Thread: assigning a structure as a unit

> bar2 = *((struct foo2 *)&bar);
Not legal.
1. The cast causes out of bound memory accesses by pretending that the struct is larger than it really is.
2. Depending on the extra members of foo2, it may have different alignment requirements as well.

O_o

3. Even if the code was well-defined, you aren't making the point you think because specifically you told the compiler the assignment is of the same type.

That said, I'm pretty sure that would be the same as "out-of-bounds" access which I believe is "undefined-behavior".

[Edit]
I posted a second example to elaborate on what Salem posted.
[/Edit]

Soma

Code:

*(&bar); // this is an instance of `foo'
*((struct foo2 *)&bar); // you've told the compiler that this is an instance of `foo2'

Code:

#include <stdio.h>

struct foo {
    int a;
    int b;
};

struct foo2 {
    int a;
    int b;
    int extra;
};

void Dump(struct foo2 * bar2){printf("a = %d b = %d extra = %d\n", bar2->a, bar2->b, bar2->extra);}

int main(void)
{
    int maybe[1] = {44};
    {
        struct foo bar = { .a = 42, .b = 43 };
        Dump((struct foo2 *)&bar);
        /* bar2->extra is set according to what memory */
        bar.a = 46; /* maybe */
    }
    {
        int bar[2] = {44, 45};
        Dump((struct foo2 *)&bar);
        /* bar2->extra is set according to what memory */
    }
    return 0;
}

A different, and possibly more realistic example. Perhaps this avoids alignment issues.

O_o

The example may avoid alignment issues, but the example still has "out-of-bounds" access.

The compiler doesn't "see" a `foo' structure having two integers; the compiler "sees" a `foo2' structure which has three integers; the compiler copies three integers.

As before, the situation you ask about isn't a "different types" situation as far as the compiler is concerned. Yes. The types are actually different, but you told the compiler that the types are the same, and so the compiler does exactly what it would do if the types actually were the same.

[Edit]
I think I possible see the problem.

Code:

printf("a = %d b = %d\n", bar2.a, bar2.b);

The use of `bar2' after being assigned is irrelevant.

Code:

bar2 = *((struct foo *)&bar);

The use of `bar' is relevant. The compiler assigns a value to `bar2.extra' because it is used with the assignment line. The "out-of-bounds" access happens because the compiler assigns `bar2.extra' from `((struct foo *)&bar)->extra' which does not exist.
[/Edit]

Soma

Code:

#include <stdio.h>

struct foo {
    int a;
    int b;
};

struct foo2 {
    struct foo link;
    int extra;
};

void Dump(struct foo2 * bar2){printf("a = %d b = %d extra = %d\n", bar2->link.a, bar2->link.b, bar2->extra);}

int main(void)
{
    struct foo2 bar2;
    int maybe[1] = {44};
    {
        struct foo bar = { .a = 42, .b = 43 };
        Dump((struct foo2 *)&bar);
        /* bar2->extra is set according to what memory */
        bar.a = 46; /* maybe */
    }
    {
        int bar[2] = {44, 45};
        Dump((struct foo2 *)&bar);
        /* bar2->extra is set according to what memory */
    }
    return 0;
}

Why would the compiler copy three integers after I've cast it to struct foo which only has two integers?

O_o

I see that you've changed the types of `bar' and `bar2' on me.

I expect the compiler would only copy two integers

In the current code from post #6, the compiler would only copy two integers because both types used in the assignment are `foo' which only has two integers.

therefore .extra would remain uninitialised... so, Dump() in your example would be accessing uninitialised memory, and that's all, as far as I can see

You are wrong; You swapped the types of `bar' and `bar2'; I did not.

Soma

I think we were just out of sync. I swapped them in my 2nd example and it seems that you were working on your post before you saw that.

O_o

Yep.

*shrug*

It happens.

Soma

It does. We need some kind of resolve conflict mechanism like git.

O_o

I like the idea; even a simple "some posts were edited since you hit reply" would be handy.

If you hit "Reply With Quote", I can see an AJAX event triggered for specific posts.

o_O

Well, that may--been a long time--not be possible with the "full weight" of "vBulletin"; the tagging over "has been edited" is often loose--up to several minutes if no one has visited the page.

Still, I like the idea enough to toss the suggestion to a "Drupal" forum developer I know.

Soma