strange integer-pointer behaviour

[matevzb]

I've ran into some fishy code that, at first glance, is buggy, but it
seems to work correctly
and none of the compilers I've tried (five so far, on various systems)
gives any warnings.
The code:
============================
#include <stdio.h>

void
fcn (char *str)
{
if (str == '\0')
{
printf ("str!\n");
}
}

int
main (void)
{
fcn ('\0');
return 0;
}
============================
My understanding so far: '\0' is passed to fcn(), it equals integer 0,
which in turn "equals"
a NULL pointer in terms of zero-ness. The 0 in fcn() is then compared
to '\0' which is also a 0.
No warnings, no errors, code works (or so it seems).
But - if I pass e.g. '\1' to fcn() than all the compilers complain
with similar warnings,
i.e. making a pointer from integer without a cast.

My question is, why don't the compilers complain when '\0' is passed?
Shouldn't they give
the same warning, as it's also an integer (albeit 0 in value)?

 

[Eric Sosman]

matevzb wrote On 08/13/07 13:35,:

I've ran into some fishy code that, at first glance, is buggy, but it
seems to work correctly
and none of the compilers I've tried (five so far, on various systems)
gives any warnings.
The code:
============================
#include <stdio.h>

void
fcn (char *str)
{
if (str == '\0')
{
printf ("str!\n");
}
}

int
main (void)
{
fcn ('\0');
return 0;
}
============================
My understanding so far: '\0' is passed to fcn(), it equals integer 0,
which in turn "equals"
a NULL pointer in terms of zero-ness. The 0 in fcn() is then compared
to '\0' which is also a 0.
No warnings, no errors, code works (or so it seems).
But - if I pass e.g. '\1' to fcn() than all the compilers complain
with similar warnings,
i.e. making a pointer from integer without a cast.

My question is, why don't the compilers complain when '\0' is passed?
Shouldn't they give
the same warning, as it's also an integer (albeit 0 in value)?

A literal constant zero[*] used in a pointer context
*is* a null pointer constant. In the code above, '\0'
is just a decorative way to write a literal constant zero.
The compiler sees that it's being used where a pointer is
expected, so the compiler recognizes the zero as a null
pointer constant instead of as an integer. Indeed, you
will find that on many systems the NULL macro is defined
as an unadorned zero.

[*] Any zero-valued constant integer expression will
do; try `fcn (42 / 100)', for example. Also, the
expression is still a null pointer constant if it
is cast to `void*'.

This special treatment applies only to

- Zero-valued expressions: If the expression's value
is non-zero, the expression cannot be a null pointer
constant. That's why the compiler complained when
you tried to use '\1' as an argument to fcn().

- Integer-valued expressions: If the expression's type
is `double' or `float' or `struct fubar', it is not
a null pointer constant. fcn(0.0) won't work.

- Constant expressions: If the expression is not a
compile-time constant, it is not a null pointer
constant. fcn(x - x) won't work; you can deduce
that the argument will always be zero, but from
the compiler's point of view it's just a remarkable
coincidence.

The FAQ has a section devoted to null pointers; you
might find it helpful.

 

[matevzb]

A literal constant zero[*] used in a pointer context
*is* a null pointer constant. In the code above, '\0'
is just a decorative way to write a literal constant zero.
The compiler sees that it's being used where a pointer is
expected, so the compiler recognizes the zero as a null
pointer constant instead of as an integer. Indeed, you
will find that on many systems the NULL macro is defined
as an unadorned zero.

Yes, sorry about even posting this. Moments after I posted, I realized
that all bets are off with 0 with regards to (pointer) type
checking...
Thanks

 

[Eric Sosman]

matevzb wrote On 08/13/07 14:30,:

A literal constant zero[*] used in a pointer context
*is* a null pointer constant. In the code above, '\0'
is just a decorative way to write a literal constant zero.
The compiler sees that it's being used where a pointer is
expected, so the compiler recognizes the zero as a null
pointer constant instead of as an integer. Indeed, you
will find that on many systems the NULL macro is defined
as an unadorned zero.

Yes, sorry about even posting this. Moments after I posted, I realized
that all bets are off with 0 with regards to (pointer) type
checking...

Well, no, not exactly. There is no need to type-check
a NULL, because NULL is a perfectly good value for any kind
of pointer at all. The confusion (what remains of it) comes
from the fact that there are many ways to spell NULL.

Gertrude Stein never said "A NULL is a '\0' is a 0,"
but she could have.

 

[CBFalconer]

I've ran into some fishy code that, at first glance, is buggy, but
it seems to work correctly and none of the compilers I've tried
(five so far, on various systems) gives any warnings. The code:

#include <stdio.h>

void fcn (char *str) {

This wants a pointer to char as the parameter.

if (str == '\0') {

This tests the parameter against a fixed char.

printf ("str!\n");
}
}

int main (void) {
fcn ('\0');

This passes an integer (char representation) to something that
wants a char. pointer.

 

[Keith Thompson]

matevzb wrote On 08/13/07 14:30,:

A literal constant zero[*] used in a pointer context
*is* a null pointer constant. In the code above, '\0'
is just a decorative way to write a literal constant zero.
The compiler sees that it's being used where a pointer is
expected, so the compiler recognizes the zero as a null
pointer constant instead of as an integer. Indeed, you
will find that on many systems the NULL macro is defined
as an unadorned zero.

Yes, sorry about even posting this. Moments after I posted, I realized
that all bets are off with 0 with regards to (pointer) type
checking...

Well, no, not exactly. There is no need to type-check
a NULL, because NULL is a perfectly good value for any kind
of pointer at all. The confusion (what remains of it) comes
from the fact that there are many ways to spell NULL.

Gertrude Stein never said "A NULL is a '\0' is a 0,"
but she could have.

On the other hand, I'd like my compiler to warn me about using '\0' as
a null pointer constant, at least optionally. It's perfectly legal,
but it usually indicates a conceptual error on the part of the
programmer.

 

[Default User]

matevzb wrote On 08/13/07 14:30,:

Well, no, not exactly. There is no need to type-check
a NULL, because NULL is a perfectly good value for any kind
of pointer at all. The confusion (what remains of it) comes
from the fact that there are many ways to spell NULL.

Gertrude Stein never said "A NULL is a '\0' is a 0,"
but she could have.

Well, her code was notoriously buggy anyway.




Brian

 

[CBFalconer]

.... snip ...

Well, no, not exactly. There is no need to type-check
a NULL, because NULL is a perfectly good value for any kind
of pointer at all. The confusion (what remains of it) comes
from the fact that there are many ways to spell NULL.

Gertrude Stein never said "A NULL is a '\0' is a 0,"
but she could have.

I don't agree that '\0' is a spelling of NULL. It is a spelling of
the int value zero. The primes make it precisely an int.

 

[somenath]

- Constant expressions: If the expression is not a

compile-time constant, it is not a null pointer
constant. fcn(x - x) won't work; you can deduce
that the argument will always be zero, but from
the compiler's point of view it's just a remarkable
coincidence.

But I compiled the following code using "gcc -Wall " options.It does
not throw even warning .
#include <stdio.h>

void
fcn (char *str)
{
if (str == NULL)
{
printf ("str!\n");
}



}


int
main (void)
{
int x;
fcn (x - x);
return 0;

}

 

[Eric Sosman]

somenath wrote On 08/14/07 10:26,:

But I compiled the following code using "gcc -Wall " options.It does
not throw even warning .
#include <stdio.h>

void
fcn (char *str)
{
if (str == NULL)
{
printf ("str!\n");
}



}


int
main (void)
{
int x;
fcn (x - x);
return 0;

}

<off-topic>

Try "-Wall -W -O2" or even "-Wall -W -ansi -pedantic -O2".
Despite its inclusive-sounding name, "-Wall" does not enable
all of the warnings gcc can generate.

</off-topic>

As for the `x - x', the Standard says (6.3.2.3p3):

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.

The question of whether `x - x' is a null pointer constant
thus hinges on whether it is an integer constant expression.
The Standard defines I.C.E. in 6.6p6:

An _integer constant expression_ shall have integer
type and shall only have operands that are integer
constants, enumeration constants, character constants,
sizeof expressions whose results are integer constants,
and floating constants that are the immediate operands
of casts. [...]

Since `x - x' has operands that are impermissible in an
I.C.E., `x - x' is not an I.C.E. and hence not an N.P.C.

 

[Eric Sosman]

CBFalconer wrote On 08/13/07 16:20,:

This wants a pointer to char as the parameter.

Right.

This tests the parameter against a fixed char.

No, t tests the parameter against a null pointer
constant.

This passes an integer (char representation) to something that
wants a char. pointer.

No, it passes a NULL-valued `char*' to fcn(). '\0' is
a zero-valued integer constant expression, hence it is a
null pointer constant. It is used in a pointer context
(thanks to the prototype), so the N.P.C. becomes a null
pointer. The N.P. is converted to type `char*' (again,
thanks to the prototype) and that value is passed to the
function.

 

[Richard Bos]

I don't agree that '\0' is a spelling of NULL.

The only spelling of NULL, which is the name of a macro, is NULL. '\0'
is, however, a spelling of the null pointer constant.

It is a spelling of the int value zero.

No, it's an integer _constant_ of type int and value zero. Big
difference. An integer constant with value zero is a null pointer
constant; an int with value zero is not, and is not guaranteed (though
highly likely) to compare equal to a null pointer object.

Yes, I do agree that writing NULL when you mean "a null pointer
constant" is a bad idea, because it adds to the confusion many newbies
have concerning null pointers.

Fix your .sig.

Richard

 

[Harald van =?UTF-8?B?RMSzaw==?=]

This wants a pointer to char as the parameter.


This tests the parameter against a fixed char.

As others have pointed out, '\0' is an allowable null pointer constant, but
also, even in other contexts, '\0' isn't a char. It's an int. You might be
thinking of C++, where it would be a char.

This passes an integer (char representation) to something that
wants a char. pointer.

As above, it passes a null pointer constant to a function expecting a
pointer argument.

 

[Army1987]

I don't agree that '\0' is a spelling of NULL. It is a spelling of
the int value zero. The primes make it precisely an int.

Nothing in the standard forbids <stddef.h> or other standard
headers to #define NULL '\0'.

 

[Keith Thompson]

The only spelling of NULL, which is the name of a macro, is NULL. '\0'
is, however, a spelling of the null pointer constant.

Of *a* null pointer constant.

No, it's an integer _constant_ of type int and value zero. Big
difference. An integer constant with value zero is a null pointer
constant;

Right. More generally, so is an integer constant expression with
value zero.

an int with value zero is not, and is not guaranteed (though
highly likely) to compare equal to a null pointer object.

An "int with value zero" presumably refers to an object that exists
during execution time. A null pointer constant exists only in source
code; there's no such thing during execution.

An int with value zero cannot compare equal (or unequal) to a pointer
whose value is a null pointer. An attempt to compare them is a
constraint violation, because the types are incompatible.

It's likely, but not guaranteed, that *converting* a (non-constant)
int with value zero to a pointer type will yield a null pointer value.
Since integer-to-pointer conversion is implementation-defined, this is
actually independent of the question of whether a null pointer value
is represented as all-bits-zero.

[...]

 

[Richard Bos]

Of *a* null pointer constant.

_The_ spelling of _a_ null pointer constant, or _a_ spelling of _the_
null pointer constant. Take your pick; IMO, the difference is of an
angels-on-needlepoints value.

Richard

 

[Keith Thompson]

_The_ spelling of _a_ null pointer constant, or _a_ spelling of _the_
null pointer constant. Take your pick; IMO, the difference is of an
angels-on-needlepoints value.

It's not a big deal, but there isn't a single null pointer constant;
there are arbitrarily many of them.

An integer constant expression with the value 0, or such an
expression cast to type void *, is called a _null pointer
constant_.

Referring to "the null pointer constant" makes no more sense than
referring to "the integer constant". (A difference is that all null
pointer constants refer to the same value, but they're still distinct
constants.)

 

[Kenneth Brody]

matevzb wrote On 08/13/07 14:30,: [...]

Yes, sorry about even posting this. Moments after I posted, I realized
that all bets are off with 0 with regards to (pointer) type
checking...

Well, no, not exactly. There is no need to type-check
a NULL, because NULL is a perfectly good value for any kind
of pointer at all. The confusion (what remains of it) comes
from the fact that there are many ways to spell NULL.

Gertrude Stein never said "A NULL is a '\0' is a 0,"
but she could have.

A NULL by any other name would smell as sweet to those nasal demons.

--
+-------------------------+--------------------+-----------------------+
| Kenneth J. Brody | www.hvcomputer.com | #include |
| kenbrody/at\spamcop.net | www.fptech.com | <std_disclaimer.h> |
+-------------------------+--------------------+-----------------------+
Don't e-mail me at: <mailto:[email protected]>