cppcheck 2.5: false positive in the return value of a function

[vix]

#define hiword(x)   ((unsigned short)((x) >> 16))
#define loword(x)   ((unsigned short)(x))

typedef struct {
  unsigned short  f;
  unsigned short  s;
} MyStruct;

MyStruct foo(double v)
{
   MyStruct retval={0};
   float u = (float)v;

   retval.f  = hiword(*(unsigned int *)&u);
   retval.s = loword(*(unsigned int *)&u);

   return(retval);
}

cppcheck gives
CWE: 562
Returning object that points to local variable 'u' that will be invalid when returning.
on the line
retval.s = loword(*(unsigned int *)&u);

But, if I'm right, the address of u is dereferenced and so its value is returned into retval.

[CHR]

Which version do you use, and how do you call cppcheck? With current head, I get

thud.c:14:16: portability: Casting from float * to unsigned int * is not portable due to different binary data representations on different platforms. [invalidPointerCast]
   retval.f  = hiword(*(unsigned int *)&u);
               ^
thud.c:15:15: portability: Casting from float * to unsigned int * is not portable due to different binary data representations on different platforms. [invalidPointerCast]
   retval.s = loword(*(unsigned int *)&u);
              ^

but not the warning you describe.

[vix]

I use cppcheck 2.5 from the GUI and I select Analyze >> Folder pointing to the folder with all my source files (.c and .h).
I get the same result if I copy and paste the code snippet into an empty file.c and select Analyze >> File.

I set the C standard to C99

I find the same warning as you, but moreover I get CWE: 562 as an error (not a warning).

[CHR]

Ok, I can reproduce it with 2.5. You can either use a current build, or wait for 2.6.

[Martin Brieg]

-- sorry, wrong thread --