ADVANCED NOTES ON C LANGUAGE
(FOR ADVANCED SYSTEM PROGRAMMERS/INSTRUCTORS/TRAINERS)
Based on ANSI C-ISO/IEC 9899:1999
Authored By:
Peter Chacko
Netdiox Computing Systems
info.nediox@gmail.com
www.netdiox.com
Keywords/index terms: C interview questions FAQ, C Programming, volatile, pointers, GNU C ,
compiler, kernel,training, Linux..
T A B L E O F C O N T E N T S
1. Introduction - Why this notes?..............................................3
2. Data types, conversions, expressions…………………………….3
2.1 Conversion /Data transformation……………………………3
3. Pointers……………………………………………………………………….5
4. Strings in C…………………………………………………………………..7
5. Composite data type……………………………………………………..9
5.1 Structures and Unions………………………………………..11
6. Hardware cache Lines and C Structures…………………………12
7. Implementation of the function call and recursion…………..12
8. Volatile Variables in C……………………………………………………14
9. Appendix
9.1 GNU C compiler extensions………………………………………14
9.2 Inline assembly……………………………………………………….18
9.3 Illegal C interview questions……………………………………. 20
10. About the author…………………………………………………………21
ii
1 . I N T R O D U C T I O N
Why this notes?
It is generally observed that few people know how much they
don’t know in C, that cause them waste enormous amount of
time debugging their C code for fixing core dumps, data
corruption, abnormal program behavior etc which they could
have avoided with better coding in C, in the first phase. Then
people move to computer programming from different
background, with non-uniform skills in fundamental Computer
Science that is vital to be an efficient C Programmer. Lastly C
depends on the underlying processor architecture,
ABI(Application binary interface specification), Language tools,
Operating System to get its job done for a program on a given
hardware- which cannot be explained in a C book completely.
Hence people think that C is vast topic to master, which is not
quite right, once you have the complete background to be a C
Programmer. This notes try to throw some lights on these loose
ends, and motivates the reader to gather more information on
different domains , as it progresses.
Before you proceed always remember that C is a short language
to finish reading out , but a vast language to understand very
very deep and at the bottom you will see its interaction with other
computer science domains , like operating systems , compilers ,
CPU architecture etc. So always be patient and be liberal in terms
of how much computer science you want to know more, to be a
master in C as a system programmer.
This article doesn’t cover C language domains in its entirety, this
is just a complementary notes on existing books on C. You also
need to refer your C manual for the specific
features/implementation of the C standard. Thats not covered
here to avoid duplication. Use this document as a directive to help
you understand your weak areas in C.
Please understand that this article is not meant for those who
don’t have a system programming background.
2 . D A T A T Y P E , C O N V E R S I O N S A N D E X P R E S S I O N S
A data type of an object indicates the number of bytes it takes to store
it , as you know. If it is signed type, the most significant bit of the most
significant byte is used to store that information. But when it comes to
arrays and functions and pointers, data type plays some role.
iii
Consider the following array,
int ar[10];
Here type of array is array of 10 integers.
How do we represent that information.? Let me give you more
background to motivate you to understand this concept further.
Consider the following declaration,
int a ;
This is an integer declaration, based on the grammar
datatype identifier;
When we declare an array of 10 integers, we follow the same rule
identifier
int ar[10];
Data type
But instead of int[10] arr; we did int arr[10], that is how a data type
construct and it’s identifier bound together in a declaration of an
array( or functions) .That is why it is called “derived type”.
Now consider an example of type defining an array of 10 integers.
We do
typedef int ARR[10];
and
ARR ar;
Nowar means an array of 10 integers.
Here we only followed that special rule of derived type to typedef. Many
experienced C programmers find it confused when they have to typedef
an array type, as they lack the above mentioned understanding of how
arrays are “typified”.
iv
Similarly function declarations and pointer declarations. ( identifier
comes in between that different parts of data type. )
Hence we have 2 types of declaration
1st is of the form
data-type identifier;
eg: int a;
2
nd
type is
datatype-part-I identifier dataype-part-ii
eg:
identifier
int ar[10];
dataype-part-ii
datatype-part-I
identifier
int fun(int a,int b);
dataype-part-ii
datatype-part-I
3
rd
type is of the form,
data-type-part-I data-type-part-ii identifier;
identifier
eg: int * a;
dataype-part-ii
datatype-part-I
v
You can use this information when you work with complex
declarations in your projects. Please think over these topics and
understand the concepts well.
2 . 1 C O N V E R S I O N / D A T A T R A N S F O R M A T I O N
Again we restrict our discussion to basic data types. Reader is
advised to refer the C standard to see the details of other specific
data types for his/her specific tasks.
Consider the following,
char A=128;
int B=A;
what is the value stored in ‘B’?
These are the kind of issues that will cause many obscene
bug in the form of data corruption.
Here when we store 128 to a char variable, we set the sign
bit as 128 , means 10000000 in binary. Now when this is stored in
to an integer variable, what is stored in B now is
11111111 11111111 11111111 10000000
as char->int assignment involves an implicit promotion with
sign-bit preservation.
So when you print B as signed quantity you will get -128.
you now know what is happening.
Now another example
unsigned int A=2;
int B=-1;
if (A>B)
{……}
else
vi
Higher memory
address
{ ….}
In this case the expression (A>B)would evaluate false as B
gets converted to an unsigned integer which is now
11111111 11111111 11111111
11111111,
(which will be read as -1 without conversion. As negative numbers are
stored as 2’s complement method)
This conversion is caused by mixing signed and unsigned in an
expression.
But most compilers doesnt promote those operands which are part of a
sibling sub expression of it’s parent expression having bigger data types
invoked. ANSI is silent on this topic.
The following operators are special.
&&, ||, ,, ? : etc
All operators are evaluated left to right and there is a sequence point
(ie, side effect- safe) after the evaluation of the first operand. ( This
means that ( i++ && i) cause i++ be fully evaluated, applied and then
the second operand is used)
3 . P O I N T E R S
A beginner or experienced, sometimes or other finds that pointer
handling in C could be easier. One reason is that many miss the underlying
memory architecture. Consider a typical implementation, as shown by the
following logical diagram.
STACK
HEAP
BSS
DATA
.rodata
TEXT
vii
When a program is built(compilation, assembler invocation, and linking)
it is stored in a file with a special format. One such format is
ELF(Executable and Linking Format)in Unix like systems. It has the
necessary information ( in the ELF header and program header) to help
the program loader to load and create the process image in memory at
runtime, and also do run time linking (for shared libraries linked
dynamically).
At the time program is built, storage for all uninitialised data (static or
global) are not allocated, but only the total size is noted in program
header table of the ELF file. Later when program is being loaded, loader
read the metadata from the ELF file and allocate one large chunk (page
aligned) of memory to hose all such variables. This section is called
BSS(Block Static Storage or Block started by symbol).
Data( storage for all initialized static and global variables) and STACK
section should be familiar to you. HEAP is created by dynamic invocation
by memory-alloc routines of your program. In addition to this there will
be shared memory mappings as well. If you process a file using mmap()
system call interface, those regions are also mapped to your program.
Now the sum total of all such memory is called the address space of your
program. You can only manipulate these sections of memory and some
which are not to be written in to. Hence all sorts of pointer problems are
due to the following primary reasons.
1. You manipulate memory that is not in your program’s address space.
2. You manipulate memory in your address space wrongly(like, writing to a
write-protected region etc) or corrupting other variables.
For instance, Consider the following declaration ,
int ar[4]; int b=100;
ar[4]=200;
Will corrupt the variable b, because you exceeded the array
bound.[manipulated memory in your region wrongly..
int * a;
*a=100;
viii
will cause 100 be written to a memory address 0’. You have
manipulated a memory region outside your address space. Most
architecture doesn’t support de-returning 0 address. So accessing
a Null pointer is considered as accessing a region outside the
address space of any program
Now what’s an address? Is that a virtual address, physical address
or logical address?
The value you see when you apply an & operator to a C variable
is actually the program-relative logical address. It has to be
processed by segmentation unit(coupled with the value in the CS
register) to create a linear address and then processed by the
paging unit to create the physical address in your RAM. So these
details are not at the control of C, but is at the control of the
memory management subsystems of the operating system. But a
knowledge in these details help you debug your pointer issues
faster, as some one who knows his game. Pointers to pointers
are well understood as another layer of this indirection. The
compelling reason why we have pointers to pointers primarily is
to modify a pointer itself across a function call(by passing the
address of the pointer). The function in question need to have a
pointer to pointer as the formal parameter.( if you are doctor,
you need another doctor-to-doctor to cure you). Other wise same
concepts apply.
4 . S T R I N G S I N C
Strings are the main object a C programmer work with, which is a
sequence of bytes. String literals are null terminated. Its generally
accepted that all library functions that process strings expect the
strings as null terminated. Consider the following,
char ch =A’;
char *ch1 =A”;
char *ch2 =;
char *ch3 = 0;
ch is a character variable, not a string. ch1 points to a location
where the stringA” is stored.( a byte sequence where ascii value
of A and then a null character). ch2 points to null string, but has a
valid memory address as it’s value. ch3 is a NULL pointer.
Typically strings in C are stored in a write-protected section called
.rodata, just above the text segment. It can be changed with the
ix
compiler option, -writable-strings in some compilers. If its stored
in .rodata, you get a segmentation fault when you try to modify
these region, through the pointer. Always understand that any
string” is an expression, that has the value of the address where
this string is stored in the process image.
5 . C O M P O S I T E D A T A T Y P E S
Some details about the implementation
of arrays in C
Arrays to C programmer is what a surgery to surgeon. If you
know C arrays well, you know C. Lets invest some time on that
here.
Consider the declaration
int ar[4];
int * p=malloc(4*sizeof(int));
the difference in the above 2 uses are.
1. ar’ is not a variable, no memory is allocated to store ‘ar’ itself.
(some say, it is a constant pointer, wrongly). ‘ar’ is resolved at
program translation time. Total memory allocated is only 4 words.
In the second case ‘p it self is allocated memory, which is a
variable. Total memory is 4+1 words(assuming a pointer takes
only 4 bytes).
2. when you apply the ‘&’ operator to ar, it gives the same value
as ar itself. When you apply ‘&’ operator to p it gives the address
of p. Reason is obvious, ‘ar has no address because its not
program symbol seen by the run time and has no address.
The second difference in semantics of arrays name brings
another point. How can operator has no meaning to an array
name? think for few minutes….
When you apply & operator to a structure or union you set the
address of the first by the of the first member(or the shared
object in case of union) and when you just refer the variable
name of the structure , you mean the entire object. All other
variables have the same semantics. But when you just mention
x
the array by name, you don’t refer it as the entire array, rather an
address of the first element. This means that in C,
You cannot pass or return an entire array to a function or from it.
C Creator has decided this special meaning of array name
because, all array members of same type, All array members are
stored consecutively. ( no padding in between), and no reason to
treat any member special.
Now consider the array, int ar[2]; which is,
[0] [1]
Now as you can see that each element of the array itself is an
array of 4 bytes. Hence we can say that this is an array of , array
of 4 bytes. Which is nothing but
char ar [2][4];
Hence a multidimensional array is just the perspective exported
by the C compiler . It’s informal representation is just a sequence
of bytes. when you interpret it as a super object, you get a multi-
dimesional array, when you interpret it as super-sub-object, you
get a single array. The matrix form is only for your understanding.
Another way to describe this is with the statement that C is a row
major language. As C is a row major language it has some
difference in the following code
int arr[100][100];
1. for(int i=0;i<100; i++)
for(int j=0;j<100;j++)
ar[i][j]=i*j;
2. for (int j=0;j<100;j++)
for(int i=0; i<100; i++)
xi
there is no such operation meaningful to entire
array”.
ar[i][j]=i*j;
Here the first version is faster as we access row by row , where as
in the second case we access it by column by column. In the first
case, hardware cache line stall only at every CACHE-LINE size of
bytes, whereas in the second case hardware cache line stall
happens on every memory access. This is one example to show
that you need to know the underlying hardware architecture , to
be an effective C programmer. Mastering C is not just learning the
C language.
We can see that arrays and pointers are similar. We have pointers
to array as well.
int (*ar)[10];
Declare a poiner to an array, each element is one large
array itself. You can store to it like this
int b [2][10];
ar=b;
Or
ar=malloc(2* sizeof(int (*)[10]); // But all pointers are of
same size anyway//
*ar=b[0]; (or b[1]);
now if you do **ar, it means the first integer stored in the element
array. When you do ++a, you are jumping off by 1 array itself.
See you have never declared a pointer to a pointer, but you de-
referenced it as a pointer-to pointer(using **ar) , this is the
beauty(or wildness?) of pointer to an array.
Many think that the following assignment is valid given the
declarations
int ** p;
int ar[2][4];
p=ar;
xii
Here compiler throws error. She is right. You are actually storing
an object of type pointer to an array of 4 integers, to a plain
pointer to pointer to an integer. You should declare a pointer to
an array of 4 integers and then store it. You should do this in the
function argument declarations as well.
5 . 1 S T R U C T U R E S / U N I O N S
structure tag, member name, structure name all fall into different
name spaces. That is you can have
struct NewStruct{
int NewStruct;
}NewStruct
without a compilation error.
As memory reads happens at the boundaries of every
word(typical 4 bytes for a 32 bit system, 8 bytes or 64 bits
system), if any multi-byte element crosses this boundary CPU has
to issue more than 1 read/write cycle to access the memory,
causing performance issues. Hence compiler add padding bytes
to cause members aligned on these boundaries compiler never
pads in the beginning. (that is address of the structure and the
address of the first element of the structure is always same. )
All bit fields representation is implementation dependant.
6 . H A R D W A R E C A C H E L I N E S A N D C S T R U C T U R E S
When you declare variables in a structure, always declare
related variables ( meaning, the variables you always access
together) as close as possible. That way, one variable access
always cause other variables also found in the underlying cache
line. When ever you declare large object, always align that on a
cache line boundary(typically 32 bytes or 64 bytes depends on
the architecture). If you have an important member variable in a
structure, which is split across multiple cache lines, then your
program will show lower performance. Important kernel data
structures in the linux kernel carefully structure its member
variables to exploit this mechanism. All frequently accessed
variables in a structure should always be moved to the front of
the structure storage space.
xiii
7 . I M P L E M E N T A T I O N O F A F U N C T I O N C A L L & R E C U S R I O N
CPU has the registers EBP, ESP ,EIP and the STACK segment at
its disposal to implement control transfer.
When a program is executing, it has to be in some function. And
that function’s frame pointer is stored in EBP and its stack
pointer is stored in ESP. And the next instruction to be executed is
stored in EIP. When we move to the new function, that new
function also has to use these registers for the same purpose.
Hence we need a way to save & restore these values in between
the calls.
First, all arguments to the call are passed to the stack(or registers
based on the implementation). C is a right pusher. (It pushes right
most argument first(after it’s evaluation). Then the return address
is saved on the stack, followed by the current frame pointer .
Now what is left to be saved is the current ESP. fortunately
current ESP is the new EBP. Hence ESP is just saved in the new
EBP after the call. Hence CPU execute a instruction and we are in
the new function. Old ESP becomes the new EBP. Old ESP
becomes the new EBP. Old ESP will be modified by the new ESP
values(We already save the content in EBP). EIP will be used by
the instruction streams of the new function and we do our job in
the new function. When the function is done, EBP is stored back
to ESP. Then previous EBP is popped. Then what is left in the
stack frame is saved EIP which is also popped to EIP. And the CPU
go to the saved instruction, which is one right after the function
call.
Recursion is the property by which a function can call by
itself. The only difference between this iterative invocation is that
Stack frames are not removed until the function start doing the
stack unwinding. Assume that a function that prints a string in
reverse order. Here is the function
void strReverse(char* st)
{ if(!st) return;
if(!*st) return;
strReverse(st + 1 );
printf(“%c”, *str);
xiv
And assume also that we call this function with the stringABC”
Then the stack frame will be ( address decrease as Go up in most
of the architecures)
First call Second call Third call Fourth call
xv
‘\0
EIP
EBP
“C
EIP
EBP
BC
EIP
EBP
“ABC”
EIP
EBP
BC
EIP
EBP
“ABC
EIP
EBP
C”
EIP
EBP
BC
EIP
EBP
ABC”
EIP
EBP
ABC
EIP
EBP
As shown on the fourth call function gets a null character and
start unwinding, each stack frame see a different argument as
shown. Hence the printf function prints the character in reverse.
The important point to understand is that, all code after the
recursive call is executed only after the stack –unwinding process
begins.
8 . V O L A T I L E V A R I A B L E S I N C
An advanced discussion in C will not be complete without
talking about volatile variables. Please refer the paper ,
“Synchronization/Locking in the Linux kernel on x86
architecture” ( published by the the same author, Peter Chacko)
to see a detailed coverage
9 . A P P E N D I C E S
9 . 1 G N U C C O M P I L E R E X T E N S I O N S
C language is a portable assembly, but when you use GNU
extensions it becomes unportable. But to make the optimum use
of the underlying hardware, you have to have a non-portable
code in the lower most layer of the abstraction.. And it will add
your strength as a kernel developer/System programmer if you
understand the important GNU extensions. Following are a partial
list of extensions. ( For full list, please refer a GNU C manual )
xvi
A. Like in ANSI C 99( it refers as a flexible array member of
size[1]), GNU C allows a variable length object by having a last
array element of size zero.
This is very useful network programming tool when you want to
send a message of variable size. Assume that you have a
function that fill the variable message data to a structure. The
following code does the trick utilizing this mechanism.
struct message {
unsigned int length;
char msg-buf[0];
};
struct message *NewMsg = (struct message *)
malloc (sizeof (struct message) + CurrentMsgLength);
NewMsg->length = CurrentMsgLength;
Here, you declare the message as a structure having a zero-
length array, and depends on the size of your current message
( denoted by CurrentMsgLength), you allocate enough memory at
the end of the structure. As a consequence of this special use,
you cannot have an array of this structure.
B. Case ranges
You can specify multiple, consecutive case values in a single case
label, like the following.
case 10 ... 20:
which is equivalent to 11 separate case labels.
C . Attributes of a function
These extensions are helpful in passing enough information the compiler, to
optimize the code better. This keyword is followed by an attribute specification
xvii
inside double parentheses. The following are some useful attributes (please refer
the GNU manual for a full list). no_return( indicates whether the function is not to
return) , pure(indicates side-effects-free function), always_inline( instruct the
compiler to in-line the function),no_inline, deprecated( will cause a warning if the
function is called), nonull( to cause a compiler error if non-null arguments are
used).
Eg: The following declaration,
: extern void *
your_strcpy(void *dest, const void *src, unsigned len)
__attribute__((nonnull (1, 2)));
Will cause a compiler error if you invoke the function with null pointers for the
arguments first and second. If nonull is used with no arguments, all arguments are
checked against NULL .
E. Attribute syntax for variables
The following are the important attributes for variables
Aligned( to specify the alignment requirements), packed( to
specify that only the minimum alignment is required. (1-bit for
bitfields, 1-bye for other objects). Which means that all are
packed together !!, transparent union(to specify that some
variables are actually a union in disguise and to avoid type
checking or optimizations based on that). For instance,
Here is a structure in which the field b is packed, so that it
immediately follows a without any padding bytes, using the
attribute syntax for variables,
struct tag
{
char a;
int b __attribute__ ((packed));
xviii
}
Now the above structure occupy only 5 bytes, not 8 !!
You can also apply these attributes to a structure or union type as
well, as if you use attribute syntax declarations for the all
variables.
F. Thread local storage :
To specify a thread specific storage for variables, use
_thread specifier, like the following example:
_thread int I; will cause the variable I hosted in a storage
that’s specific to the calling thread.
G. Some built-in functions for atomic operations :
You can use the family of the following functions , to implement
memory access operations atomically(for example as the basic
building block of a synchronization primitives. These family of
built-ins issue enough “lock” prefixed instructions to realize
atomic access to the shared memory objects. (Not yet stable
though in many architectures, today). ( Please refer a GNU
manual for details)
_sync_add_and_fetch (type *ptr, type value, ...) or
_sync_fetch_and_add() style
__sync_sub_and_fetch (type *ptr, type value, ...) or
__sync_fetch_and_sub() style
__sync_or_and_fetch (type *ptr, type value, ...) or
__sync_fetch_and_or() style
__sync_and_and_fetch (type *ptr, type value, ...) or
_sync_fetch_and_and() style
__sync_xor_and_fetch(type *ptr, type value,…) or
__sync_fetch_and_xor() style.
xix
H. Built in function to avoid the cache-miss latency.
void __builtin_prefetch (const void *addr, ...)
This function cause the memory object pointed to by the addr are in the cache line,
after the execution this function. 2 optional arguments can be passed. The first one
specify whether it is a read (value 0) or write(value 1). Second argument specify
how local it is (meaning how to replicate it in all hierarchies of the caches). 0 means
global( should be there in the outer most cache) and 3 being local( should be kept
only in the on-chip CPU cache) and 2 means in-between.
I. Miscellenous GC features used by the kernel developers:
likely(0, unlikely() macros are very much used by kernel code to pass hints to
the compiler for the branch prediction( unlikely(0) cause to avoid prefetching
the code that follows, to instruction cache.). You can refer kernel sources to see
examples. Inline functions are also heavily used by the kernel code. New
structure initializer syntax of c99 is another GCC extensions you can find in
kernel code in many files.
9 . 2 I N L I N E A S S E M B L Y
Invoking assembly code from C is pretty simple, once you
understand the basics.
asm( “nop \t\n”); will cause a “nop” assembly instruction
executed from your C function. You can also invoke multiple
assembly instructions , as multiple, string encoded instructions as
shown below. ( This implement exit system call library. )
asm("movl $1,%eax\t\n”
“xor %ebx,%ebx\t\n”
“int $0x80 \t\n”);
we are moving 1 to eax as the syscall number of exit is 1. Then
clearing the ebx(as we are not to return from the control
path), and then making the software interrupt 128 to trap into
the kernel.
You can also use your C expressions as part of the operands in
your assembly code, using extended inline assembly. When you
make some registers clobbered, you need to tell the GCC about
that..You use extended in-line assembly for this. It has the
following form,
xx
asm ( assembler template
: output operands constraints ( optional)
: input operands constraints (optional)
: list of clobbered registers ( optinal)
);
following section shows some examples of how to use these constraints.
( Please take a look at various Linux kernel sources/header files to find rich
examples in action).
A. example for specific register constrains
Now let's take a look at how to specify individual registers as constraints for the
operands. In the following example, the cpuid instruction takes the input in the
%eax register and gives output in four registers: %eax, %ebx, %ecx, %edx. CPUID
gets it’s input from “option”in the eax register, as cpuid expects it to. The a, b, c,
and d constraints are used to collect the results
int main() {
int var1,var2,var3,var4,option;
asm ("cpuid"
: "=a" (var1),
"=b" (var2),
"=c" (var3),
"=d" (var4)
: "a" (option));
}
from the registers EAX, EBX,ECX,EDX to the variables
var1,var2,var3,var4 respectively.
B. memory constraints
xxi
Typically atomic_inc() dec functions use this in the kernel as we want a memory-
to-memory operations in this case. Example follows.
Use of memory operand constraint
Consider the following atomic increment operation:
myLock;
asm __volatile__(
"lock; incl %0"
:"=m" (myLock)
:"m" (myLock));
Here the memory constraint is necessary to get the desired
behaviour as otherwise myLock wouldn’t be consistent.
C . Matching operand constraints :
Consider a typical libc implementation of read system call.
int read (int fd, void *dataBuf, size len)
{
long ret;
__asm__ __volatile__ ("int $0x80"
: "=a" (ret)
: "0" (SYS_READ), "b" ((long) fd),
"c" ((long) dataBuf), "d" ((long)len):
"bx");
Above code specify that ret value of the invocation to be
stored in EAX, first input is on the matching 0
th
output
register(EAX itself)(we assume that SYS_READ contains the value
of the syscall number of the read system call), and the other
input values are stored in the EBX, ECX, and EDX registers. And
gcc is also instructed to note that ebx will be clobbered. it
invokes interrupt 128 and go to kernel mode with all input
values, and output the result into EAX.
xxii
9 . 3 S E M I - I L L E G E L C I N T E R V I E W Q U E S T I O N S / C F A Q
This is not an interview note. But it is our interest that right
candidates are picked during the interview process. Many
candidates prepare from C FAQs’ or ‘C for tests’ like notes. The
right candidate for the job may score less as his knowledge
comes from his C experience, not the incorrect C questions that
interviewers see in the net . Here is just a partial list.
1. Can we call main function from main itself?
How many times you do you need to do this? The behavior
depends on how the “hosted behavior is defined for the
specific implementation. C is silent here. And also if he does
know what is the specific behavior , that is not a great skill.
Invocation of the main is done , normally by the program
interpreter(stored in the .interp section of the ELF binary
file). The initial stack layout/main program arguments are
laid out at this time by the runtime linker/loader tools which
all are implementation dependent. Hence when you
successfully see that you can call main recursively, it
doesn’t mean that its universally implemented.
2. int i=1;
printf(“%d%d”,i++,++i);
what is the output?
This depends on the candidate’s knowledge that
a) C -compiler is a right pusher. b) there is sequence point
after every argument evaluation. And that how arguments
are pushed to the stack.
This is a good question to compiler writers , not to a C
programmer, as a conservative C programmer never use
side effect operators like this.
3. Can a function return more than one value?
This is stupid interview question called by many in-
experienced C programmers . A function only return
one value. Even though you return C pointers that
xxiii
points to a location containing multiple values, what
you return is a single value, pointer.
4. Never ask any C question whose behavior depends on the
underlying operating system, libraries, compilers or the
underlying architecture for a specific answer.
5. If I do malloc(1) , does that mean malloc only return 1
bytes?
This is another semi-legal question floated in the
net(and dutifully followed by many great
interviewers)
It is true in most OS that malloc will make the
underlying system call (like brk() in UNIX), which
allocates in chunks of 4k multiple of pages, which will
be part of the requesting program’s address space,
which they can access, without causing the core
dump. That doesn’t mean that we should re-interpret
the meaning of malloc. Its just the way OS allocation
does allocation which can be modifies, without
changing C, crossing beyond the 1 byte, is semi-
begun. malloc libraries can be re-implemented at any
time and this is not at the control of C language.
6. Some companies ask questions related to the specifics of
library functions, which is not uniformly implemented.
Some mix questions related to threads, which are also not
yet standardized. When you evaluate a C programmer,
always check on his understanding of operators, pointers,
and his ability in implementing efficient data structures for
the right purpose. C language is for efficiency, and evaluate
C programmer for his efficient programming, rather than
some details which are not used in 99% of the projects.
Check his knowledge in mixed mode arithmetic/operator
conversions and his skills in writing great C software having
impeccable algorithmic content.
xxiv
A B O U T T H E A U T H O R
Peter Chacko has been working on
system/networking/storage systems development since
1994. His career spans Hard-real time/embedded systems,
OS kernels like Unix and Linux, distributed storage and
networking systems, virtualization, data security and
related computer science disciplines. He had been
working for companies like Bellcore (Bell labs spin-off), HP,
IBM, US West communications and a couple of startups as a
consultant in the United states for 6+ years. He holds a
Masters degree in Computer science &applications(MCA),
Bachelors in Physics. He is currently the founder & CTO of
Bangalore-based cloudStorage startup(Sciendix data
systems pvt.ltd). He also run NetDiox computing systems
as a not-for-profit research center-cum-educational center
on system and networking software alongside. Prior to this
he was working for calsoftlabs heading all the R&D
activities of storage networking business unit. His linked-in
profile is at http://www.linkedin.com/pub/peter-
chacko/b/608/608
xxv