The kind of comments telling you to add checks against null are typically coming from programmers mostly used to deal with higher level programming languages. They think that generally, more explicit error handling is always a good thing. And that is true in most cases, but not in this one.

The spirit of C has always been performance over safety. Give the programmer the freedom to do what they want, as fast as possible, at the price of potential risks. For example arrays were purposely not designed with a stored size or any bounds-checking (The Development of the C Language - Dennis M. Ritchie).

If we look at the standard library, lots of functions like memcpy/strcpy do for example not support overlapping parameters even though they could. Instead a specialized function memmove was designed for that purpose - safer but slower. Similarly, malloc does not needlessly initialize the allocated memory to zero/null, because that would lead to execution overhead. A specialized function calloc was designed for that purpose. And so on.

We should never add checks against null because they add needless overhead code.

Instead the function should be explicitly documented to not handle the case where an argument which is a null pointer is passed, thereby leaving error handling to the caller.

The reason for this is simple: there are a whole lot of scenarios where the caller is sure that the passed arguments are most definitely not null pointers. Having the function needlessly checking for null pointers only adds bloat in the form of additional branches. Take this example:

char* array[n] = { ... };
for(size_t i=0; i<n; i++)
{
  func(array[i]);
}

Now if this snippet is performance-critical, we stall the whole loop in case func repeatedly checks the passed array for null. We know it isn't a null pointer, but the repeated check may lead to branch prediction problems or cache misses on some systems. On any system, it means a useless check taking up performance. And it cannot get optimized away unless the function is inlined, perhaps not even then.

To give the caller the freedom to do as they like, we should let them handle the null pointer check. Which should ideally be done as close to the point where something might end up as a null pointer, rather than inside some unrelated library function.

As a side note, some very modern compilers like recent gcc or clang versions have limited possibilities of static analysis at compile time, in case we use the exotic static declarator feature (-Wall is necessary for this in gcc). But it has very limited use:

// must point at array of at least 1 item and not null
void func (int a[static 1]); 

int main (void) 
{
  func(NULL); // warning null passed to a callee that requires a non-null argument 
  
  int* x=NULL;
  func(x);   // no warning, the compiler can't predict what x contains
}

It's better to use dedicated static analyser tools to find bugs like this and then we don't need the exotic language feature either. Compilers are still to this date not very good at finding application bugs through static analysis.

posted about 2 years ago

CC BY-SA 4.0

2y ago

Lundin‭

8077 reputation 33 109 919 152