pardo@june.cs.washington.edu (David Keppel) (03/27/89)
This is a request for information.
I recently made the following assertions about pANSI C.
Are they correct?
* The number zero (0) compares equal with any null pointer of any type.
* The bit storage of a null pointer is not necessarily all zeros.
* Null pointers of two types do not necessarily compare equal.
* A null pointer of one type, cast in to another pointer type, will
always be cast to a null pointer in the destination type.
;-D on ( All this, for a bunch of zeros! ) Pardo
--
pardo@cs.washington.edu
{rutgers,cornell,ucsd,ubc-cs,tektronix}!uw-beaver!june!pardobph@buengc.BU.EDU (Blair P. Houghton) (03/28/89)
In article <7672@june.cs.washington.edu> pardo@cs.washington.edu (David Keppel) writes: >I recently made the following assertions about pANSI C. >Are they correct? > >* The number zero (0) compares equal with any null pointer of any type. >* Null pointers of two types do not necessarily compare equal. If this is true, how long will it be until someone accidentally uses this fact to prove that one can violate the first and second laws of thermodynamics? --Blair "...since the zeroth is apparently conditional on data typing..."
henry@utzoo.uucp (Henry Spencer) (03/28/89)
In article <7672@june.cs.washington.edu> pardo@cs.washington.edu (David Keppel) writes: >I recently made the following assertions about pANSI C. >Are they correct? > >* The number zero (0) compares equal with any null pointer of any type. The integer *constant* 0, an integer *constant* expression that evaluates to 0, or either of those cast to "void *", compares equal with any null pointer. Comparing a general number, e.g. an int variable, to a pointer is illegal. And there is no guarantee that casting a general zero value to a pointer will produce a null pointer. The real problem here is that C has the notion of a null pointer constant, which is a magic value of pointer type, but instead of writing it as "nil" like some other languages do, C says that an integer constant with the value zero, when found in a pointer context, means null. The transformation into the magic pointer value is done at compile time, not at run time, so just any zero won't do -- it has to be something that the compiler knows to be zero. It's just an odd way of writing the null pointer *constant*; it does not imply any general relationship between integer zeros and null pointers. >* The bit storage of a null pointer is not necessarily all zeros. Correct. >* Null pointers of two types do not necessarily compare equal. Comparing two pointers of different types is, in general, illegal. The results of casting a pointer of one type to a different type is, in general, implementation-defined, but there is a specific exemption for null pointers. Any two null pointers, when cast to pointer types that are sufficiently similar for comparison to be legal, compare equal. >* A null pointer of one type, cast in to another pointer type, will > always be cast to a null pointer in the destination type. I do not actually find this to be guaranteed in the general case, although the requirement that it compare equal to a null pointer of the destination type will convince most implementors to make it so. The two properties of null pointers are that they compare equal to each other, and that they compare unequal to a pointer to any object or function (subject, in all cases, to the requirement that types be similar enough for comparison to be legal). The wording of 3.2.2.3 guarantees that casting to other pointer types preserves the first property, but the draft is silent about the effects of general pointer casts on the second property. (Note that a null pointer constant does not become a null pointer until it is converted to a specific pointer type, so *that* transformation is exempt from this.) (Note also that some specific pointer casts carry stronger guarantees.) (The above is based on the Oct draft, but is probably correct for more recent versions too.) -- Welcome to Mars! Your | Henry Spencer at U of Toronto Zoology passport and visa, comrade? | uunet!attcan!utzoo!henry henry@zoo.toronto.edu