Can the `restrict` keyword be used in the function definition only? - c

I am wondering if it is possible to include the restrict keyword only in the function definition and not in the function declaration like so:
void foo(char *bar);
void foo(char * restrict bar)
{
// do something
}
Since foo only takes one argument, any pointer aliasing would have to take place inside foo. There would be no need for the person calling the function to know about the restrict modifier. Would it be fine to omit the keyword in only the function declaration, just like with const?

You may use restrict on parameters in function declarations whether they are definitions or not, as it is allowed by the C grammar and there is no rule against it. However, they have no effect to the compiler in declarations that are not definitions. This is because 6.5.2.2 7 says qualifiers are removed when passing arguments to functions with prototypes:
… the arguments are implicitly converted, as if by assignment, to the types of the corresponding parameters, taking the type of each parameter to be the unqualified version of its declared type.
Thus, if a function declaration has a parameter of type int * restrict a, whatever argument you pass is converted to the unqualified type, int *.
Further, two otherwise identical function declarations are compatible even if the qualifiers on parameters are changed, because C 2018 6.7.6.3 15 says:
… (In the determination of type compatibility and of a composite type, … each parameter declared with qualified type is taken as having the unqualified version of its declared type.)
However, this applies only to the parameter itself. The parameter is not affected by a restrict that qualifies it. But it can point to a pointer that is restrict-qualified. For example, void foo(void * restrict *a); and void foo(void **a); declare different function types.
Although qualifiers on parameters in declarations have no effect to the compiler, they can signal to humans that the arguments are expected to conform to the restriction. Inside the function definition, the parameter is restrict-qualified, and anybody calling the function should respect that.

The next C99 source code can show you that the output of the program depends on restrict :
__attribute__((noinline))
int process(const int * restrict const a, int * const b) {
*b /= (*a + 1) ;
return *a + *b ;
}
int main(void) {
int data[2] = {1, 2};
return process(&data[0], &data[0]);
}
The software terminates with code 1 using restrict and 0 without restrict qualifier.
The compilation is done with gcc -std=c99 -Wall -pedantic -O3 main.c.
The flag -O1 do the job too.
And so on.

Related

Compatibility of function types that does not include a prototype

There is a rule for function type compatibility N2310 6.7.6.3(p15):
If one type has a parameter type list and the other type is specified
by a function declarator that is not part of a function definition and
that contains an empty identifier list, the parameter list shall not
have an ellipsis terminator and the type of each parameter shall be
compatible with the type that results from the application of the
default argument promotions.
I can imagine an example:
#include <stdio.h>
int foo();
float bar();
int main(void){
printf("%d\n", foo(1, 3)); //fine, int is unchanged by default argument promotion
printf("%f\n", bar(1.0f, 2.0f)); //error, float is promoted to double
}
int foo(int a, int b){
return a + b;
}
float bar(float b, float c){
return b + c;
}
The thing that I found contradictory was that 6.5.2.2(p6) mentions that :
If the number of arguments does not equal the number of parameters,
the behavior is undefined.
In the case of int foo() it has an empty identifier-list. So does a call printf("%d\n", foo(1, 3)); yield UB (2 arguments were supplied)?
Anyway the rules look pretty strange and kind of unnatural. What was the reason for that? I suppose some backward compatibility with previous versions of the Standard... ?
C 2018 6.7.6.3 15 tells you whether two types are compatible. So it can be used to compare two declarations. For foo, you have:
int foo();
int foo(int a, int b) {...}
Of these, the second has a parameter list, and the first is specified by a function declaration that is not part of a function definition and that contains an empty identifier list. So the rule is 6.7.6.3 15 applies. It says the parameter list shall not have an ellipsis terminator (and it does not), and that the type of each parameter shall be compatible with the type that results from the default argument promotions (and they are, since int produces int).
Then, for bar, we have:
float bar();
float bar(float b, float c) {...}
Again, 6.7.6.3 15 applies. But in this case, each parameter has a type that does not result from the default argument promotions, since the default promotions convert float to double. So these two declarations declare bar with incompatible types.
Regarding 6.5.2.2 6:
… If the number of arguments does not equal the number of parameters, the behavior is undefined…
This refers to the number of parameters of the actual function, not the number of parameters appearing in the (empty) list in the declaration.
Anyway the rules look pretty strange and kind of unnatural. What was the reason for that? I suppose some backward compatibility with previous versions of the Standard... ?
Yes, C was originally lax about function declarations, allowing functions to be declared with empty parameter lists, if I recall correctly, all arguments were passed with the promoted types. Support for stricter and more precise declarations came later, and the rules are written to allow old code to continue to work.
Note that the rule about compatibility of function types is relevant for the declarations of the functions, but not the calls.
When a function call is being analyzed by the compiler, the rules in 6.5.2.2 are used to prepare the call. These rules say the arguments are treated in various ways according to the declaration of the function that is visible at the point of the call. (Technically, to the type of the expression that denotes the called function. This is often a function name but could be a pointer to a function, including one computed by a cast expression.)
The rules about compatibility assure you that, if you call a function using a type that is compatible with the type of the actual function definition, then the call has defined behavior.

Function without prototype called with non-compatible type

I read the Standard N1570 section 6.5.2.2 Function calls and was confused about the special meaning of the function type that includes prototype. Precisely 6.5.2.2(p6)
If the function is defined with a type that does not include a
prototype, and the types of the arguments after promotion are not
compatible with those of the parameters after promotion, the behavior
is undefined, except for the following cases:
— one promoted type is a signed integer type, the other promoted type
is the corresponding unsigned integer type, and the value is
representable in both types;
— both types are pointers to qualified or unqualified versions of a
character type or void.
6.5.2.2(p7) provides a rule of calling function with prototype:
If the expression that denotes the called function has a type that
does include a prototype, the arguments are implicitly converted, as
if by assignment, to the types of the corresponding parameters, taking
the type of each parameter to be the unqualified version of its
declared type.
Consider the following example:
struct test_arg{
int a;
};
void test_no_prototype(const struct test_arg a){ }
void test_with_prototype(const struct test_arg a);
void test_with_prototype(const struct test_arg a){ }
int main(){
struct test_arg test = {.a = 42};
test_no_prototype(test); //1 UB?
test_with_prototype(test); //2 Fine?
}
I think that 1 is UB because test_no_prototype does not include prototype and test has non-qualified version of struct test_arg, but the argument has type const struct test_arg which is non-compatible with struct test_arg because of different qualification.
I think that 2 is fine because test_with_prototype includes prototype and the simple assignment constraints from 6.5.16.1(p1) allow assignment of to a variable of qualified struct type from non-qualified version of the same struct.
This seems strange and for now I cannot imagine any reason of why we treat functions with and without prototype differently. Probably I understood the rule incorrectly... If so could you explain what it means?
The term prototype does not mean a declaration of a function preceding its definition. It means a declaration of a function that declares the types of its parameters (C 2018 6.2.1 2).
test_no_prototype has a prototype because void test_no_prototype(const struct test_arg a){ } declares the type of its parameter, const struct test_arg.
An example of a declaration without a prototype is void test_no_prototype();. It is an old style of declaration that should not be used in new code.

The void type in C

The void type in C seems to be strange from various different situations. Sometimes it behaves like a normal object type, such as int or char, and sometimes it just means nothing (as it should).
Look at my snippet. First of all, it seems strange that you can declare a void object, meaning you just declare nothing.
Then I created an int variable and casted its result to void, discarding it:
If an expression of any other type is evaluated as a void
expression, its value or designator is discarded. (ISO/IEC 9899:201x, 6.3.2.2 void)
I tried to call my function with a void cast, but my compiler gave me (Clang 10.0):
error: too many arguments to function call, expected 0, have 1
So the void in a prototype means nothing, and not the type void.
But then, I created a pointer to void, dereferenced it, and assigning the “result” to my int variable. I got the “incompatible type” error. That means the void type does exist here.
extern void a; // Why is this authorised ???
void foo(void); // This function takes no argument. Not the 'void' type.
int main(void)
{
int a = 42;
void *p;
// Expression result casted to 'void' which discards it (per the C standard).
(void)a;
// Casting to 'void' should make the argument inexistant too...
foo((void)a);
// Assigning to 'int' from incompatible type 'void': so the 'void' type does exists...
a = *p;
// Am I not passing the 'void' type ?
foo(*p);
return 0;
}
Is void an actual type, or a keyword to means nothing ? Because sometimes it behaves like the instruction “nothing is allowed here”, and sometimes like an actual type.
EDIT: This questions is NOT a duplicate. It is a purely about the semantics of the void type. I do not want any explanation about how to use void, pointers to void or any other things. I want an answer per the C standard.
In C language the void type has been introduced with the meaning of 'don't care' more than 'null' or 'nothing', and it's used for different scopes.
The void keyword can reference a void type, a reference to void, a void expression, a void operand or a void function. It also explicitly defines a function having no parameters.
Let's have a look at some of them.
The void type
First of all void object exists and have some special properties, as stated in ISO/IEC 9899:2017, §6.2.5 Types:
The void type comprises an empty set of values; it is an incomplete object type that cannot be completed.
Pointers
The more useful reference to void, or void *, is a reference to an incomplete type, but itself is well defined, and then is a complete type, have a size, and can be used as any other standard variable as stated in ISO/IEC 9899:2017, §6.2.5 Types:
A pointer to void shall have the same representation and alignment requirements as a pointer to a character type.
Similarly, pointers to qualified or unqualified versions of compatible types shall have the same representation and alignment requirements.
All pointers to structure types shall have the same representation and alignment requirements as each other.
All pointers to union types shall have the same representation and alignment requirements as each other.
Pointers to other types need not have the same representation or alignment requirements.
Casting to void
It can be used as cast to nullify an expression, but allowing the completion of any side effect of such expression. This concept is explained in the standard at ISO/IEC 9899:2017, §6.3 Conversions, §6.3.2.2 void:
The (nonexistent) value of a void expression (an expression that has type void) shall not be used in any way, and implicit or explicit conversions (except to void) shall not be applied to such an expression.
If an expression of any other type is evaluated as a void expression, its value or designator is discarded. (A void expression is evaluated for its side effects.)
A practical example for the casting to void is its use to prevent warning for unused parameters in function definition:
int fn(int a, int b)
{
(void)b; //This will flag the parameter b as used
... //Your code is here
return 0;
}
The snippet above shows the standard practice used to mute compiler warnings. The cast to void of parameter b acts as an effective expression that don't generate code and marks b as used preventing compiler complains.
void Functions
The paragraph §6.3.2.2 void of the standard, covers also some explanation about void functions, that are such functions that don't return any value usable in an expression, but functions are called anyway to implement side effects.
void pointers properties
As we said before, pointers to void are much more useful because they allow to handle objects references in a generic way due to their property explained in ISO/IEC 9899:2017, §6.3.2.3 Pointers:
A pointer to void may be converted to or from a pointer to any object type.
A pointer to any object type may be converted to a pointer to void and back again; the result shall compare equal to the original pointer.
As practical example imagine a function returning a pointer to different objects depending on input parameters:
enum
{
FAMILY, //Software family as integer
VERSION, //Software version as float
NAME //Software release name as char string
} eRelease;
void *GetSoftwareInfo(eRelease par)
{
static const int iFamily = 1;
static const float fVersion = 2.0;
static const *char szName = "Rel2 Toaster";
switch(par)
{
case FAMILY:
return &iFamily;
case VERSION:
return &fVersion;
case NAME:
return szName;
}
return NULL;
}
In this snippet you can return a generic pointer that can be dependent on input par value.
void as functions parameter
The use of void parameter in functions definitions was introduced after the, so called, ANSI-Standard, to effectively disambiguate functions having variable number of arguments from functions having no arguments.
From standard ISO/IEC 9899:2017, 6.7.6.3 Function declarators (including prototypes):
The special case of an unnamed parameter of type void as the only item in the list specifies that the function has no parameters.
Actual compilers still support function declaration with empty parenthesis for backward compatibility, but this is an obsolete feature that will eventually be removed in future release of standard. See Future directions - §6.11.6 Function declarators:
The use of function declarators with empty parentheses (not prototype-format parameter type declarators) is an obsolescent
feature.
Consider the following example:
int foo(); //prototype of variable arguments function (backward compatibility)
int bar(void); //prototype of no arguments function
int a = foo(2); //Allowed
int b = foo(); //Allowed
int c = bar(); //Allowed
int d = bar(1); //Error!
Now resembling your test, if we call the function bar as follows:
int a = 1;
bar((void)a);
Triggers an error, because casting to void an object doesn't null it. So you are still trying to pass a void object as parameter to a function that don't have any.
Side effects
As requested this is a short explain for side effects concept.
A side effect is whichever alteration of objects and values derived from the execution of a statement, and which are not the direct expected effect.
int a = 0;
(void)b = ++a;
In the snippet above the void expression lose the direct effect, assigning b, but as side effect increase the value of a.
The only reference, explaining the meaning, in the standard can be found in 5.1.2.3 Program execution:
Accessing a volatile object, modifying an object, modifying a
file, or calling a function that does any of those operations are all
side effects, which are changes in the state of the execution
environment.
Evaluation of an expression in general includes both value
computations and initiation of side effects.
void is a type. Per C 2018 6.2.5 19, the type has no values (the set of values it can represent is empty), it is incomplete (its size is unknown), and it cannot be completed (its size cannot be known).
Regarding extern void a;, this does not define an object. It declares an identifier. If a were used in an expression (except as part of a sizeof or _Alignof operator), there would have to be a definition for it somewhere in the program. Since there cannot a definition of void object in strictly conforming C, a cannot be used in an expression. So I think this declaration is allowed in strictly conforming C but is not useful. It might be used in C implementations as an extension that allows getting the address of an object whose type is not known. (For example, define an actual object a in one module, then declare it as extern void a; in another module and use &a there to get its address.)
The declaration of functions with (void) as a parameter list is a kludge. Ideally, () might be used to indicate a function takes no parameters, as is the case in C++. However, due to the history of C, () was used to mean an unspecified parameter list, so something else had to be invented to mean no parameters. So (void) was adopted for that. Thus, (void) is an exception to the rules that would say (int) is for a function taking an int, (double) is for a function taking a double, and so on—(void) is a special case meaning that a function takes no parameters, not that it takes a void.
In foo((void) a), the cast does not make the value “not exist.” It converts a to the type void. The result is an expression of type void. That expression “exists,” but it has no value and cannot be used in an expression, so using it in foo((void) a) results in an error message.
From C Standard#6.2.5p19:
19 The void type comprises an empty set of values; it is an incomplete object type that cannot be completed.
This indicate that the void type exists.
Doubt 1:
void foo(void); // This function takes no argument. Not the 'void' type.
Correct.
From C Standard#6.7.6.3p10 [emphasis mine]:
10 The special case of an unnamed parameter of type void as the only item in the list specifies that the function has no parameters.
This is a special case they had to add to the language syntax because void foo(); already meant something different (void foo(); doesn't specify anything about foo's parameters). If it weren't for the old meaning of void foo();, void foo(); would have been the syntax to declare a no-argument function. You can't generalize anything from this. It's just a special case.
Doubt 2:
// Casting to 'void' should make the argument inexistant too...
foo((void)a);
No, it will not because void is also an object type though it is incomplete.
Doubt 3:
// Assigning to 'int' from incompatible type 'void': so the 'void' type does exists...
a = *p;
Yes, it does exist and hence the compiler is reporting error on this statement.
Doubt 4:
// Am I not passing the 'void' type ?
foo(*p);
Declaration of foo() function:
void foo(void);
^^^^
The void in parameter list indicates that function will not take any argument because it has been declared with no parameters.
Just for reference, check this from C Standard#5.1.2.2.1p1 [emphasis mine]:
1 The function called at program startup is named main. The implementation declares no prototype for this function. It shall be defined with a return type of int and with no parameters:
int main(void) { /* ... */ }
^^^^
Doubt 5:
extern void a; // Why is this authorised ???
This is authorized because void is a valid type and it is just a declaration. No storage will allocate to a.
In C, void can't be considered as a data type, it is a keyword used as a placeholder in place of a data type to show that actually there is no data. Hence this
void a;
is not valid.
while here
void foo(void);
void keyword is used to inform to the compiler that foo is not going to take any input argument nor it has return type.
In below case
int a = 42;
void *p;
a = *p; /* this causes error */
a = *p; is wrong because you can't dereference void pointer directly, you need to perform proper type casting first. for e.g
a = *(int*)p; /* first typecast and then do dereference */
Also this
foo(*p);
is wrong because of two reason,
firstly foo() doesn't expects any argument.
secondly you can't do *p as p is void pointer. Correct one is foo(*(int*)p); if foo() declaration is void foo(int);.
Note that this
(void)a;
doesn't do anything so your compiler might not giving any warning but when you do like
int b = (void)a;
compiler won't allow as void is not consider as data type.
Finally this
extern void a; // Why is this authorised ???
this is just a declaration not definition, a doesn't exist until you define it, since a is having extern storage class, you need to define somewhere & when you are going define like
a = 10;
compiler throws a error as
error: ‘a’ has an incomplete type
From C standard 6.2.5 Types
The void type comprises an empty set of values; it is an
incomplete object type that cannot be completed.
6.3.2.2 void
The (nonexistent) value of a void expression (an expression that has
type void) shall not be used in any way, and implicit or explicit
conversions (except to void) shall not be applied to such an
expression. If an expression of any other type is evaluated as a
void expression, its value or designator is discarded. (A void
expression is evaluated for its side effects.)
6.3.2.3 Pointers
A pointer to void may be converted to or from a pointer to any
object type. A pointer to any object type may be converted to a
pointer to void and back again; the result shall compare equal to the
original pointer.
A storage-class specifier or type qualifier modifies the keyword
void as a function parameter type list (6.7.6.3).
An attempt is made to use the value of a void expression, or an
implicit or explicit conversion (except to void) is applied to a
void expression (6.3.2.2).
First of all, it seems strange that you can declare a void object, meaning you just declare nothing.
void is an incomplete object type that cannot be completed. This mostly defines its uses in regular contexts, i.e. contexts that do not provide special treatment for void. Your extern declaration is one of such regular contexts. It is OK to use an incomplete data type in a non-defining declaration.
However, you will never be able to provide a matching definition for that declaration.
So the void in a prototype means nothing, and not the type void.
Correct. The parameter must be unnnamed. And the (void) combination is given special treatment: it is not one parameter of type void, but rather no parameters at all.
But then, I created a pointer to void, dereferenced it, and assigning the “result” to my int variable. I got the “incompatible type” error. That means the void type does exist here.
No. It is illegal to apply unary * operator to a void * pointer. Your code is invalid for that reason already. Your compiler issued a misleading diagnostic message. Formally, diagnostic messages are not required to properly describe the root of the problem. The compiler could've just said "Hi!".
Is void an actual type, or a keyword to means nothing ?
It is a type. It is an incomplete object type that cannot be completed.

In C, is it legal to add `const` only in function definitions, not declarations?

Is adding additional const specifiers to function arguments allowed by the standard, like in the following?
foo.h:
int foo(int x, char * data);
foo.c:
// does this match the prototype?
int foo(const int x, char * const data) {
// this implementation promises not to change x or move data inside the function
}
GCC accepts it with -std=c99 -Wpedantic -Wall -Werror, but that's not necessarily the same as standard-compliant.
This answer shows that the C++ standard allows this - does the C (99) standard allow this too?
There's another question here and a good answer here for C++
This is explicitly allowed by a special case in the rules for function parameter lists. N1570 §6.7.6.3p131 says:
In the determination of type compatibility and of a composite type, each parameter declared with function or array type is taken as having the adjusted type and each parameter declared with qualified type is taken as having the unqualified version of its declared type.
But you must also understand that the "unqualified version" of a type like const char * is still const char *, because the type const char * is derived from the type const char, and §6.2.5p26 says
A derived type is not qualified by the qualifiers (if any) of the type from which it is derived.
That means that the declaration
void foo (const int x);
is compatible with the definition
void foo (int x) { ... }
but the declaration
void bar (const char *x)
is not compatible with the definition
void foo (char *x) { ... }
You might be wondering why these rules are the way they are. The short version is that in C, all arguments are always passed by copying the value (but not any data pointed to by the value, if there are pointers involved), so it doesn't matter whether an actual argument is const T; the callee receives it as a regular old T regardless. But if you copy a pointer to constant data, the copy still points to constant data, so it does matter and that qualifier should be preserved.
1 Document N1570 is the closest approximation to the 2011 ISO C standard that is publicly available at no charge.
To the best of my knowledge, these rules have not changed significantly since the original 1989 standard. Pre-C89 "K&R" C didn't have prototypes, nor did it have const, so the entire question would be moot.
From the C99 spec, 6.7.5.3.15 Function declarators, this is legal:
If one type has a parameter type list and the other type is specified by a function definition that contains a (possibly empty) identifier list, both shall agree in the number of parameters, and the type of each prototype parameter shall be compatible with the type that results from the application of the default argument promotions to the type of the corresponding identifier. (In the determination of type compatibility and of a composite type, each parameter declared with function or array type is taken as having the adjusted type and each parameter declared with qualified type is taken as having the unqualified version of its declared type.)
cv-qualifiers on the parameter (and not on the type of the parameter) do not affect the type of the parameter, so do not affect the prototype. Things that don't affect the prototype like this can be different between the declaration and definition with no problem. Similarly, the name of the parameter does not affect the prototype, so can also be different between the declaration and definition.
Now what is confusing is that a const appearing here may be on the parameter or may be on the type of the parameter, depending on exactly where it is. If it is part of the type of the parameter, then it does affect the prototype, so must be consistent between the declaration and definition:
int foo(const int x); // const on the parameter
int foo(int * const x); // also const on the parater
int foo(const int *x); // const in the type, not on the parameter
int foo(int const *x); // also const in the type

Can a function pointer with a const argument be used as a function pointer with a nonconst argument?

Perhaps the title isn't clear in itself...
I have a function f (provided by some library) that takes as an argument a function pointer of signature void g(int*), i.e.
void f(void (*g)(int*));
However, I would like to use it using a function g (that I defined) with signature void g(const int*). A priori, I can't see how this can violate any const-correctness, as all the signature of f says is that g will only ever be called with a (non-const) int* (non-const), and indeed I can call a void (const int*) function with a non-const int* argument.
But GCC complains and says,
expected 'void (*)(int *)', but argument is of type 'void (*)(const int *)'
I can't see how this complaint can be legitimate, so does anyone know whether my understanding of that is wrong, or if there is a way around that?
You seem to have found something that the compiler writers and standards writers did not account for. From C99 draft n1256, §6.7.5.3 paragraph 15,
corresponding parameters shall have compatible types.
Note that const int * is not compatible with int *. However, int * may be converted to const int *. From §6.3.2.3, paragraph 2,
For any qualifier q, a pointer to a non-q-qualified type may be converted to a pointer to the q-qualified version of the type
More sophisticated rules for inferring when it is acceptable to substitute types derived from qualified or unqualified versions of the same type are simply not present in the standard. Therefore, your code is technically in violation of the standard.
My conclusion: It seems to me that this error should be treated as "pedantic" by the compiler: your code does not technically conform to the standard, but the meaning is unambiguous and the code is absolutely safe. Feel free to write a feature request to your compiler vendor. There are plenty of nonconformant practices which do not generate warnings without -pedantic.
As a final note, I compiled with Clang and the compiler informed me that the warning was pedantic. However, I had not requested pedantic warnings... so there appears to be no way to turn it off.
warning: incompatible pointer types passing 'void (int const *)', expected 'void (*)(int *)'
[-pedantic]
Workaround: Use an explicit cast.
void g(const int *);
f((void (*)(int *)) g);
You are right, there's no reason C should disallow that call (other than because the C standard says it should). U(*)(T*) should be a sub-type of U(*)(const T*) because int* is a sub-type of const int* through substitutability.
Why C does not allow this, I don't know.
As for work-arounds, you can provide a proxy function:
void foo(const int* x) { ... } // <-- the function you want to pass in
void bar(int* x) { foo(x); } // proxy
f(bar); // instead of f(foo)
The fact that using a safe, standard-compliant proxy like this works at all should be proof enough that the call should have been valid in the first place.

Resources