为什么 GCC 只有在使用结构名称而不是 typedef 时才会抛出错误和不必要的警告？

Question

I have a program consisting of two source files (farm.c, init.c) and two corresponding header files (farm.h, init.h) Both source files contain header guards and each other, because they both require functions/variables from each other.

init.h:

#ifndef INIT_H
#define INIT_H

#include<stdio.h>
#include<stdlib.h>
#include"farm.h"

#define PIG_SOUND "oink"
#define CALF_SOUND "baa"

enum types {PIG, CALF};

typedef struct resources {
    size_t pork;
    size_t veal;
    size_t lamb;
    size_t milk;
    size_t eggs;
} resources;

typedef struct animal {
    size_t legs;
    char* sound;
    int efficiency;
    void (*exclaim)(struct animal*);
    void (*work)(struct animal*, struct resources*);
} animal;

/* I have tried various ways of declaring structs in addition to
   the typedef such as this */

//animal stock;
//animal farm;

void make_pig(struct animal* a, int perf);
void make_calf(struct animal* a, int perf);

#endif

farm.h:

#ifndef FARM_H
#define FARM_H

#include<stdio.h>
#include<stdlib.h>
#include<time.h>
#include<string.h>
#include"init.h"

/* GCC does not recognise the typedef or struct identifier 
   until these forward declarations have been made in 
   addition to the included init.h header file */

//typedef struct animal animal;
//typedef struct resources resources;

void exclaim(animal* b);
void work(struct animal* b, struct resources* s);

#endif

init.c:

#include"init.h"

void make_pig(animal* a, int perf) {
    a->legs = 4;
    a->sound = PIG_SOUND;
    a->efficiency = perf;
    a->exclaim = exclaim;
    a->work = work;
}

void make_calf(animal* a, int perf) {
    a->legs = 4;
    a->sound = CALF_SOUND;
    a->efficiency = perf;
    a->exclaim = exclaim;
    a->work = work;
}

farm.c:

#include"farm.h"

int main() {
    return 0;
}

void exclaim(animal* a) {
    for (int i = 0; i < 3; i++) {
        printf("%s ", a->sound);
    }
    printf("\n");
}

void work(animal* a, struct resources* r) {
    if (!strcmp(a->sound, PIG_SOUND)) {
        r->pork += a->efficiency;
    }

    if (!strcmp(a->sound, CALF_SOUND)) {
        r->veal += a->efficiency;
    }
}

Using both types of names (ie, struct ani and animal ) normally works perfectly fine on my Linux system with the C99 standard. However when I use struct ani instead of animal here, I get the below warnings for every instance of the type usage for struct ani and struct resources .

lib/farm.h:10:21: warning: ‘struct ani’ declared inside parameter list will not be visible outside of this definition or declaration
   10 | void exclaim(struct ani* a);
      |                     ^~~
lib/farm.h:11:33: warning: ‘struct resources’ declared inside parameter list will not be visible outside of this definition or declaration
   11 | void work(struct ani* a, struct resources* r);

And 10 warnings in total for every usage of function pointers of the form:

src/init.c:17:16: warning: assignment to ‘void (*)(struct ani *)’ from incompatible pointer type ‘void (*)(struct ani *)’ [-Wincompatible-pointer-types]
   17 |     a->exclaim = exclaim;
      |                ^
src/init.c:18:13: warning: assignment to ‘void (*)(struct ani *, struct resources *)’ from incompatible pointer type ‘void (*)(struct ani *, struct resources *)’ [-Wincompatible-pointer-types]
   18 |     a->work = work;

Can someone please explain why such behaviour occurs and how I can avoid problems? It typically takes me an unfeasible amount of time to solve these errors and I still don't truly understand my mistake in the first place.

Answer 1

You've hit one of the odd corner cases of C scoping rules.

Informally, a tagged struct (or union , but I'm not going to repeat that over and over) springs into existence when it is named if no declaration for it is visible. "Springs into existence" means that it is considered declared in the current scope . Also, if a tagged struct was previously named in a scope and you then declare a struct with the same tag, the two structs are considered the same. Until a declaration for the struct is completed, the struct is considered an incomplete type, but a pointer to an incomplete type is a complete type, so you can declare a pointer to a tagged struct before you actually complete the definition of the struct.

Most of the time, that just works with minimal thought. But function prototypes are a bit special, because a function prototype is a scope, all by itself. (The scope lasts only until the end of the function declaration.)

When you put that together, you end up with the issue you're facing. You cannot use a pointer to a tagged struct in a function prototype unless the tagged struct was known before the function prototype appears. If it had been mentioned before, even in an outer scope, the tag is visible and therefore will be considered to be the same struct (even if it is still incomplete). But if the tag was not previously visible, a new struct type will be created within the prototype scope , which will not be visible after that scope ends (which is almost immediately).

Concretely, if you write the following:

extern struct animal * barnyard;
void exclaim(struct animal*);

then the two uses of struct animal refer to the same struct type, which will presumably be completed later (or in another translation unit).

But without the extern struct animal * barnyard; declaration, the struct animal named in the exclaim prototype is not previously visible, so thus is declared only in the prototype scope, so it is not the same type as some subsequent use of struct animal . Had you put the declarations in the opposite order, you would have seen a compiler warning (assuming you'd asked for compile warnings):

void exclaim(struct animal*);
extern struct animal * barnyard;

( On godbolt, bless it's heart )

A typedef declaration performs the same way as the extern declaration above; the fact that it is a type alias is not relevant. What's important is that the use of struct animal in the declaration causes the type to spring into existence, and you can subsequently use it freely in a prototype. That's the same reason that the function pointers inside your struct definitions are OK; the start of the struct definition was sufficient to cause the tag to be declared, so the prototype sees it.

In fact, any syntactic construction which contains a tagged struct ( struct whatever ) will serve the same purpose, because what matters is the effect of mentioning a tagged struct with no visible declaration. Above, I used extern global declarations as examples because they are lines which might appear in a header, but there are many other possibilities, including even the declaration of a function which returns a pointer to a struct (because the return type of a function declaration is not in the prototype scope).

See below for some additional comments about the edited question.

---

My personal preference is to always use typedefs as forward declarations of tags, and never use struct foo anywhere in my code other than the typedef and the subsequent definition:

typedef struct Animal Animal;

void exclaim(Animal*);

// ...

// Later or in a different header
struct Animal {
  Animal* next;
  void (*exclaim)(Animal *);
  // etc.
};

Note that I always use the same identifier for the tag and the typedef. Why not? There's no confusion and tags have not been in the same namespace as other identifiers since C was premordial.

For me, a big advantage of this style is that it lets me separate implementation details; the public header only contains the typedef declarations (and prototypes which use that type) and only the implementation needs to contain the actual definitions (after first having included the public header).

---

Note: since this answer was written, the question was edited to add a more detailed code sample. For now, I'll just leave these additional notes here:

On the whole, you get better answers when you provide better information. Since I couldn't see your actual code, I did the best I could, which was to try to explain what is going on, leaving you to apply that to your actual code.

In the code you have now added to the question, there is a circular header dependency. These should be avoided; it's almost impossible to get them right. The circular dependency means that you don't control the order of inclusion, so a declaration in one header might not come before the use in another header. You no longer have a forward declaration, because, depending on the inclusion order, it might be a backward declaration.

To resolve the circular dependency, abstract out the shared components and put them in a new header file. Here, forward declarations of structs (using, for example, typedefs) are highly useful because they don't depend on anything used in the definition of the struct. A shared header might include only typedefs, or it might also include prototypes which don't require additional dependencies.

Also, avoid putting long lists of library includes in your header files; include only those headers actually necessary to define types actually used in the header.