page faults occurring in a short time, called "page thrashing," can drastically cut the performance of a system.

Programs that frequently access many widely separated locations in memory are more likely to cause page
thrashing on a system. So is running many small programs that all continue to run even when you are not
actively using them. To reduce page thrashing, you can run fewer programs simultaneously. Or you can try
changing the way a large program works to maximize the capability of the operating system to guess which
pages won't be needed. You can achieve this effect by caching values or changing lookup algorithms in large
data structures, or sometimes by changing to a memory allocation library which provides an implementation
of malloc() that allocates memory more efficiently. Finally, you might consider adding more RAM to the
system to reduce the need to page out.

11. What is the benefit of using const for declaring constants?

The benefit of using the const keyword is that the compiler might be able to make optimizations based on
the knowledge that the value of the variable will not change. In addition, the compiler will try to ensure that
the values won't be changed inadvertently.

Of course, the same benefits apply to #defined constants. The reason to use const rather than #define to
define a constant is that a const variable can be of any type (such as a struct, which can't be represented by
a #defined constant). Also, because a const variable is a real variable, it has an address that can be used, if
needed, and it resides in only one place in memory.

12. What is the difference between declaring a variable and defining a variable?

Declaring a variable means describing its type to the compiler but not allocating any space for it. Defining
a variable means declaring it and also allocating space to hold the variable. You can also initialize a variable
at the time it is defined. Here is a declaration of a variable and a structure, and two variable definitions, one
with initialization:

extern int decl1;  /* this is a declaration */
struct decl2 
{
    int member;
}; /* this just declares the type--no variable mentioned */
int     def1 = 8;      /* this is a definition */
int     def2;          /* this is a definition */

To put it another way, a declaration says to the compiler, "Somewhere in my program will be a variable with
this name, and this is what type it is." A definition says, "Right here is this variable with this name and this
type."

A variable can be declared many times, but it must be defined exactly once. For this reason, definitions do
not belong in header files, where they might get #included into more than one place in your program.

13. When should a type cast be used?

There are two situations in which to use a type cast. The first use is to change the type of an operand to an arithmetic operation so that the operation will be performed properly. The variable f1 is set to the result of dividing the integer i by the integer
j. The result is 0, because integer division is used. The variable f2 is set to the result of dividing i by j as well. However, the (float) type cast causes i to be converted to a float. That in turn causes floating-point division to be used and gives the result 0.75.

#include <stdio.h>
main()
{
    int i = 3;
    int j = 4;
    float f1 = i / j;
    float f2 = (float) i / j;
    printf("3 / 4 == %g or %g depending on the type used.\n", f1, f2);
}

The second case is to cast pointer types to and from void * in order to interface with functions that expect
or return void pointers. For example, the following line type casts the return value of the call to malloc()
to be a pointer to a foo structure.

struct foo      *p = (struct foo *) malloc(sizeof(struct foo));

14. When should a type cast not be used?

A type cast should not be used to override a const or volatile declaration. Overriding these type modifiers
can cause the program to fail to run correctly.

A type cast should not be used to turn a pointer to one type of structure or data type into another. In the rare events in which this action is beneficial, using a union to hold the values makes the programmer's intentions clearer.

15. When should the const modifier be used?

There are several reasons to use const pointers. First, it allows the compiler to catch errors in which code
accidentally changes the value of a variable, as in

while (*str = 0) /* programmer meant to write *str != 0 */
{
    /* some code here */
    str++;
}

in which the = sign is a typographical error. Without the const in the declaration of str, the program would compile but not run properly.

Another reason is efficiency. The compiler might be able to make certain optimizations to the code generated if it knows that a variable will not be changed.

Any function parameter which points to data that is not modified by the function or by any function it calls should declare the pointer a pointer to const. Function parameters that are passed by value (rather than through a pointer) can be declared const if neither the function nor any function it calls modifies the data.

 

In practice, however, such parameters are usually declared const only if it might be more efficient for the
compiler to access the data through a pointer than by copying it.

16. When should the register modifier be used? Does it really help?

The register modifier hints to the compiler that the variable will be heavily used and should be kept in the
CPU's registers, if possible, so that it can be accessed faster. There are several restrictions on the use of the
register modifier.

First, the variable must be of a type that can be held in the CPU's register. This usually means a single value
of a size less than or equal to the size of an integer. Some machines have registers that can hold floating-point
numbers as well.

Second, because the variable might not be stored in memory, its address cannot be taken with the unary and
operator. An attempt to do so is flagged as an error by the compiler.
Some additional rules affect how useful the register modifier is. Because the number of registers is limited,
and because some registers can hold only certain types of data (such as pointers or floating-point numbers),
the number and types of register modifiers that will actually have any effect are dependent on what machine
the program will run on. Any additional register modifiers are silently ignored by the compiler.

Also, in some cases, it might actually be slower to keep a variable in a register because that register then
becomes unavailable for other purposes or because the variable isn't used enough to justify the overhead of
loading and storing it.

So when should the register modifier be used? The answer is never, with most modern compilers. Early C
compilers did not keep any variables in registers unless directed to do so, and the register modifier was a
valuable addition to the language. C compiler design has advanced to the point, however, where the compiler
will usually make better decisions than the programmer about which variables should be stored in registers.
In fact, many compilers actually ignore the register modifier, which is perfectly legal, because it is only a hint
and not a directive.

In the rare event that a program is too slow, and you know that the problem is due to a variable being stored
in memory, you might try adding the register modifier as a last resort, but don't be surprised if this action
doesn't change the speed of the program.

17. When should the volatile modifier be used?

The  volatile modifier is a directive to the compiler's optimizer that operations involving this variable should
not be optimized in certain ways. There are two special cases in which use of the volatile modifier is
desirable. The first case involves memory-mapped hardware (a device such as a graphics adaptor that appears
to the computer's hardware as if it were part of the computer's memory), and the second involves shared
memory (memory used by two or more programs running simultaneously).

Most computers have a set of registers that can be accessed faster than the computer's main memory. A good
compiler will perform a kind of optimization called "redundant load and store removal." The compiler looks
for places in the code where it can either remove an instruction to load data from memory because the value
is already in a register, or remove an instruction to store data to memory because the value can stay in a register
until it is changed again anyway.

If a variable is a pointer to something other than normal memory, such as memory-mapped ports on a
peripheral, redundant load and store optimizations might be detrimental. For instance, here's a piece of code
that might be used to time some operation:

time_t time_addition(volatile const struct timer *t, int a)
{
        int     n;
        int     x;
        time_t  then;
        x = 0;
        then = t->value;
        for (n = 0; n < 1000; n++)
        {
                x = x + a;
        }
       return t->value - then;
}

In this code, the variable t->value is actually a hardware counter that is being incremented as time passes.
The function adds the value of a to x 1000 times, and it returns the amount the timer was incremented by
while the 1000 additions were being performed.

Without the volatile modifier, a clever optimizer might assume that the value of t does not change during
the execution of the function, because there is no statement that explicitly changes it. In that case, there's
no need to read it from memory a second time and subtract it, because the answer will always be 0. The
compiler might therefore "optimize" the function by making it always return 0.

If a variable points to data in shared memory, you also don't want the compiler to perform redundant load
and store optimizations. Shared memory is normally used to enable two programs to communicate with each
other by having one program store data in the shared portion of memory and the other program read the
same portion of memory. If the compiler optimizes away a load or store of shared memory, communication
between the two programs will be affected.

18. Where in memory are my variables stored?

Variables can be stored in several places in memory, depending on their lifetime. Variables that are defined
outside any function (whether of global or file static scope), and variables that are defined inside a function
as static variables, exist for the lifetime of the program's execution. These variables are stored in the "data
segment." The data segment is a fixed-size area in memory set aside for these variables. The data segment is
subdivided into two parts, one for initialized variables and another for uninitialized variables.

Variables that are defined inside a function as auto variables (that are not defined with the keyword static)
come into existence when the program begins executing the block of code (delimited by curly braces {})
containing them, and they cease to exist when the program leaves that block of code. 

Variables that are the arguments to functions exist only during the call to that function. These variables are stored on the "stack".
The stack is an area of memory that starts out small and grows automatically up to some predefined limit.
In DOS and other systems without virtual memory, the limit is set either when the program is compiled or
when it begins executing. In UNIX and other systems with virtual memory, the limit is set by the system,
and it is usually so large that it can be ignored by the programmer.

The third and final area doesn't actually store variables but can be used to store data pointed to by variables.
Pointer variables that are assigned to the result of a call to the malloc() function contain the address of a
dynamically allocated area of memory. This memory is in an area called the "heap." The heap is another area
that starts out small and grows, but it grows only when the programmer explicitly calls malloc() or other
memory allocation functions, such as calloc(). The heap can share a memory segment with either the data
segment or the stack, or it can have its own segment. It all depends on the compiler options and operating
system. The heap, like the stack, has a limit on how much it can grow, and the same rules apply as to how
that limit is determined.