I am dynamically allocating a struct which has a different struct as a member:
struct a {
// other members
struct b;
}
struct b
The difference is really equivalent to any other situation where you're comparing "automatic" and "dynamic" allocation.1
In terms of a guide as when you should use a pointer member, I would say that you should avoid it unless there's a good reason not to, due to the programmer overhead in dealing with manual memory management (and the bugs that it inevitably leads to).
An example of a good reason would be if you needed your struct a to refer to an existing struct b.
If you dynamically allocate (malloc) struct a as in
struct a *temp = (struct a *)malloc(sizeof(struct a));
you malloc space for a pointer to struct b (assuming that's what is in struct a) but you don't malloc space for struct b. That means later you'll have to do
temp->b = (struct b *)malloc(sizeof(struct b));
before you try and use struct b.
If you don't store a pointer to struct b but rather struct b directly then you'll get the automatic allocation when you define struct a.
First, let's get some real definitions in place to make this concrete.
struct b {
int x;
};
struct a_with_b {
struct b b;
}
struct a_with_b_ptr {
struct b *bp;
}
When you encapsulate a struct, you just need to allocate the outer struct (and since the inner struct is not a pointer, you use . to reference members of the innert struct):
struct a_with_b *a1 = malloc(sizeof(struct a_with_b));
a1->b.x = 3;
But when you encapsulate a pointer, you have to allocate each independently and use -> when referencing members of the inner struct:
struct a_with_b_ptr *a2 = malloc(sizeof(struct a_with_b_ptr));
a1->b = malloc(sizeof(struct b));
a1->b->x = 3;