char * cast on NULL?

[Roland Illig]

Leonard den Ottolander wrote:
> Hello Oswald,
> 
> On Sun, 2004-09-26 at 13:15, Oswald Buddenhagen wrote:
> 
>>the outcome is, that it doesn't matter anywhere but when passing null
>>pointers through varargs. in all other cases it's about style, and taste
>>differs.
> 
> 
> All in all it seems rather pointless to do this in only certain parts of
> the code and at this point in time (approaching 4.6.1). If I need to
> know the type of the last argument I can find it elsewhere.

That's almost always correct, due to function prototypes. But there's 
one exception to it: varargs. (As Oswald noted above)

Imagine a machine where pointers take 32 bit and ints take 16.

/* concat strings */
void concat(const char *first, ...);

Then you call:

/*1*/ concat ("first", "second", (char *) NULL);
/*2*/ concat ("first", "second", NULL);
/*3*/ concat ("first", "second", 0);

The function concat expects their arguments to be strings, but the 
compiler cannot guarantee that only strings are passed. So see what 
could happen: Let's say the address of first is 0xc0000040, the address 
of second is 0xc0000050. Then the arguments (grouped in 16 bits) could be:

/*1*/ [0xc000] [0x0040] [0xc000] [0x0050] [0x0000] [0x0000] [garbage...]
/*3*/ [0xc000] [0x0040] [0xc000] [0x0050] [0x0000] [garbage...]

Some systems actually #define NULL to be 0, so case /*2*/ could be 
either like /*1*/ or like /*3*/. But in case /*3*/, the string list is 
not terminated correctly, and a segmentation fault will occur.

Roland