Applied duck-typing in plain C

Question

In my open-source plain C code I use this simple structure to read and parse data from a string buffer:

typedef struct lts_LoadState
{
  const unsigned char * pos;
  size_t unread;
} lts_LoadState;

The buffer is accessed with this simple API:

/* Initialize buffer */
void ltsLS_init(lts_LoadState * ls,const unsigned char * data, size_t len);

/* Do we have something to read? (Actually a macro.) */
BOOL ltsLS_good(ls);

/* How much do we have to read? (Actually a macro.) */
size_t ltsLS_unread(ls);

/* Eat given number of characters, return pointer to beginning of eaten data */
const unsigned char * ltsLS_eat(lts_LoadState * ls, size_t len);

Note: ltsLS_unread may be replaced with return (ltsLS_good(ls)) ? SIZE_MAX : 0 without breaking the current implementation.

This code is used to load some data in a custom format from a string buffer. (This may be a better illustration.)

---

Now I need to load data not from a string buffer, but from a FILE pointer.

I would hate to copy-paste the implementation, and would like to reuse existing code instead. (I'm OK with refactoring/adapting it, of course.)

This is a textbook stuff in C++, but how to do that in plain C without incurring runtime overhead?

---

Here is an example function that uses the lts_LoadState API and that is not to be copy-pasted (but may be changed, of course, to support both string buffer and FILE *):

static int ltsLS_readline(
    lts_LoadState * ls,
    const unsigned char ** dest,
    size_t * len
  )
{
  const unsigned char * origin = ls->pos;
  unsigned char last = 0;
  size_t read = 0;

  while (ltsLS_good(ls))
  {
    if (ltsLS_unread(ls) > 0)
    {
      unsigned char b = *ls->pos; /* OK, this should be ltsLS_eat_char macro. */
      ++ls->pos;
      --ls->unread;

      if (b == '\n')
      {
        *dest = origin;
        *len = (last == '\r') ? read - 1 : read;

        return LUATEXTS_ESUCCESS;
      }

      last = b;
      ++read;
    }
    else
    {
      ls->unread = 0;
      ls->pos = NULL;
    }
  }

  return LUATEXTS_ECLIPPED;
}

Answer 1

It sounds like you want function variables, which you would pass as a parameter. C can do them, but the syntax isn't very pretty.

There probably is a bit of runtime overhead, but not much.

How do you pass a function as a parameter in C?

Answer 2

I hate to open this back up but this is something I was thinking about today and I don't think this has a great answer yet.

I think to implement duck typing in C what you're after is a global vtable. Every struct (object) should have the vtable as the first element in the struct. Basically whenever there's a behaviour you want to access through duck typing you would add it to this global vtable; then you can call it no matter what object is passed to your function, you'd be able to cast the object to the table, look to the location the behaviour should be, check it's non-null, and call it.

//Would be declared in some global.h or similar
struct global_v_table_t = 
{
    char* (*toString)(void);
    //... other functions being accessed through duck typing go here
}

//--------------------
//In some other files: 
//Then we create some objects:
struct bicycle_t
{
    struct global_v_table;
    void (*ride)(void);
};

//When we initialise a bicycle
bicycle_t * bicycyle_init(void)
{
    bicycle_t * bike = malloc(sizeof(bicycle_t));
    //Req'd basically for every object within the project:
    //Either do this or call calloc() instead of malloc():
    globalVtableInit((global_v_table_init)bike);//NULL the vtable
    //Set the behaviours that this object exhibits:
    bike->global_v_table.toString = BIKE_toString;    
}


static char * bikeString = "I'm a bike!";
char * BIKE_toString(void)
{
    return bikeString;
}

//----------------

//Now anyone can ask that an object provide it's toString:

//The example uses an error logging function:

void logError(void * obj)
{
    char * (toStringMethod)(void) = ((global_v_table *)obj)->toString;
    if (NULL != toStringMethod)
    {//As long as the object implements the toString behaviour: 
        printf(toStringMethod()); //Print the object's toString.
    }
}

//Will tidy this code up a bit later but this is what I'm thinking.
//Hopefully is at least partly understandable. The obvious drawback
//to this implementation is that for every object you get this massive
//v_table which is full of mostly NULL's for each object as it scales.
//If you want to make C behave like other languages though you have
//to expect some sort of penalty I guess...