Thread: `struct declared inside parameter list' problem, seems ineluctable in such a case

Hi, I got this warning message when compiling a project under MinGW using gcc. Now, to simplify, I'll try to give an example:

There are 3 files included in this problem, which belong to a large project:

1) a.h
2) b.h
3) test.c

a.h:

Code:

#ifndef A_H
#define A_H

#include "b.h"

struct A
{
	int a;
	int b;
};

void a_func(struct A *a, struct B *b);

#endif

As you can see, it declares a function named `a_func', which accepts a pointer to `struct A', and a pointer to `struct B' (so I included `b.h'）. `a_func' might be defined in `a.c' and used elsewhere, which is not the problem.

b.h:

Code:

#ifndef B_H
#define B_H

#include "a.h"

struct B
{
	double i;
	double j;
	struct A *a;
};

#endif

`struct B' contains a pointer to `struct A', so I included `a.h'.

and test.c:

Code:

#include <stdio.h>
#include "b.h"

int main(void)
{
	struct B b;

	b.i = 2.71;
	b.j = 3.14;
	printf("i is %f and j is %f\n", b.i, b.j);
	
	return 0;
}

Now compile test.c:

Code:

gcc -Wall -c test.c

it says:

Code:

In file included from b.h:4,
                 from test.c:2:
a.h:12: warning: "struct B" declared inside parameter list
a.h:12: warning: its scope is only this definition or declaration, which is probably not what you want

***

Now I understand that it is because after we included `b.h' in `test.c', `B_H' has been defined, and since `b.h' includes `a.h' while `a.h' includes `b.h' also (then `b.h' is actually not included in `a.h' because `B_H' has already been defined), the definition of `struct B' is not known by `a.h'.

I also understand that, if i put `void a_func(struct A *a, struct B *b);' into `b.h', this problem is solved in this case, but this breaks the organization of this project since `a_func' is defined in `a.c', which is supposed to be with `a.h'.

So if anyone has any idea to keep the declaration of `a_func' in `a.h' and meanwhile avoid the `struct declared inside parameter list' problem, I'll be very grateful.

You can do without the definition of struct B in a.h, so in fact you could remove the inclusion of b.h, replacing it with a forward declaration:

Code:

struct B;

Likewise, the definition of struct B does not need the definition of struct A since struct B stores a pointer to struct A. As such, you can remove the inclusion of a.h from b.h, replacing it with a forward declaration of struct A.

Originally Posted by babyifan

If we remove the inclusion of a.h from b.h, replacing it with a declaration of struct A, then how does GCC know the size of struct A when we malloc() space to `a' (the pointer to struct A)?

You would include the relevant header files in the source files instead.

Originally Posted by laserlight

You would include the relevant header files in the source files instead.

Sorry, I still can't get the point.

I mean, in this minimized case, when a.h needs definition of some type in b.h, at the same time b.h needs definition of some type in a.h, but there were conditional directives guard in both headers:

Code:

#ifndef HEADER
#define HEADER

...

#endif

and when i include one header in a C file, the side effect of the conditional directives guard appears: in my example, a.h won't include the content of b.h at all because B_H has already been defined, and the pitiful declarations in a.h that needs definitions in b.h will never get those definitions.

I know I can solve this problem by moving the declaration of a_func to b.h, or moving the definition of struct B to a.h, or by using only one header for this problem, but it seems i'm a mysophobia in coding and src structure, so i do want to know if there exists other solutions.

Originally Posted by babyifan

I mean, in this minimized case, when a.h needs definition of some type in b.h, at the same time b.h needs definition of some type in a.h

No, a.h does not need the "definition of some type in b.h". a.h contains the definition of struct A, which does not involve struct B at all. It also contains a declaration of a_func, which does involve struct B, but only needs a declaration, not definition, of struct B. b.h does not need the "definition of some type in a.h", because struct B is defined in b.h, but its use of struct A involves a pointer to struct A, not a struct A object, as a member. Therefore, a declaration of struct A will suffice.

Originally Posted by babyifan

and when i include one header in a C file, the side effect of the conditional directives guard appears: in my example, a.h won't include the content of b.h at all because B_H has already been defined, and the pitiful declarations in a.h that needs definitions in b.h will never get those definitions.

Yes, but if you do it the way that I am recommending, then you will #include "b.h" in the source file where you define a_func (along with #include "a.h", presumably).

Hi laserlight, I've tried your solution but it seems not to work. If we remove the inclusion of a.h from b.h, replacing it with a declaration of struct A, then how does GCC know the size of struct A when we malloc() space to `a' (the pointer to struct A)? The type of struct A is incomplete because after removing a.h, we don't know where to get definition of struct A, dereferencing pointer `a' will be impossible.

That's right. C compilers will accept forward references to structures when you only have pointers to the structure. So if you have a function prototype in a header file which involves pointers to a structure, you can get away with a forward declaration.

When you go to implement this function in a source file, then you must #include the header file to get the full structure declaration. At this point, however, the prototype has already been accepted with the forward declaration, so the recursive inclusion problem doesn't matter.

Try it. Use a forward declaration to define struct B in a.h. Then in a.c, #include "b.h". All should be well.

Today I checked out Forward declaration - Wikipedia, the free encyclopedia , and I know that I've learned new things. laserlight, thanks for your patience and dwks, thanks for your clear explanation.