Using memset() after malloc() causes assertion

Question

I have a struct like this:

struct state {
 bool isfinal;
 bool *isfull;
 char **board;
};

isfull is an array and board is a 2D array.

I can allocate memory for the arrays with this function:

struct state new_state(struct state state)
{
 int i;
 struct state new_state = {};
 new_state.isfull = (bool *)malloc(BOARD_WIDTH * sizeof(bool));
 new_state.board = (char **)malloc(BOARD_HIGHT * sizeof(char *));
 for (i = 0; i < BOARD_HIGHT; ++i)
 new_state.board[i] = (char *)malloc(BOARD_WIDTH * sizeof(char));
 if (state.board) {
 memcpy(new_state.isfull, state.isfull, BOARD_WIDTH * sizeof(bool));
 for (i = 0; i < BOARD_HIGHT; ++i)
 memcpy(new_state.board[i], state.board[i], BOARD_WIDTH * sizeof(char));
 } else {
 getchar();
 memset(new_state.isfull, 0, BOARD_WIDTH * sizeof(bool));
 for (i = 0; i < BOARD_HIGHT; ++i)
 memset(new_state.board[i], 0, BOARD_WIDTH * sizeof(char *));
 }
 return new_state;
}

First time this function is called with 0 and the else block will be executed. This results in an assertion failure.
Also in case of assertion, the function executes until the return statement and only then the program breaks.

I've solved the problem by using calloc() instead.
this works:

struct state new_state(struct state state)
{
 int i;
 struct state new_state = {};
 new_state.isfull = (bool *)calloc(BOARD_WIDTH, sizeof(bool));
 new_state.board = (char **)calloc(BOARD_HIGHT, sizeof(char *));
 for (i = 0; i < BOARD_HIGHT; ++i)
 new_state.board[i] = (char *)calloc(BOARD_WIDTH, sizeof(char));
 if (state.board) {
 memcpy(new_state.isfull, state.isfull, BOARD_WIDTH * sizeof(bool));
 for (i = 0; i < BOARD_HIGHT; ++i)
 memcpy(new_state.board[i], state.board[i], BOARD_WIDTH * sizeof(char));
 } else {
 // memset(new_state.isfull, 0, BOARD_WIDTH * sizeof(bool));
 // for (i = 0; i < BOARD_HIGHT; ++i)
 // memset(new_state.board[i], 0, BOARD_WIDTH * sizeof(char *));
 }
 return new_state;
}

and interestingly using a getchar() function as a form of delay also works!
like this:

struct state new_state(struct state state)
{
 int i;
 struct state new_state = {};
 new_state.isfull = (bool *)malloc(BOARD_WIDTH * sizeof(bool));
 new_state.board = (char **)malloc(BOARD_HIGHT * sizeof(char *));
 for (i = 0; i < BOARD_HIGHT; ++i)
 new_state.board[i] = (char *)malloc(BOARD_WIDTH * sizeof(char));
 if (state.board) {
 memcpy(new_state.isfull, state.isfull, BOARD_WIDTH * sizeof(bool));
 for (i = 0; i < BOARD_HIGHT; ++i)
 memcpy(new_state.board[i], state.board[i], BOARD_WIDTH * sizeof(char));
 } else {
 getchar();
 memset(new_state.isfull, 0, BOARD_WIDTH * sizeof(bool));
 for (i = 0; i < BOARD_HIGHT; ++i)
 memset(new_state.board[i], 0, BOARD_WIDTH * sizeof(char *));
 }
 return new_state;
}

searching the internet, I've found using memset() after malloc() is a common practice.
so first question: why doesn't the first version of my function work?
and second: why using getchar() make it work?

• G.M.: memcpy(new_state.board, state.board, BOARD_HIGHT + BOARD_HIGHT * BOARD_WIDTH); doesn't look right. Think about what that will do to the various new_state.board[i] assigned in the previous loop.
• pmg: Unless you're using C23 struct state new_state = {}; is an invalid initializer. If you're compiling with a C++ compiler, all bets are off.
• 4386427: "... first time this function is called with 0". Don't describe what you do! Post the coding do it. Post the shortest code that can give the assertion. BTW: What does "called with 0" mean?? The function takes a struct so 0 makes no sense.
• 4386427: And what are values of BOARD_WIDTH and BOARD_HIGHT? Are they compile time constants?

Answer 1

It looks like a typo, you use:

memset(new_state.board[i], 0, BOARD_HIGHT * sizeof (char *));

instead of:

memset(new_state.board[i], 0, BOARD_WIDTH * sizeof (char *));

If BOARD_HIGHT != BOARD_WIDTH then you can either leave elements uninitialized (and with indeterminate values) or you will go out of bounds and have definitive undefined behavior.

I'm assuming you did a copy-paste of the memset calls, and forgot to change this.

If you use calloc, you will save a few CPU cycles, and won't have the problem with the typo.

---

And as mentioned in a comment, you need a loop for copying the board member. Using memcpy will not work.

See e.g. this old answer of mine to understand why.

• amir z: thanks. hopefully in my case height and width are equal (both 10). so my question still remains... I will edit the question
• amir z: for the second part, thanks again. I was experimenting with using one malloc call for initializing the whole 2d array. I forgot to edit that part of the code. I will edit the question again. but that still does not answer my question.
• Some programmer dude: @amirz There could be something wrong in the code you don't show, something leading to undefined behavior. Please try to create a minimal reproducible example to show us.

Answer 2

The problem is an incorrect size passed to memset in:

memset(new_state.board[i], 0, BOARD_WIDTH * sizeof(char *));

You mean to clear the char array allocated for new_state.board[i] to all bits zero, but you compute the number of bytes using the size of a char * instead of a char, which produces a much larger size, causing memset to overwrite bytes beyond the end of the allocated objects. As written, the code has undefined behavior, and in your case it causes an assertion violation later, probably because you overwrote internal data structures used by malloc to keep track of the memory allocation state, which a later call to malloc, realloc or free detects and reports.

Your second version using calloc() is fine because you removed the bogus code. Using calloc() is a good approach because the memory returned by the C library call is in a consistent known state, at little or no penalty, which will prevent many potential cases of unpredictable behavior.

The third example, you have the same undefined behavior as the initial code, but because of differing circumstances (an extra call to getchar() that might allocate memory for its own purpose), the undefined behavior has different consequences... This is inherent to undefined behavior: anything can happen and it is impossible to predict if, when and how undesired side effects might manifest.

Here are basic recommendations to try and avoid undefined behavior:

• always test for allocation failure and leave data structures in a consistent state for the caller to detect such failures (you do not test any of the allocation calls in your code, whether using malloc or calloc.

• use the actual object type instead of the assumed type to compute the number of bytes for malloc, calloc, memcpy, memset as:

  memset(new_state.board[i], 0, BOARD_WIDTH * sizeof(*new_state.board[i]));

• use calloc() instead of malloc() to avoid non deterministic memory content for allocated objects.

Also note these remarks:

• {} is not a portable initializer in C.
• passing the structures by value both as arguments and return value is potentially risky and inefficient. It also makes it harder to report failures to the caller.
• if the matrix has a fixed size, consider inlining the isfull array and the board matrix in the state structure, and allocating that in a single call to calloc.
• use structure assignment instead of memcpy when possible.

Here is a modified version:

#include <stdlib.h>
// assuming BOARD_WIDTH and BOARD_HEIGHT have been defined as constant expressions
struct state {
 bool isfinal;
 bool isfull[BOARD_WIDTH];
 char board[BOARD_HEIGHT][BOARD_WIDTH];
};
/* allocate a copy of a state */ 
struct state *new_state(const struct state *state)
{
 struct state *new_state = calloc(1, sizeof(*new_state));
 if (new_state && state) {
 *new_state = *state;
 new_state->isfinal = false;
 }
 return new_state;
}

Answer 3

At least this issue:

---

Incorrct sizing

Code has

 for (i = 0; i < BOARD_HIGHT; ++i)
 new_state.board[i] = (char *) malloc(BOARD_WIDTH *sizeof (char));
...
 for (i = 0; i < BOARD_HIGHT; ++i)
 memcpy(new_state.board[i], state.board[i], BOARD_WIDTH * sizeof (char));
...
 for (i = 0; i < BOARD_HIGHT; ++i)
 memset(new_state.board[i], 0, BOARD_WIDTH * sizeof (char *));

The size of some are based on a char and another is based on a pointer. It is unlikely these are all correct.

Instead of trying to size to the type, consider, for these and other lines of code, sizing to the referenced object.
It is easier to code right, review and maintain.

for (i = 0; i < BOARD_HIGHT; ++i)
 new_state.board[i] = malloc(sizeof new_state.board[i][0] * BOARD_WIDTH);
...
for (i = 0; i < BOARD_HIGHT; ++i)
 memcpy(new_state.board[i], state.board[i], sizeof new_state.board[i][0] * BOARD_WIDTH);
...
for (i = 0; i < BOARD_HIGHT; ++i)
 memset(new_state.board[i], 0, sizeof new_state.board[i][0] * BOARD_WIDTH);