pocketpy/docs/bindings.md

icon	title	order
cpu	Write C Bindings	18

In order to use a C/C++ library in python, you need to write bindings for it.

pkpy uses an universal signature to wrap a C function pointer as a python function or method, i.e py_CFunction.

typedef bool (*py_CFunction)(int argc, py_Ref argv);

• argc is the number of arguments passed to the function.
• argv is the pointer to the first argument.

If successful, the function should return true and set the return value in py_retval(). In case there is no return value, you should use py_newnone(py_retval()). If an error occurs, the function should raise an exception and return false.

Steps

Bind a simple function

Say you have a function add that takes two integers and returns their sum.

int add(int a, int b) {
    return a + b;
}

Here is how you can write the binding for it:

// 1. Define a wrapper function with the signature `py_CFunction`.
bool py_add(int argc, py_Ref argv) {
    // 2. Check the number of arguments.
    PY_CHECK_ARGC(2);
    // 3. Check the type of arguments.
    PY_CHECK_ARG_TYPE(0, tp_int);
    PY_CHECK_ARG_TYPE(1, tp_int);
    // 4. Convert the arguments into C types.
    int _0 = py_toint(py_arg(0));
    int _1 = py_toint(py_arg(1));
    // 5. Call the original function.
    int res = add(_0, _1);
    // 6. Set the return value.
    py_newint(py_retval(), res);
    // 7. Return `true`.
    return true;
}

Once you have the wrapper function, you can bind it to a python module via py_bindfunc.

py_GlobalRef mod = py_getmodule("__main__");
py_bindfunc(mod, "add", py_add);

Alternatively, you can use py_bind with a signature, which allows you to specify some default values.

py_GlobalRef mod = py_getmodule("__main__");
py_bind(mod, "add(a, b=1)", py_add);

Bind a struct

If you have a struct like this:

typedef struct MyStruct{
    int x;
    int datasize;
    int* data;
}MyStruct;

x is some kind of property of the struct, and this struct is used for store datasize numbers.

Here's how you can create a MyStruct:

// 1. Define a wrapper function with the signature `py_CFunction`.
bool MyStruct__new__(int argc, py_Ref argv) {
    // 2. Check the number of arguments.
    PY_CHECK_ARGC(3);
    // 3. Check the type of arguments.
    PY_CHECK_ARG_TYPE(0, tp_type);
    PY_CHECK_ARG_TYPE(1, tp_int);
    PY_CHECK_ARG_TYPE(2, tp_int);
    // 4. Convert the arguments into C types.
    py_Type cls = py_totype(py_arg(0));
    int x = py_toint(py_arg(1));
    int datasize = py_toint(py_arg(2));
    // 5. Create a MyStruct instance, where `datasize` gives correspond slots to store numbers.
    MyStruct* res = py_newobject(py_pushtmp(), cls, datasize, sizeof(MyStruct));
    // 6. Set the values.
    res->x = x;
    res->datasize = datasize;
    // 7. `data` is in the head of slots, init `data` with zeros.
    res->data = py_getslot(py_peek(-1), 0);
    for (int i = 0; i < datasize; i++) {
        res->data[i] = 0;
    }
    // 8. Put the created struct into the return value register.
    py_assign(py_retval(), py_peek(-1));
    // 9. Pop the struct safely.
    py_pop();
    return true;
}

Function for getting the property x from MyStruct:

bool MyStruct_x(int argc, py_Ref argv) {
    // 1. Check the number of arguments.
    PY_CHECK_ARGC(1);
    // 2. Convert the arguments into C types.
    MyStruct* self = py_touserdata(argv);
    // 3. Set the x value.
    py_newint(py_retval(), self->x);
    // 4. Return `true`.
    return true;
}

Function for getting a specified number from data:

bool MyStruct_data_get(int argc, py_Ref argv) {
    // 1. Check the number and type of arguments.
    PY_CHECK_ARGC(2);
    PY_CHECK_ARG_TYPE(1, tp_int);
    // 2. Convert the arguments into C types.
    MyStruct* self = py_touserdata(argv);
    int index = py_toint(py_arg(1));
    // 3. Exception if the index is out of range.
    if (index >= self->datasize) {
        IndexError("Not a valid index");
    }
    // 4. Return the value.
    py_newint(py_retval(), self->data[index]);
    // 5. Return `true`.
    return true;
}

Function for setting a number's value in data:

bool MyStruct_data_set(int argc, py_Ref argv) {
    // 1. Check the number and type of arguments.
    PY_CHECK_ARGC(3);
    PY_CHECK_ARG_TYPE(1, tp_int);
    PY_CHECK_ARG_TYPE(2, tp_int);
    // 2. Convert the arguments into C types.
    MyStruct* self = py_touserdata(argv);
    int index = py_toint(py_arg(1));
    int value = py_toint(py_arg(2));
    // 3. Exception if the index is out of range.
    if (index >= self->datasize) {
        IndexError("Not a valid index");
    }
    // 4. Set the value.
    self->data[index] = value;
    // 5. All functions should have a return value. None is returned here.
    py_newnone(py_retval());
    // 6. Return `true`.
    return true;
}

Now you can bind the functions to the new module mmystruct:

py_GlobalRef mod = py_newmodule("mystruct");
// 1. Add a custom type.
py_Type mystruct = py_newtype("custom_struct", tp_object, mod, NULL);
// 2. Bind the function of creating MyStruct.
py_bind(py_tpobject(mystruct), "__new__(cls, x: int, datasize: int)", MyStruct__new__);
// 3. Bind the property `x`.
py_bindproperty(mystruct, "x", MyStruct_x, NULL);
// 4. Bind magic methods of operating numbers in `data`.
py_bindmagic(mystruct, __getitem__, MyStruct_data_get);
py_bindmagic(mystruct, __setitem__, MyStruct_data_set);

You can use it like this:

import mystruct
test = mystruct.custom_struct(3,4) # x=3, 4 slots for data
print(test.x)
print(test[1]) # 0
test[1] = 100
print(test[1]) # 100

Bind a function with arbitrary argument lists

Sometimes you want a function that takes arbitrary input arguments. For example, sum several numbers in the table, or make a simple print function.

Sum several numbers

Say you have 2,3,4,5,6 and put them into the sum function. Here's an implementation:

bool py_sum(int argc, py_Ref argv) {
    // 1. These numbers are packed as a tuple
	PY_CHECK_ARG_TYPE(0, tp_tuple);
    // 2. Get the length of the tuple
	int len = py_tuple_len(py_arg(0));
	int res = 0;
    // 3. Sum the numbers up.
	for (int i = 0; i < len; i++) {
		int _0 = py_toint(py_tuple_getitem(py_arg(0), i));
		res += _0;
	}
    // 4. Set the result.
	py_newint(py_retval(), res);
    // 5. Return `true`.
	return res;
}

And then bind it:

py_GlobalRef mod = py_newmodule("sumary");
py_bind(mod, "sum(*values: tuple[int])", py_sum);

It can be used like this:

import sumary
print(sumary.sum(2,3,4,5,6))

Make a simple print function

Let's make a simple print function now. It takes arbitrary argument *values, and end/sep is not necessary. It's so simple that only string argument is acceptable.

Here's an implementation:

bool py_print(int argc, py_Ref argv) {
    // 1. *values is always a tuple.
	PY_CHECK_ARG_TYPE(0, tp_tuple);
    // 2. Get the length of tuple.
	int len = py_tuple_len(py_arg(0));
	const char* end = "\n";
	const char* sep = " ";
    // 3. First arg is sep, but it could be None. 
	if (!py_isnone(py_arg(1))) {
		PY_CHECK_ARG_TYPE(1, tp_str);
		sep = py_tostr(py_arg(1));
	}
    // 4. Second arg is end, it also can be None.
	if (!py_isnone(py_arg(2))) {
		PY_CHECK_ARG_TYPE(2, tp_str);
		end = py_tostr(py_arg(2));
	}
    // 5. Print.
	for (int i = 0; i < len; i++) {
		if (i > 0) {
			printf("%s", sep);
		}
        // 6. It can print iterable like `list` if you modify this line.
		printf("%s", py_tostr(py_tuple_getitem(py_arg(0), i)));
	}
	printf("%s", end);
    // 7. All the functions should return a value, here None is returned.
	py_newnone(py_retval());
	return true;
}

And then bind:

py_GlobalRef mod = py_newmodule("myprint");
py_bind(mod, "my_print(*values: object, sep: str | None = None, end: str | None = None)", py_print);

It can print names like this:

import myprint
myprint.my_print('Bob','Mary', end = 'Cake', sep = '|')

See also:

• py_bind
• py_bindmethod
• py_bindfunc
• py_bindproperty
• py_newmodule
• py_newtype