[Swan-dev] pointer initialization and fcfcde7422a4805a56a3a4a175271c56fbbbab12
D. Hugh Redelmeier
hugh at mimosa.com
Mon Jul 4 19:15:17 UTC 2016
| From: Lennart Sorensen <lsorense at csclub.uwaterloo.ca>
| On Sat, Jun 25, 2016 at 10:24:15PM -0400, D. Hugh Redelmeier wrote:
| > This commit is technically wrong. I admit that the problem will not show
| > up on most machines.
| > The C standard does not require that the binary representation of a NULL
| > pointer be 0. Initializing a struct by zeroing bytes does not guarantee
| > that pointer fields are initialized to NULL.
| C11 does require that in fact.
| It states (in section 188.8.131.52.3):
| An integer constant expression with the value 0, or such an expression
| cast to type void *, is called a null pointer constant. If a null
| pointer constant is converted to a pointer type, the resulting pointer,
| called a null pointer, is guaranteed to compare unequal to a pointer to
| any object or function.
| And I have never heard of an implementation that didn't do that.
| Doing otherwise made things like "if (!pointer) ..." too damn hard to
| get right.
Let me be clearer. A pointer value that is null must test equal to
zero. Not true of the raw bytes of a null pointer (no conversion
happens when you deal with the raw bytes of a pointer).
passert(NULL = 0); /* must not fail */
memset(&p, 0, sizeof(p));
passert(p == 0); /* may fail */
p = 0;
passert(p == 0); /* must not fail */
static const unsigned char zeros[sizeof(void *)];
p = 0;
passert(memcmp(&p, zeros, sizeof(void*)) == 0); /* may fail */
Some machines with segmented memory take advantage of this liberty.
But most of those have died out.
More information about the Swan-dev