Thread: dynamicly adding data members to structs

C only please, no C++.

My eventual goal is to be able to translate Python into C to be compiled by the C compiler. As I see it there's two steps, create a dynamic typing library for C, and create the actual translator. I'm already working on a basic dynamic type and the results so far are good.

Here's the problem, in python the data in classes are created dynamicly, there's no way to predefine them. Here's some example Python code:

Code:

#create an empty class named "person"
class person:
    pass
#create a "person" instance
some_name = person()
#assign data members to "some_name"
some_name.name = "Homer Simpson"
some_name.age = 39 #this is a guess, I'm not sure how old he is
some_name.gender = "M"
#print the newly assigned data
print some_name.name, some_name.age, some_name.gender
Homer Simpson 39 M

Python handles those dynamic class members (and all variables) with string-based hash tables. I wish to accomplish the same thing in C without the hash tables if possible. If I have to use hash tables could I use memory addresses instead of strings? Integer comparisons are much faster than string comparisons. Any ideas on how I can do this without hash tables?

Thanks in advance.

I if have to use hash tables could I use memory addresses instead of strings? Integer comparisons are much faster than string comparisons.

Huh? If you're thinking what I think you're thinking, then no. You should key on the string, not the address where the string is located. If you did key on the address, how am I supposed to do a lookup into this hashtable? I have the string I want, which may or may not exist at the same address as the string you have in the table. Key on the string. Look into hashes if you absolutely have to, but only after profiling says the hash table is a bottleneck. It's a hash table. If you're Doing It Right™, then it's a single comparision. Maybe two or three in the worst case.

Originally Posted by Cactus_Hugger

Huh? If you're thinking what I think you're thinking, then no. You should key on the string, not the address where the string is located. If you did key on the address, how am I supposed to do a lookup into this hashtable? I have the string I want, which may or may not exist at the same address as the string you have in the table. Key on the string. Look into hashes if you absolutely have to, but only after profiling says the hash table is a bottleneck. It's a hash table. If you're Doing It Right™, then it's a single comparision. Maybe two or three in the worst case.

Hashtable is not the right word if I do use the memory addresses. Let's say my struct has variables a, b, and c. My struct has a list of pointers to the data members, a, b, and c to store the memory addresses. Let's say I want to add variable d to the struct, I would go through the list of variable pointers and compare the address of d to the addresses of a, b, and c. If I got a match then the variable d is already in the struct and I would throw an error, if i don't get a match then I add d the address of d to the list. I would then follow a similar protocall for all operations related to these variables.

What I really want is a way to do this without hashtables. (or address lists)

Thanks for the speedy reply though.

> I would go through the list of variable pointers and compare the address of d to the addresses of a, b, and c.
That's not going to tell you much at all. You still want to compare the values of at the addresses. i.e. *a == *d

If you don't want to use a hash table, then consider a binary tree of people. Or even just a sorted list of people.

Your explanations are very blurry, considering you basically explained (not legal C):

Code:

struct x
{
   int a, b, c;
   int * a, * b, * c;
}

And you want to add 'd' to the struct.

You mean, given a list of people how do I find out if member x in any of the people has a value of d?

I would also be concerned, that there have been a few "Simpsons" questions posted here, and there is a new member called "Homer_Simpson" that has been replying to them. So you're probably being watched

How about:

Code:

struct dyn_obj {
      void **array;  /* array of addresses */
      char *names[]; /* parallel array of variable descriptors */
      int *types; /*pointer to parallel array of variable types */
      int length; /* length of array
};

Originally Posted by ITAmember

I've got the dynamic object thingy down, I want to be able to add objects to a struct without predeclaring them. I'm not sure if that's what the code is as it looks like a dynamic object type, not a way to implement a dynamic struct.

Just an idea. Whatever you want to call it, you could use it to dynamically add pointers to a struct, my impression was that was the goal, there would have to be a function like:

Code:

int (struct dyn_obj *ptr, void *data_addr, char *name, int type); /* returns success or error */

Using some realloc to expand the struct to hold new members whenever.

I kind of see what you mean now, I've been giving it some thought and this is what I came up with.

Here's my dynamic object definition code:

Code:

typedef struct
{
    int memory_address; /* stores the memory address of the current object, accessed/mutated via assembly */
    int type; /* stores the type of the object, this may be removed as it might not be needed */
} object;

The memory_address value stores the address of some object type defined by a struct. All those object types need to have data members and methods (function pointers) added to them at runtime, without predeclaring them. Here's an example:

Code:

typedef struct
{
    linked_list_of_object_addresses data;
} dynamic_struct;

This will be the base of all objects, built in types will have the linked_list_of_object_addresses filled in with some data members and methods before the program starts. Here's a function that would add a data member if it's not already in the list:

Code:

 int append(object &a, object &b) /* a is the object to append to, b is the object to append */
{
    for(int i = 0; i < len(*a); i++) /* len returns the length of linked_list_of_object_addresses */
    {
        if(b = (*a)[i])
        {
            return 0;
        }
    (*a).linked_list_of_object_addresses.add(b);
    return 1;
}

I know it's pseudocode but the idea would work right?

Originally Posted by ITAmember

I know it's pseudocode but the idea would work right?

Good question. I should admit I don't know python and I've never done anything like this. I would have separate lists for "data members" and "methods", esp. considering you might want two different styles of node for them.

Not sure why you're worried about adding data members that already exist; if this has to do with your own pre-runtime set-up (ie, of built in classes and methods), I don't think this method of keeping your tish sorted inspires confidence If it is to prevent a user from doing the same, I think you should let them rather than incorporate (performance reducing) error checks to handle bad programming.

On the topic of optimization, a lil' note:

Code:

for(int i = 0; i < len(*a); i++)

Be aware that "len(*a)" will get called EVERY time the loop iterates, not just once. So, since AFAICT that value does not change during the loop, call it once and put it into a variable:

Code:

int x = len(*a); 
for(int i = 0; i < x; i++)