How to fix strict-aliasing with functions that use void**

Stephen Hill steve.d.hill.uk@googlemail.com
Sat Jan 10 22:46:00 GMT 2009


I'm a newbie to GCC 4.x and I am getting lots of warnings about type 
punning with code that use void** (I am using -O3).
I have functions declared to perform dynamic memory allocation that use 
void**:

   bool RiMalloc(void **pBlock, int nSize);
   void RiFree(void **pBlock);

So, for example, we call the functions thus:

   typedef struct {
      int x;
      int y;
   } tWibble;

   tWibble* void test_func(void)
   {  
      tWibble* pZ = NULL;
      if (RiMalloc(&pZ, sizeof(tWibble)))
      {
         pZ->x = 5;
         pZ->y = 10
      }
      return pZ;
   }

Compiling this code with strict-aliasing enabled produces a warning.

Putting a void* cast in front of &pZ in the RiMalloc() call stops the 
warning being produced, but I am unsure as to whether I am really fixing 
the problem or hiding it:

   tWibble* void test_func(void)
   {  
      tWibble* pZ = NULL;
      if (RiMalloc((void*)&pZ, sizeof(tWibble)))
      {
         pZ->x = 5;
         pZ->y = 10
      }
      return pZ;
   }


Another way that stops the warning being produced is to create another 
pointer with void * type:

   tWibble* void test_func(void)
   {
      void *pvZ = NULL;
      if (RiMalloc(&pvZ, sizeof(tWibble)))
      {
         tWibble* pZ;
        
         pZ = pvZ;
         pvZ->x = 5;
         pvZ->y = 10
      }
      return pvZ;
   }

But isn't this creating an aliasing problem, as now I have two pointers 
of different types that point to the same memory location?
What is the correct way of fixing these strict-aliasing warnings without 
changing the functions API?
Using a union of the pointers appears to be invalid.
I really do not want to change the RiMalloc() and RiFree() interface as 
it is used extensively in our code.
I am no expert on the assembly language produced for the processor I am 
using, so I am dubious of my ability to check the compiler output.

Thanks for your help,
Steve.



More information about the Gcc-help mailing list