Lesson #27

Interaction between structures, arrays, pointers, and functions.


In this lesson I will cover the following topics

  1. How to declare and use an array of structures.

  2. How to pass structures to functions and return structures from functions.

  3. How to use pointers to structures with functions to avoid copying structures too much.


Arrays of structures

Structures are very useful, but it turns out that arrays of structures are also very useful. A structure is like a record in a database table. An array of structures is like the database table itself.

Suppose I wanted to write a program that analyzed the logs of an Internet Service Provider. Suppose each log entry contained information about the person who logged in, the date and time they logged in, the date and time they logged out, and how much network traffic they generated. I can express all of this information with structures very nicely.

// This structure holds a date and time.
struct datetime {
  int seconds, minutes, hours;
  int day, month, year;

// This structure holds one record of information from the log file.
struct log_entry {
  char username[16];       // Holds the username of the user.
  struct datetime login;   // When the user logged in.
  struct datetime logout;  // When the user logged out.
  long TX_bytes;           // Number of bytes sent over the network.
  long RX_bytes;           // Number of bytes received from the network.

In this example I'm assuming that user names are no more than 15 characters (I'm leaving one space in the array for the null byte). This is fairly realistic.

Reading the log file will be somewhat complicated. My program will have to open the file with fopen, read it a line at a time using fgets, and process each line. You might imagine that each line of the log format is something like (I'm showing a single long line broken into two shorter lines to prevent ugly word wrapping from happening).

li=23-Jul-1999:15:37:24, u=pchapin, lo=23-Jul-1999:15:45:28,
  tx=1134587, rx=89721

Here "li=" shows the log in time, "u=" shows the user name, and so forth. Extracting the necessary values from this string would require some work but hopefully you could imagine doing it. (Actually it is highly unlikely that a log file would show both login and logout times in one entry. More likely those two events would be in different log entries. Thus our program would really have to handle that and that makes it a good deal trickier).

Right now I'm not interested in how you interpret the text of the log entries. I'm only interested in how you would manage the structures after you got them filled up with data. Since there are many log entries in a typical log file you will probably want an array of structures.

struct log_entry log[1024];

This declares an array of 1024 (exactly 210... a nice, round number) struct log_entry variables. The name of the array is log. Now imagine that you have some function that actually reads the log file, interprets each entry and loads the array with data. Maybe someone else wrote it. Here is the declaration of that function.

int read_log(struct log_entry *buffer, int size);

This function expects a pointer to a struct log_entry which, as is traditional, is really to be the address of the first element of an array. The second parameter is the size of the array. The function uses the size to insure that it does not overrun the array. The function returns the number of log entries that it actually fills.

Armed with this function you can do

int log_count;

log_count = read_log(log, 1024);

The name log is the name of an array (of structures) without an index. As always, the compiler interprets this to be the address of the first element of the array. This is exactly what read_log expects. The read_log function then works its magic and returns the number of log entries it found. Maybe that's a number like 334.

So what are you going to do with this information? Suppose you wanted to print out all log entries where the user logged in during the middle of the night. Perhaps you are worried about late night hacking attempts. Here is what you might do.

int i;

for (i = 0; i < log_count; i++) {
  if (log[i].login.hours < 5) {
    printf("User %s logged in during early morning of %d-%d-%d\n",
      log[i].login.day, log[i].login.month, log[i].login.year

This loop runs over all the legitimate log entries (not necessarily all 1024 of them). For each log entry it examins the hours member of the login date and time. If that member is less than 5 (early morning) the if statement will trigger.

The expression log[i].login.hours works as desired. The member selection operator and the array access operator have the same precedence but they associate from left to right. Thus the expression is really ((log[i]).login).hours.

Inside the if statement I print out the relevant information. I print the username array of characters into a %s format specifier so that the string is printed out. I then print out the data part of log[i].login as three separate integers. The printf statement is quite long because some of the things I'm printing require elaborate expressions to properly specify. That is fine.

Pointers to structures

My loop above uses array indicies to access the log array. However, I can also use pointers. Just as I can take the address of a simple variable and get a pointer to that variable, I can also take the address of a structure and get a pointer to that structure. Here is how it might look

struct log_entry *p;

for (p = log; p < log + log_count; p++) {
  if ((*p).login.hours < 5) {

This loop starts by making p point at the first structure in the array. When it later increments p, the address in p will be advanced by an amount equal to the size of the entire structure (large). Thus after p++, p will be pointing at the next structure in the array. This is exactly how pointers to ordinary variables work. When I add the integer log_count to the pointer log, I get a new pointer log_count structure sizes downstream. Again this is exactly how pointers to ordinary variables work.

The expression in the if statement looks complicated


Here I start by considering the thing p is pointing at. Since p is a pointer to a structure, *p is a structure. In particular *p is a struct log_entry. Since such structures have a login member, the reference to it is completely legal and correct.

In this expression the parentheses are necessary. Otherwise I have *p.login.hours and because of the higher precedence of the member selection operator this looks like *((p.login).hours). However, this is an error. I can't apply the member selection operator to a pointer so p.login doesn't make sense. Pointers aren't structures and they don't have members (pointers are addresses). It is true that in this case p is a pointer to a structure, but that does not change the fact that p itself is just a plain old pointer.

Since having to type the parentheses around *p all the time in expressions like this is very tedious. C has a special operator just for this situation. It is sometimes called the "arrow operator". It looks like this


When you use the arrow operator, the left operand must be a pointer to a structure. The right operand must be one of the members in the structure pointed at by the left operand. The arrow operator has the same precedence as the member selection operator and associates from left to right. Thus the above is really


and that makes perfect sense. The pointer p is a pointer to a structure and login is one of the members of that structure. I then apply the member selection operator to the login member as desired.

Putting this all together I get

struct log_entry *p;

for (p = log; p < log + log_count; p++) {
  if (p->login.hours < 5) {
    printf("User %s logged in during early morning of %d-%d-%d\n",
      p->username, p->login.day, p->login.month, p->login.year

Notice how neat and clean the arrow operator makes this look?

Passing structures to functions

Since structures are first class variables you can pass them to a function just like any other variable. For example consider the following function that prints out a log_entry is a nice format. It might start like this

void print_log_entry(struct log_entry entry)
  printf("Session for %s:\n", entry.username);
  // etc...

Here I'm naming the formal parameter entry. It has the type struct log_entry. When I pass a struct log_entry variable to this function, that variable is copied and the function gets its own copy. Inside the function I can do whatever I want with the copy without changing the original in any way. This is exactly how ordinary variables are passed to a function.

While this works, you should do this carefully. Structures tend to be quite large and copying large variables can take a lot of time. Passing a whole structure into a function can cause the function to be sluggish. Unless the structure is small or unless you really need to give the function a copy, it is usually better to pass a pointer to the structure instead. Here is how that looks

void print_log_entry(struct log_entry *entry)
  printf("Session for %s:\n", entry->username);
  // etc...

In this case I'm only giving an address to the function. Since addresses are small passing an address is quick. The function uses that address to access the (possibly huge) structure as it sits back in the caller's list of variables. Since C has that handy-dandy arrow operator, accessing the members of the structure pointed at by the parameter is easy. This is, in fact, the main use of the arrow operator and the main reason for it existing in the language.

It turns out that you can also return whole structures from functions as well. The following works just fine.

struct log_entry get_next_entry(void)
  struct log_entry an_entry;

  strcpy(an_entry.username, ...);
  // etc...

  return an_entry;

Here I define get_next_entry to return a variable of type struct log_entry. Inside the function I declare a local variable of type struct log_entry and I name that variable an_entry. Then I do whatever is necessary to fill in the members of an_entry with appropriate data. (Perhaps I read the next line from the log file and interpret that line). Finally when an_entry is ready, I return the whole structure in one operation.

In my main program I might do

struct log_entry next;

next = get_next_entry();

Here I create a variable of type struct log_entry and I use that variable to "catch" the value returned by get_next_entry. This method works very well. The problem with it is that again it involves copying whole structures. That can be time consuming. Another technique that is often used instead looks like this

void get_next_entry(struct log_entry *);

struct log_entry next;


Here I'm imagining that get_next_entry takes a pointer to a struct log_entry. It then uses that pointer to "fill in" the members of the structure pointed at. In my main program I create a suitable placeholder variable and then pass the address of that variable to get_next_entry so that the variable can be filled up with data. This is basically the same idea that scanf uses to fill in integers.

What doesn't work is this

struct log_entry *get_next_entry(void);

struct log_entry *next;

next = get_next_entry();

Here I'm assuming that get_next_entry just returns a small pointer to a struct log_entry. In my main program I declare a pointer variable to receive get_next_entry's return value and I call get_next_entry to get things ready. This is fine so far. But how should get_next_entry look? Here is one attempt

struct log_entry *get_next_entry(void)
  struct log_entry an_entry;

  strcpy(an_entry.username, ...);
  // etc...

  return &an_entry;

Here I create a local struct log_entry named an_entry to hold the next values. I then do whatever is necessary to load up the members of an_entry with appropriate data. Finally I return the address of an_entry since I'm only supposed to be returning a pointer.

This looks good except... after get_next_entry returns all its local data will vanish. Thus the pointer it is returning will point at meaningless memory and the main program will become confused when it tries to access that memory. This is why it is usually better to have the main program allocate the structure and pass an address into the function rather than have the function allocate the structure and try to pass the address out.

Actually the above example can be fixed quite easily by making appropriate use of the static keyword. Do you see how? While this works, there are other issues with using static data that cause this approach to still be less favored.

So far I've sent time explaining ways of returning a structure without actually returning it! Because structures tend to be large, it is good to avoid copying them when possible. Thus despite the fact that you can pass structures to functions and return them from functions, it is often not done. Most of the time programmers use pointers as I've described to get the same effect. However, there are times when you do want to return a whole structure from a function. In C functions can only return a single value. Sometimes this is awkward. However, you could define a structure with several members, load up one of those structures in your function, and return the whole structure at once. This, in effect, gives you a way to return more than one value at a time. This technique is sometimes used.


  1. Since a structure is just another type, you can create arrays of structures the same way you create arrays of any type.

    struct log_entry log[1024];

    Each element in such an array is a structure and so you can apply the member selection operator to any element.

    printf("%s", log[i].username);
  2. Since a structure is just another type you can pass structure variables to functions the same way you pass other variables to functions.

    void f(struct log_entry input)
      // etc...

    As always, the structure you use for an argument is copied and the function works on the copy. Returning structures from functions is also straightforward.

    struct log_entry next(void)
      struct log_entry result;
      // Fill in the members of Result.
      return result;

    Keep in mind that passing structures to functions and returning structures from functions involves copying the structure. For large structures that might be time consuming.

  3. Many C programmers use pointers to structures with functions to avoid copying structures.

    struct log_entry result;
      /* Give function next the address of result so that it can access
         the members of result and/or update those members with the
         values it wants to return. */

    When working with a pointer to a structure, you will want to make use of the arrow operator to access the members of the structure pointed at by that pointer. For example

    void next(struct log_entry *p)
      strcpy(p->username, ...);
      // etc...
© Copyright 2003 by Peter C. Chapin.
Last Revised: July 8, 2003