如何使用Python C API创建生成器/迭代器？

Question

如何使用Python C API复制以下Python代码？

class Sequence():
    def __init__(self, max):
        self.max = max
    def data(self):
        i = 0
        while i < self.max:
            yield i
            i += 1

到目前为止,我有这个:

#include <Python/Python.h>
#include <Python/structmember.h>

/* Define a new object class, Sequence. */
typedef struct {
    PyObject_HEAD
    size_t max;
} SequenceObject;

/* Instance variables */
static PyMemberDef Sequence_members[] = {
    {"max", T_UINT, offsetof(SequenceObject, max), 0, NULL},
    {NULL} /* Sentinel */
};

static int Sequence_Init(SequenceObject *self, PyObject *args, PyObject *kwds)
{
    if (!PyArg_ParseTuple(args, "k", &(self->max))) {
        return -1;
    }
    return 0;
}

static PyObject *Sequence_data(SequenceObject *self, PyObject *args);

/* Methods */
static PyMethodDef Sequence_methods[] = {
    {"data", (PyCFunction)Sequence_data, METH_NOARGS,
     "sequence.data() -> iterator object\n"
     "Returns iterator of range [0, sequence.max)."},
    {NULL} /* Sentinel */
};

/* Define new object type */
PyTypeObject Sequence_Type = {
   PyObject_HEAD_INIT(NULL)
   0,                         /* ob_size */
   "Sequence",                /* tp_name */
   sizeof(SequenceObject),    /* tp_basicsize */
   0,                         /* tp_itemsize */
   0,                         /* tp_dealloc */
   0,                         /* tp_print */
   0,                         /* tp_getattr */
   0,                         /* tp_setattr */
   0,                         /* tp_compare */
   0,                         /* tp_repr */
   0,                         /* tp_as_number */
   0,                         /* tp_as_sequence */
   0,                         /* tp_as_mapping */
   0,                         /* tp_hash */
   0,                         /* tp_call */
   0,                         /* tp_str */
   0,                         /* tp_getattro */
   0,                         /* tp_setattro */
   0,                         /* tp_as_buffer */
   Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /* tp_flags*/
   "Test generator object",   /* tp_doc */
   0,                         /* tp_traverse */
   0,                         /* tp_clear */
   0,                         /* tp_richcompare */
   0,                         /* tp_weaklistoffset */
   0,                         /* tp_iter */
   0,                         /* tp_iternext */
   0,                         /* tp_methods */
   Sequence_members,          /* tp_members */
   0,                         /* tp_getset */
   0,                         /* tp_base */
   0,                         /* tp_dict */
   0,                         /* tp_descr_get */
   0,                         /* tp_descr_set */
   0,                         /* tp_dictoffset */
   (initproc)Sequence_init,   /* tp_init */
   0,                         /* tp_alloc */
   PyType_GenericNew,         /* tp_new */
};

static PyObject *Sequence_data(SequenceObject *self, PyObject *args)
{
    /* Now what? */
}

但我不知道下一步该去哪里.有人可以提供一些建议吗？

编辑

我想我遇到的主要问题是模拟yield语句.据我所知,它是一个非常简单的,但实际上是复杂的声明 - 它创建了一个具有自己的生成器__iter__()和next()自动调用的方法.搜索文档,它似乎与PyGenObject相关联; 但是,如何创建此对象的新实例尚不清楚.PyGen_New()作为它的参数a PyFrameObject,我能找到的唯一参考是PyEval_GetFrame(),这似乎不是我想要的(或者我错了？).有没有人有这方面的经验可以分享？

进一步编辑

当我(基本上)扩展Python在幕后所做的事情时,我发现这更清楚了:

class IterObject():
    def __init__(self, max):
        self.max = max
    def __iter__(self):
        self.i = 0
        return self
    def next(self):
        if self.i >= self.max:
            raise StopIteration
        self.i += 1
        return self.i

class Sequence():
    def __init__(self, max):
        self.max = max
    def data(self):
        return IterObject(self.max)

从技术上讲,顺序是一个人关闭,但你明白了.

唯一的问题是每次需要一个生成器时创建一个新对象都非常烦人 - 在Python中比C更是如此,因为定义一个新类型需要的怪物.yieldC中没有语句,因为C没有闭包.我做了什么(因为我在Python API中找不到它 -  请指出我已经存在的标准对象!)是创建一个简单的通用生成器对象类,它为每个next()方法调用回调一个C函数.这是(注意我还没有尝试编译它,因为它不完整 - 见下文):

#include <Python/Python.h>
#include <Python/structmember.h>
#include <stdlib.h>

/* A convenient, generic generator object. */

typedef PyObject *(*callback)(PyObject *callee, void *info) PyGeneratorCallback;

typedef struct {
    PyObject HEAD
    PyGeneratorCallback callback;
    PyObject *callee;
    void *callbackInfo; /* info to be passed along to callback function. */
    bool freeInfo; /* true if |callbackInfo| should be free'()d when object
                    * dealloc's, false if not. */
} GeneratorObject;

static PyObject *Generator_iter(PyObject *self, PyObject *args)
{
    Py_INCREF(self);
    return self;
}

static PyObject *Generator_next(PyObject *self, PyObject *args)
{
    return self->callback(self->callee, self->callbackInfo);
}

static PyMethodDef Generator_methods[] = {
    {"__iter__", (PyCFunction)Generator_iter, METH_NOARGS, NULL},
    {"next", (PyCFunction)Generator_next, METH_NOARGS, NULL},
    {NULL} /* Sentinel */
};

static void Generator_dealloc(GenericEventObject *self)
{
    if (self->freeInfo && self->callbackInfo != NULL) {
        free(self->callbackInfo);
    }
    self->ob_type->tp_free((PyObject *)self);
}

PyTypeObject Generator_Type = {
   PyObject_HEAD_INIT(NULL)
   0,                         /* ob_size */
   "Generator",               /* tp_name */
   sizeof(GeneratorObject),   /* tp_basicsize */
   0,                         /* tp_itemsize */
   Generator_dealloc,         /* tp_dealloc */
   0,                         /* tp_print */
   0,                         /* tp_getattr */
   0,                         /* tp_setattr */
   0,                         /* tp_compare */
   0,                         /* tp_repr */
   0,                         /* tp_as_number */
   0,                         /* tp_as_sequence */
   0,                         /* tp_as_mapping */
   0,                         /* tp_hash */
   0,                         /* tp_call */
   0,                         /* tp_str */
   0,                         /* tp_getattro */
   0,                         /* tp_setattro */
   0,                         /* tp_as_buffer */
   Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /* tp_flags*/
   0,                         /* tp_doc */
   0,                         /* tp_traverse */
   0,                         /* tp_clear */
   0,                         /* tp_richcompare */
   0,                         /* tp_weaklistoffset */
   0,                         /* tp_iter */
   0,                         /* tp_iternext */
   0,                         /* tp_methods */
   0,                         /* tp_members */
   0,                         /* tp_getset */
   0,                         /* tp_base */
   0,                         /* tp_dict */
   0,                         /* tp_descr_get */
   0,                         /* tp_descr_set */
   0,                         /* tp_dictoffset */
   0,                         /* tp_init */
   0,                         /* tp_alloc */
   PyType_GenericNew,         /* tp_new */
};

/* Returns a new generator object with the given callback function
 * and arguments. */
PyObject *Generator_New(PyObject *callee, void *info,
                        bool freeInfo, PyGeneratorCallback callback)
{
    GeneratorObject *generator = (GeneratorObject *)_PyObject_New(&Generator_Type);
    if (generator == NULL) return NULL;

    generator->callee = callee;
    generator->info = info;
    generator->callback = callback;
    self->freeInfo = freeInfo;

    return (PyObject *)generator;
}

/* End of Generator definition. */

/* Define a new object class, Sequence. */
typedef struct {
    PyObject_HEAD
    size_t max;
} SequenceObject;

/* Instance variables */
static PyMemberDef Sequence_members[] = {
    {"max", T_UINT, offsetof(SequenceObject, max), 0, NULL},
    {NULL} /* Sentinel */
}

static int Sequence_Init(SequenceObject *self, PyObject *args, PyObject *kwds)
{
    if (!PyArg_ParseTuple(args, "k", &self->max)) {
        return -1;
    }
    return 0;
}

static PyObject *Sequence_data(SequenceObject *self, PyObject *args);

/* Methods */
static PyMethodDef Sequence_methods[] = {
    {"data", (PyCFunction)Sequence_data, METH_NOARGS,
     "sequence.data() -> iterator object\n"
     "Returns generator of range [0, sequence.max)."},
    {NULL} /* Sentinel */
};

/* Define new object type */
PyTypeObject Sequence_Type = {
   PyObject_HEAD_INIT(NULL)
   0,                         /* ob_size */
   "Sequence",                /* tp_name */
   sizeof(SequenceObject),    /* tp_basicsize */
   0,                         /* tp_itemsize */
   0,                         /* tp_dealloc */
   0,                         /* tp_print */
   0,                         /* tp_getattr */
   0,                         /* tp_setattr */
   0,                         /* tp_compare */
   0,                         /* tp_repr */
   0,                         /* tp_as_number */
   0,                         /* tp_as_sequence */
   0,                         /* tp_as_mapping */
   0,                         /* tp_hash */
   0,                         /* tp_call */
   0,                         /* tp_str */
   0,                         /* tp_getattro */
   0,                         /* tp_setattro */
   0,                         /* tp_as_buffer */
   Py_TPFLAGS_DEFAULT | Py_TPFLAGS_BASETYPE, /* tp_flags*/
   "Test generator object",   /* tp_doc */
   0,                         /* tp_traverse */
   0,                         /* tp_clear */
   0,                         /* tp_richcompare */
   0,                         /* tp_weaklistoffset */
   0,                         /* tp_iter */
   0,                         /* tp_iternext */
   0,                         /* tp_methods */
   Sequence_members,          /* tp_members */
   0,                         /* tp_getset */
   0,                         /* tp_base */
   0,                         /* tp_dict */
   0,                         /* tp_descr_get */
   0,                         /* tp_descr_set */
   0,                         /* tp_dictoffset */
   (initproc)Sequence_init,   /* tp_init */
   0,                         /* tp_alloc */
   PyType_GenericNew,         /* tp_new */
};

static PyObject *Sequence_data(SequenceObject *self, PyObject *args)
{
    size_t *info = malloc(sizeof(size_t));
    if (info == NULL) return NULL;
    *info = 0;

    /* |info| will be free'()d by the returned generator object. */
    GeneratorObject *ret = Generator_New(self, info, true,
                                         &Sequence_data_next_callback);
    if (ret == NULL) {
        free(info); /* Watch out for memory leaks! */
    }
    return ret;
}

PyObject *Sequence_data_next_callback(PyObject *self, void *info)
{
    size_t i = info;
    if (i > self->max) {
        return NULL; /* TODO: How do I raise StopIteration here? I can't seem to find
                      *       a standard exception. */
    } else {
        return Py_BuildValue("k", i++);
    }
}

然而,不幸的是,我还没有完成.我留下的唯一问题是:如何StopIteration使用C API 引发异常？我似乎无法在标准例外中找到它.另外,也许更重要的是,这是解决这个问题的正确方法吗？

感谢任何仍在关注此事的人.

Answer 1

下面是一个模块的简单实现,spam其中一个函数myiter(int)返回迭代器:

import spam
for i in spam.myiter(10):
    print i

打印0到9之间的数字.

它比你的情况简单,但显示了要点:用标准__iter__()和next()方法定义对象,以及实现迭代器行为,包括StopIteration适当时提高.

在你的情况下,迭代器对象需要保持对Sequence的引用(所以你需要deallocator方法来为Py_DECREF它).序列本身需要在其中实现__iter()__并创建一个迭代器.

包含迭代器状态的结构.(在您的版本而不是m中,它将引用Sequence.)

typedef struct {
  PyObject_HEAD
  long int m;
  long int i;
} spam_MyIter;

迭代器的__iter__()方法.它总是简单地返回self.它允许迭代器和集合在构造中被视为相同for ... in ....

PyObject* spam_MyIter_iter(PyObject *self)
{
  Py_INCREF(self);
  return self;
}

迭代的实现:next()方法.

PyObject* spam_MyIter_iternext(PyObject *self)
{
  spam_MyIter *p = (spam_MyIter *)self;
  if (p->i < p->m) {
    PyObject *tmp = Py_BuildValue("l", p->i);
    (p->i)++;
    return tmp;
  } else {
    /* Raising of standard StopIteration exception with empty value. */
    PyErr_SetNone(PyExc_StopIteration);
    return NULL;
  }
}

我们需要扩展版本的PyTypeObject结构来为Python提供有关__iter__()和的信息next().我们希望它们被有效地调用,因此在字典中没有基于名称的查找.

static PyTypeObject spam_MyIterType = {
    PyObject_HEAD_INIT(NULL)
    0,                         /*ob_size*/
    "spam._MyIter",            /*tp_name*/
    sizeof(spam_MyIter),       /*tp_basicsize*/
    0,                         /*tp_itemsize*/
    0,                         /*tp_dealloc*/
    0,                         /*tp_print*/
    0,                         /*tp_getattr*/
    0,                         /*tp_setattr*/
    0,                         /*tp_compare*/
    0,                         /*tp_repr*/
    0,                         /*tp_as_number*/
    0,                         /*tp_as_sequence*/
    0,                         /*tp_as_mapping*/
    0,                         /*tp_hash */
    0,                         /*tp_call*/
    0,                         /*tp_str*/
    0,                         /*tp_getattro*/
    0,                         /*tp_setattro*/
    0,                         /*tp_as_buffer*/
    Py_TPFLAGS_DEFAULT | Py_TPFLAGS_HAVE_ITER,
      /* tp_flags: Py_TPFLAGS_HAVE_ITER tells python to
         use tp_iter and tp_iternext fields. */
    "Internal myiter iterator object.",           /* tp_doc */
    0,  /* tp_traverse */
    0,  /* tp_clear */
    0,  /* tp_richcompare */
    0,  /* tp_weaklistoffset */
    spam_MyIter_iter,  /* tp_iter: __iter__() method */
    spam_MyIter_iternext  /* tp_iternext: next() method */
};

myiter(int) function创建迭代器.

static PyObject *
spam_myiter(PyObject *self, PyObject *args)
{
  long int m;
  spam_MyIter *p;

  if (!PyArg_ParseTuple(args, "l", &m))  return NULL;

  /* I don't need python callable __init__() method for this iterator,
     so I'll simply allocate it as PyObject and initialize it by hand. */

  p = PyObject_New(spam_MyIter, &spam_MyIterType);
  if (!p) return NULL;

  /* I'm not sure if it's strictly necessary. */
  if (!PyObject_Init((PyObject *)p, &spam_MyIterType)) {
    Py_DECREF(p);
    return NULL;
  }

  p->m = m;
  p->i = 0;
  return (PyObject *)p;
}

其余的很无聊......

static PyMethodDef SpamMethods[] = {
    {"myiter",  spam_myiter, METH_VARARGS, "Iterate from i=0 while i<m."},
    {NULL, NULL, 0, NULL}        /* Sentinel */
};

PyMODINIT_FUNC
initspam(void)
{
  PyObject* m;

  spam_MyIterType.tp_new = PyType_GenericNew;
  if (PyType_Ready(&spam_MyIterType) < 0)  return;

  m = Py_InitModule("spam", SpamMethods);

  Py_INCREF(&spam_MyIterType);
  PyModule_AddObject(m, "_MyIter", (PyObject *)&spam_MyIterType);
}

Answer 2

在Sequence_data,您必须返回一个新的PyInt实例或抛出一个StopIteration异常,告诉外面的代码没有更多的值.有关详细信息,请参阅PEP 255和9.10发生器.

请参阅Python/C API中有关辅助函数的迭代器协议.