clang-check wrongly adds forward definitions

Clang Frontend
1

Hello,

I'm using clang-check (combined with cproto) to fix old K&R
code that needs to compile with a compiler that doesn't like
that.

I have an issue with the way it adds an identifier forward
definition with it's not yet defined, e.g.:

#include <stddef.h>

struct foo {
    int a;
    struct s* b;
};

int main(void) {
    struct foo f;

    f.b = NULL;

    return 0;
}

`struct s` was not defined, but this is not a problem for a pointer definition.
Then I run (v4.0.1):

$ clang-check -ast-print -fix-what-you-can in.c
typedef long ptrdiff_t;
typedef unsigned long size_t;
typedef int wchar_t;
typedef struct {
    long long __clang_max_align_nonce1 __attribute__((aligned(_Alignof(long long))));
    long double __clang_max_align_nonce2 __attribute__((aligned(_Alignof(long double))));
} max_align_t;
struct foo {
    int a;
    struct s;
    struct s *b;
};
int main() {
    struct foo f;
    f.b = ((void *)0);
    return 0;
}

The `struct s` right in the middle of the definition of `struct foo` confuse
all compilers. My understanding of the C standard is that this definition
should be either *before* the `struct foo` definition, or not exist at all
(which I think is the usual way to do it). Am I missing some reason it would
appear there?

I'll see if I manage to fix that by myself if no-one is interested :slight_smile:

Cheers,
Y.

2

Hello,

I'm using clang-check (combined with cproto) to fix old K&R
code that needs to compile with a compiler that doesn't like
that.

I have an issue with the way it adds an identifier forward
definition with it's not yet defined, e.g.:

#include <stddef.h>

struct foo {
    int a;
    struct s* b;
};

int main(void) {
    struct foo f;

    f.b = NULL;

    return 0;
}

`struct s` was not defined, but this is not a problem for a pointer
definition.
Then I run (v4.0.1):

$ clang-check -ast-print -fix-what-you-can in.c
typedef long ptrdiff_t;
typedef unsigned long size_t;
typedef int wchar_t;
typedef struct {
    long long __clang_max_align_nonce1 __attribute__((aligned(_Alignof(long
long))));
    long double __clang_max_align_nonce2 __attribute__((aligned(_Alignof(long
double))));
} max_align_t;
struct foo {
    int a;
    struct s;
    struct s *b;
};
int main() {
    struct foo f;
    f.b = ((void *)0);
    return 0;
}

The `struct s` right in the middle of the definition of `struct foo`
confuse
all compilers. My understanding of the C standard is that this definition
should be either *before* the `struct foo` definition, or not exist at all
(which I think is the usual way to do it). Am I missing some reason it
would
appear there?

Putting it there would be right for C++, which allows nested record types.
It's possible that the clang-check code doesn't take account of C's rather
different rules in this area.

I'll see if I manage to fix that by myself if no-one is interested :slight_smile:

That'd be great. I don't know whether it makes sense to have one piece of
functionality that tries to handle both C and C++, or to separate them
out. That might depend on how much they have in common.

-- James