Specifically, what's dangerous about casting the result of malloc?

Now before people start marking this a dup, I've read all the following, none of which provide the answer I'm looking for:

1. C FAQ: What's wrong with casting malloc's return value?
2. SO: Should I explicitly cast malloc()’s return value?
3. SO: Needless pointer-casts in C
4. SO: Do I cast the result of malloc?

Both the C FAQ and many answers to the above questions cite a mysterious error that casting malloc's return value can hide; however, none of them give a specific example of such an error in practice. Now pay attention that I said error, not warning.

Now given the following code:

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

int main(int argc, char** argv) {

    char * p = /*(char*)*/malloc(10);
    strcpy(p, "hello");
    printf("%s\n", p);

    return 0;
}

Compiling the above code with gcc 4.2, with and without the cast gives the same warnings, and the program executes properly and provides the same results in both cases.

anon@anon:~/$ gcc -Wextra nostdlib_malloc.c -o nostdlib_malloc
nostdlib_malloc.c: In function ‘main’:
nostdlib_malloc.c:7: warning: incompatible implicit declaration of built-in function ‘malloc’
anon@anon:~/$ ./nostdlib_malloc 
hello

So can anyone give a specific code example of a compile or runtime error that could occur because of casting malloc's return value, or is this just an urban legend?

Edit I've come across two well written arguments regarding this issue:

1. In Favor of Casting: CERT Advisory: Immediately cast the result of a memory allocation function call into a pointer to the allocated type
2. Against Casting (404 error as of 2012-02-14: use the Internet Archive Wayback Machine copy from 2010-01-27.{2016-03-18:"Page cannot be crawled or displayed due to robots.txt."})

Solution 1:

You won't get a compiler error, but a compiler warning. As the sources you cite say (especially the first one), you can get an unpredictable runtime error when using the cast without including stdlib.h.

So the error on your side is not the cast, but forgetting to include stdlib.h. Compilers may assume that malloc is a function returning int, therefore converting the void* pointer actually returned by malloc to int and then to your pointer type due to the explicit cast. On some platforms, int and pointers may take up different numbers of bytes, so the type conversions may lead to data corruption.

Fortunately, modern compilers give warnings that point to your actual error. See the gcc output you supplied: It warns you that the implicit declaration (int malloc(int)) is incompatible to the built-in malloc. So gcc seems to know malloc even without stdlib.h.

Leaving out the cast to prevent this error is mostly the same reasoning as writing

if (0 == my_var)

instead of

if (my_var == 0)

since the latter could lead to a serious bug if one would confuse = and ==, whereas the first one would lead to a compile error. I personally prefer the latter style since it better reflects my intention and I don't tend to do this mistake.

The same is true for casting the value returned by malloc: I prefer being explicit in programming and I generally double-check to include the header files for all functions I use.

Solution 2:

One of the good higher-level arguments against casting the result of malloc is often left unmentioned, even though, in my opinion, it is more important than the well-known lower-level issues (like truncating the pointer when the declaration is missing).

A good programming practice is to write code, which is as type-independent as possible. This means, in particular, that type names should be mentioned in the code as little as possible or best not mentioned at all. This applies to casts (avoid unnecessary casts), types as arguments of sizeof (avoid using type names in sizeof) and, generally, all other references to type names.

Type names belong in declarations. As much as possible, type names should be restricted to declarations and only to declarations.

From this point of view, this bit of code is bad

int *p;
...
p = (int*) malloc(n * sizeof(int));

and this is much better

int *p;
...
p = malloc(n * sizeof *p);

not simply because it "doesn't cast the result of malloc", but rather because it is type-independent (or type-agnositic, if you prefer), because it automatically adjusts itself to whatever type p is declared with, without requiring any intervention from the user.

Solution 3:

Non-prototyped functions are assumed to return int.

So you're casting an int to a pointer. If pointers are wider than ints on your platform, this is highly risky behavior.

Plus, of course, that some people consider warnings to be errors, i.e. code should compile without them.

Personally, I think the fact that you don't need to cast void * to another pointer type is a feature in C, and consider code that does to be broken.

Solution 4:

If you do this when compiling in 64-bit mode, your returned pointer will be truncated to 32-bits.

EDIT: Sorry for being too brief. Here's an example code fragment for discussion purposes.

main()
{
   char * c = (char *)malloc(2) ;
   printf("%p", c) ;
}

Suppose that the returned heap pointer is something bigger than what is representable in an int, say 0xAB00000000.

If malloc is not prototyped to return a pointer, the int value returned will initially be in some register with all the significant bits set. Now the compiler say, "okay, how do I convert and int to a pointer". That's going to be either a sign extension or zero extension of the low order 32-bits that it has been told malloc "returns" by omitting the prototype. Since int is signed I think the conversion will be sign extension, which will in this case convert the value to zero. With a return value of 0xABF0000000 you'll get a non-zero pointer that will also cause some fun when you try to dereference it.