/* Ordered Dictionary object implementation. This implementation is necessarily explicitly equivalent to the pure Python OrderedDict class in Lib/collections/__init__.py. The strategy there involves using a doubly-linked-list to capture the order. We keep to that strategy, using a lower-level linked-list. About the Linked-List ===================== For the linked list we use a basic doubly-linked-list. Using a circularly- linked-list does have some benefits, but they don't apply so much here since OrderedDict is focused on the ends of the list (for the most part). Furthermore, there are some features of generic linked-lists that we simply don't need for OrderedDict. Thus a simple custom implementation meets our needs. Alternatives to our simple approach include the QCIRCLE_* macros from BSD's queue.h, and the linux's list.h. Getting O(1) Node Lookup ------------------------ One invariant of Python's OrderedDict is that it preserves time complexity of dict's methods, particularly the O(1) operations. Simply adding a linked-list on top of dict is not sufficient here; operations for nodes in the middle of the linked-list implicitly require finding the node first. With a simple linked-list like we're using, that is an O(n) operation. Consequently, methods like __delitem__() would change from O(1) to O(n), which is unacceptable. In order to preserve O(1) performance for node removal (finding nodes), we must do better than just looping through the linked-list. Here are options we've considered: 1. use a second dict to map keys to nodes (a la the pure Python version). 2. keep a simple hash table mirroring the order of dict's, mapping each key to the corresponding node in the linked-list. 3. use a version of shared keys (split dict) that allows non-unicode keys. 4. have the value stored for each key be a (value, node) pair, and adjust __getitem__(), get(), etc. accordingly. The approach with the least performance impact (time and space) is #2, mirroring the key order of dict's dk_entries with an array of node pointers. While _Py_dict_lookup() does not give us the index into the array, we make use of pointer arithmetic to get that index. An alternative would be to refactor _Py_dict_lookup() to provide the index, explicitly exposing the implementation detail. We could even just use a custom lookup function for OrderedDict that facilitates our need. However, both approaches are significantly more complicated than just using pointer arithmetic. The catch with mirroring the hash table ordering is that we have to keep the ordering in sync through any dict resizes. However, that order only matters during node lookup. We can simply defer any potential resizing until we need to do a lookup. Linked-List Nodes ----------------- The current implementation stores a pointer to the associated key only. One alternative would be to store a pointer to the PyDictKeyEntry instead. This would save one pointer de-reference per item, which is nice during calls to values() and items(). However, it adds unnecessary overhead otherwise, so we stick with just the key. Linked-List API --------------- As noted, the linked-list implemented here does not have all the bells and whistles. However, we recognize that the implementation may need to change to accommodate performance improvements or extra functionality. To that end, we use a simple API to interact with the linked-list. Here's a summary of the methods/macros: Node info: * _odictnode_KEY(node) * _odictnode_VALUE(od, node) * _odictnode_PREV(node) * _odictnode_NEXT(node) Linked-List info: * _odict_FIRST(od) * _odict_LAST(od) * _odict_EMPTY(od) * _odict_FOREACH(od, node) - used in place of `for (node=...)` For adding nodes: * _odict_add_head(od, node) * _odict_add_tail(od, node) * _odict_add_new_node(od, key, hash) For removing nodes: * _odict_clear_node(od, node, key, hash) * _odict_clear_nodes(od, clear_each) Others: * _odict_find_node_hash(od, key, hash) * _odict_find_node(od, key) * _odict_keys_equal(od1, od2) And here's a look at how the linked-list relates to the OrderedDict API: ============ === === ==== ==== ==== === ==== ===== ==== ==== === ==== === === method key val prev next mem 1st last empty iter find add rmv clr keq ============ === === ==== ==== ==== === ==== ===== ==== ==== === ==== === === __del__ ~ X __delitem__ free ~ node __eq__ ~ X __iter__ X X __new__ X X __reduce__ X ~ X __repr__ X X X __reversed__ X X __setitem__ key __sizeof__ size X clear ~ ~ X copy X X X items X X X keys X X move_to_end X X X ~ h/t key pop free key popitem X X free X X node setdefault ~ ? ~ values X X ============ === === ==== ==== ==== === ==== ===== ==== ==== === ==== === === __delitem__ is the only method that directly relies on finding an arbitrary node in the linked-list. Everything else is iteration or relates to the ends of the linked-list. Situation that Endangers Consistency ------------------------------------ Using a raw linked-list for OrderedDict exposes a key situation that can cause problems. If a node is stored in a variable, there is a chance that the node may have been deallocated before the variable gets used, thus potentially leading to a segmentation fault. A key place where this shows up is during iteration through the linked list (via _odict_FOREACH or otherwise). A number of solutions are available to resolve this situation: * defer looking up the node until as late as possible and certainly after any code that could possibly result in a deletion; * if the node is needed both before and after a point where the node might be removed, do a check before using the node at the "after" location to see if the node is still valid; * like the last one, but simply pull the node again to ensure it's right; * keep the key in the variable instead of the node and then look up the node using the key at the point where the node is needed (this is what we do for the iterators). Another related problem, preserving consistent ordering during iteration, is described below. That one is not exclusive to using linked-lists. Challenges from Subclassing dict ================================ OrderedDict subclasses dict, which is an unusual relationship between two builtin types (other than the base object type). Doing so results in some complication and deserves further explanation. There are two things to consider here. First, in what circumstances or with what adjustments can OrderedDict be used as a drop-in replacement for dict (at the C level)? Second, how can the OrderedDict implementation leverage the dict implementation effectively without introducing unnecessary coupling or inefficiencies? This second point is reflected here and in the implementation, so the further focus is on the first point. It is worth noting that for overridden methods, the dict implementation is deferred to as much as possible. Furthermore, coupling is limited to as little as is reasonable. Concrete API Compatibility -------------------------- Use of the concrete C-API for dict (PyDict_*) with OrderedDict is problematic. (See http://bugs.python.org/issue10977.) The concrete API has a number of hard-coded assumptions tied to the dict implementation. This is, in part, due to performance reasons, which is understandable given the part dict plays in Python. Any attempt to replace dict with OrderedDict for any role in the interpreter (e.g. **kwds) faces a challenge. Such any effort must recognize that the instances in affected locations currently interact with the concrete API. Here are some ways to address this challenge: 1. Change the relevant usage of the concrete API in CPython and add PyDict_CheckExact() calls to each of the concrete API functions. 2. Adjust the relevant concrete API functions to explicitly accommodate OrderedDict. 3. As with #1, add the checks, but improve the abstract API with smart fast paths for dict and OrderedDict, and refactor CPython to use the abstract API. Improvements to the abstract API would be valuable regardless. Adding the checks to the concrete API would help make any interpreter switch to OrderedDict less painful for extension modules. However, this won't work. The equivalent C API call to `dict.__setitem__(obj, k, v)` is `PyDict_SetItem(obj, k, v)`. This illustrates how subclasses in C call the base class's methods, since there is no equivalent of super() in the C API. Calling into Python for parent class API would work, but some extension modules already rely on this feature of the concrete API. For reference, here is a breakdown of some of the dict concrete API: ========================== ============= ======================= concrete API uses abstract API ========================== ============= ======================= PyDict_Check PyMapping_Check (PyDict_CheckExact) - (PyDict_New) - (PyDictProxy_New) - PyDict_Clear - PyDict_Contains PySequence_Contains PyDict_Copy - PyDict_SetItem PyObject_SetItem PyDict_SetItemString PyMapping_SetItemString PyDict_DelItem PyMapping_DelItem PyDict_DelItemString PyMapping_DelItemString PyDict_GetItem - PyDict_GetItemWithError PyObject_GetItem _PyDict_GetItemIdWithError - PyDict_GetItemString PyMapping_GetItemString PyDict_Items PyMapping_Items PyDict_Keys PyMapping_Keys PyDict_Values PyMapping_Values PyDict_Size PyMapping_Size PyMapping_Length PyDict_Next PyIter_Next _PyDict_Next - PyDict_Merge - PyDict_Update - PyDict_MergeFromSeq2 - PyDict_ClearFreeList - - PyMapping_HasKeyString - PyMapping_HasKey ========================== ============= ======================= The dict Interface Relative to OrderedDict ========================================== Since OrderedDict subclasses dict, understanding the various methods and attributes of dict is important for implementing OrderedDict. Relevant Type Slots ------------------- ================= ================ =================== ================ slot attribute object dict ================= ================ =================== ================ tp_dealloc - object_dealloc dict_dealloc tp_repr __repr__ object_repr dict_repr sq_contains __contains__ - dict_contains mp_length __len__ - dict_length mp_subscript __getitem__ - dict_subscript mp_ass_subscript __setitem__ - dict_ass_sub __delitem__ tp_hash __hash__ _Py_HashPointer ..._HashNotImpl tp_str __str__ object_str - tp_getattro __getattribute__ ..._GenericGetAttr (repeated) __getattr__ tp_setattro __setattr__ ..._GenericSetAttr (disabled) tp_doc __doc__ (literal) dictionary_doc tp_traverse - - dict_traverse tp_clear - - dict_tp_clear tp_richcompare __eq__ object_richcompare dict_richcompare __ne__ tp_weaklistoffset (__weakref__) - - tp_iter __iter__ - dict_iter tp_dictoffset (__dict__) - - tp_init __init__ object_init dict_init tp_alloc - PyType_GenericAlloc (repeated) tp_new __new__ object_new dict_new tp_free - PyObject_Free PyObject_GC_Del ================= ================ =================== ================ Relevant Methods ---------------- ================ =================== =============== method object dict ================ =================== =============== __reduce__ object_reduce - __sizeof__ object_sizeof dict_sizeof clear - dict_clear copy - dict_copy fromkeys - dict_fromkeys get - dict_get items - dictitems_new keys - dictkeys_new pop - dict_pop popitem - dict_popitem setdefault - dict_setdefault update - dict_update values - dictvalues_new ================ =================== =============== Pure Python OrderedDict ======================= As already noted, compatibility with the pure Python OrderedDict implementation is a key goal of this C implementation. To further that goal, here's a summary of how OrderedDict-specific methods are implemented in collections/__init__.py. Also provided is an indication of which methods directly mutate or iterate the object, as well as any relationship with the underlying linked-list. ============= ============== == ================ === === ==== method impl used ll uses inq mut iter ============= ============== == ================ === === ==== __contains__ dict - - X __delitem__ OrderedDict Y dict.__delitem__ X __eq__ OrderedDict N OrderedDict ~ dict.__eq__ __iter__ __getitem__ dict - - X __iter__ OrderedDict Y - X __init__ OrderedDict N update __len__ dict - - X __ne__ MutableMapping - __eq__ ~ __reduce__ OrderedDict N OrderedDict ~ __iter__ __getitem__ __repr__ OrderedDict N __class__ ~ items __reversed__ OrderedDict Y - X __setitem__ OrderedDict Y __contains__ X dict.__setitem__ __sizeof__ OrderedDict Y __len__ ~ __dict__ clear OrderedDict Y dict.clear X copy OrderedDict N __class__ __init__ fromkeys OrderedDict N __setitem__ get dict - - ~ items MutableMapping - ItemsView X keys MutableMapping - KeysView X move_to_end OrderedDict Y - X pop OrderedDict N __contains__ X __getitem__ __delitem__ popitem OrderedDict Y dict.pop X setdefault OrderedDict N __contains__ ~ __getitem__ __setitem__ update MutableMapping - __setitem__ ~ values MutableMapping - ValuesView X ============= ============== == ================ === === ==== __reversed__ and move_to_end are both exclusive to OrderedDict. C OrderedDict Implementation ============================ ================= ================ slot impl ================= ================ tp_dealloc odict_dealloc tp_repr odict_repr mp_ass_subscript odict_ass_sub tp_doc odict_doc tp_traverse odict_traverse tp_clear odict_tp_clear tp_richcompare odict_richcompare tp_weaklistoffset (offset) tp_iter odict_iter tp_dictoffset (offset) tp_init odict_init tp_alloc (repeated) ================= ================ ================= ================ method impl ================= ================ __reduce__ odict_reduce __sizeof__ odict_sizeof clear odict_clear copy odict_copy fromkeys odict_fromkeys items odictitems_new keys odictkeys_new pop odict_pop popitem odict_popitem setdefault odict_setdefault update odict_update values odictvalues_new ================= ================ Inherited unchanged from object/dict: ================ ========================== method type field ================ ========================== - tp_free __contains__ tp_as_sequence.sq_contains __getattr__ tp_getattro __getattribute__ tp_getattro __getitem__ tp_as_mapping.mp_subscript __hash__ tp_hash __len__ tp_as_mapping.mp_length __setattr__ tp_setattro __str__ tp_str get - ================ ========================== Other Challenges ================ Preserving Ordering During Iteration ------------------------------------ During iteration through an OrderedDict, it is possible that items could get added, removed, or reordered. For a linked-list implementation, as with some other implementations, that situation may lead to undefined behavior. The documentation for dict mentions this in the `iter()` section of http://docs.python.org/3.4/library/stdtypes.html#dictionary-view-objects. In this implementation we follow dict's lead (as does the pure Python implementation) for __iter__(), keys(), values(), and items(). For internal iteration (using _odict_FOREACH or not), there is still the risk that not all nodes that we expect to be seen in the loop actually get seen. Thus, we are careful in each of those places to ensure that they are. This comes, of course, at a small price at each location. The solutions are much the same as those detailed in the `Situation that Endangers Consistency` section above. Potential Optimizations ======================= * Allocate the nodes as a block via od_fast_nodes instead of individually. - Set node->key to NULL to indicate the node is not-in-use. - Add _odict_EXISTS()? - How to maintain consistency across resizes? Existing node pointers would be invalidated after a resize, which is particularly problematic for the iterators. * Use a more stream-lined implementation of update() and, likely indirectly, __init__(). */ /* TODO sooner: - reentrancy (make sure everything is at a thread-safe state when calling into Python). I've already checked this multiple times, but want to make one more pass. - add unit tests for reentrancy? later: - make the dict views support the full set API (the pure Python impl does) - implement a fuller MutableMapping API in C? - move the MutableMapping implementation to abstract.c? - optimize mutablemapping_update - use PyObject_Malloc (small object allocator) for odict nodes? - support subclasses better (e.g. in odict_richcompare) */ #include "Python.h" #include "pycore_call.h" // _PyObject_CallNoArgs() #include "pycore_ceval.h" // _PyEval_GetBuiltin() #include "pycore_critical_section.h" //_Py_BEGIN_CRITICAL_SECTION #include "pycore_dict.h" // _Py_dict_lookup() #include "pycore_object.h" // _PyObject_GC_UNTRACK() #include "pycore_pyerrors.h" // _PyErr_ChainExceptions1() #include <stddef.h> // offsetof() #include "clinic/odictobject.c.h" /*[clinic input] class OrderedDict "PyODictObject *" "&PyODict_Type" [clinic start generated code]*/ /*[clinic end generated code: output=da39a3ee5e6b4b0d input=ca0641cf6143d4af]*/ _ODictNode; /* PyODictObject */ struct _odictobject { … }; /* ---------------------------------------------- * odict keys (a simple doubly-linked list) */ struct _odictnode { … }; #define _odictnode_KEY(node) … #define _odictnode_HASH(node) … /* borrowed reference */ #define _odictnode_VALUE(node, od) … #define _odictnode_PREV(node) … #define _odictnode_NEXT(node) … #define _odict_FIRST(od) … #define _odict_LAST(od) … #define _odict_EMPTY(od) … #define _odict_FOREACH(od, node) … /* Return the index into the hash table, regardless of a valid node. */ static Py_ssize_t _odict_get_index_raw(PyODictObject *od, PyObject *key, Py_hash_t hash) { … } #define ONE … /* Replace od->od_fast_nodes with a new table matching the size of dict's. */ static int _odict_resize(PyODictObject *od) { … } /* Return the index into the hash table, regardless of a valid node. */ static Py_ssize_t _odict_get_index(PyODictObject *od, PyObject *key, Py_hash_t hash) { … } /* Returns NULL if there was some error or the key was not found. */ static _ODictNode * _odict_find_node_hash(PyODictObject *od, PyObject *key, Py_hash_t hash) { … } static _ODictNode * _odict_find_node(PyODictObject *od, PyObject *key) { … } static void _odict_add_head(PyODictObject *od, _ODictNode *node) { … } static void _odict_add_tail(PyODictObject *od, _ODictNode *node) { … } /* adds the node to the end of the list */ static int _odict_add_new_node(PyODictObject *od, PyObject *key, Py_hash_t hash) { … } /* Putting the decref after the free causes problems. */ #define _odictnode_DEALLOC … /* Repeated calls on the same node are no-ops. */ static void _odict_remove_node(PyODictObject *od, _ODictNode *node) { … } /* If someone calls PyDict_DelItem() directly on an OrderedDict, we'll get all sorts of problems here. In PyODict_DelItem we make sure to call _odict_clear_node first. This matters in the case of colliding keys. Suppose we add 3 keys: [A, B, C], where the hash of C collides with A and the next possible index in the hash table is occupied by B. If we remove B then for C the dict's looknode func will give us the old index of B instead of the index we got before deleting B. However, the node for C in od_fast_nodes is still at the old dict index of C. Thus to be sure things don't get out of sync, we clear the node in od_fast_nodes *before* calling PyDict_DelItem. The same must be done for any other OrderedDict operations where we modify od_fast_nodes. */ static int _odict_clear_node(PyODictObject *od, _ODictNode *node, PyObject *key, Py_hash_t hash) { … } static void _odict_clear_nodes(PyODictObject *od) { … } /* There isn't any memory management of nodes past this point. */ #undef _odictnode_DEALLOC static int _odict_keys_equal(PyODictObject *a, PyODictObject *b) { … } /* ---------------------------------------------- * OrderedDict mapping methods */ /* mp_ass_subscript: __setitem__() and __delitem__() */ static int odict_mp_ass_sub(PyODictObject *od, PyObject *v, PyObject *w) { … } /* tp_as_mapping */ static PyMappingMethods odict_as_mapping = …; /* ---------------------------------------------- * OrderedDict number methods */ static int mutablemapping_update_arg(PyObject*, PyObject*); static PyObject * odict_or(PyObject *left, PyObject *right) { … } static PyObject * odict_inplace_or(PyObject *self, PyObject *other) { … } /* tp_as_number */ static PyNumberMethods odict_as_number = …; /* ---------------------------------------------- * OrderedDict methods */ /* fromkeys() */ /*[clinic input] @classmethod OrderedDict.fromkeys iterable as seq: object value: object = None Create a new ordered dictionary with keys from iterable and values set to value. [clinic start generated code]*/ static PyObject * OrderedDict_fromkeys_impl(PyTypeObject *type, PyObject *seq, PyObject *value) /*[clinic end generated code: output=c10390d452d78d6d input=1a0476c229c597b3]*/ { … } /* __sizeof__() */ /* OrderedDict.__sizeof__() does not have a docstring. */ PyDoc_STRVAR(odict_sizeof__doc__, ""); static PyObject * odict_sizeof(PyODictObject *od, PyObject *Py_UNUSED(ignored)) { … } /* __reduce__() */ PyDoc_STRVAR(odict_reduce__doc__, "Return state information for pickling"); static PyObject * odict_reduce(register PyODictObject *od, PyObject *Py_UNUSED(ignored)) { … } /* setdefault(): Skips __missing__() calls. */ /*[clinic input] OrderedDict.setdefault key: object default: object = None Insert key with a value of default if key is not in the dictionary. Return the value for key if key is in the dictionary, else default. [clinic start generated code]*/ static PyObject * OrderedDict_setdefault_impl(PyODictObject *self, PyObject *key, PyObject *default_value) /*[clinic end generated code: output=97537cb7c28464b6 input=38e098381c1efbc6]*/ { … } /* pop() */ static PyObject * _odict_popkey_hash(PyObject *od, PyObject *key, PyObject *failobj, Py_hash_t hash) { … } /* Skips __missing__() calls. */ /*[clinic input] OrderedDict.pop key: object default: object = NULL od.pop(key[,default]) -> v, remove specified key and return the corresponding value. If the key is not found, return the default if given; otherwise, raise a KeyError. [clinic start generated code]*/ static PyObject * OrderedDict_pop_impl(PyODictObject *self, PyObject *key, PyObject *default_value) /*[clinic end generated code: output=7a6447d104e7494b input=7efe36601007dff7]*/ { … } /* popitem() */ /*[clinic input] OrderedDict.popitem last: bool = True Remove and return a (key, value) pair from the dictionary. Pairs are returned in LIFO order if last is true or FIFO order if false. [clinic start generated code]*/ static PyObject * OrderedDict_popitem_impl(PyODictObject *self, int last) /*[clinic end generated code: output=98e7d986690d49eb input=d992ac5ee8305e1a]*/ { … } /* keys() */ /* MutableMapping.keys() does not have a docstring. */ PyDoc_STRVAR(odict_keys__doc__, ""); static PyObject * odictkeys_new(PyObject *od, PyObject *Py_UNUSED(ignored)); /* forward */ /* values() */ /* MutableMapping.values() does not have a docstring. */ PyDoc_STRVAR(odict_values__doc__, ""); static PyObject * odictvalues_new(PyObject *od, PyObject *Py_UNUSED(ignored)); /* forward */ /* items() */ /* MutableMapping.items() does not have a docstring. */ PyDoc_STRVAR(odict_items__doc__, ""); static PyObject * odictitems_new(PyObject *od, PyObject *Py_UNUSED(ignored)); /* forward */ /* update() */ /* MutableMapping.update() does not have a docstring. */ PyDoc_STRVAR(odict_update__doc__, ""); /* forward */ static PyObject * mutablemapping_update(PyObject *, PyObject *, PyObject *); #define odict_update … /* clear() */ PyDoc_STRVAR(odict_clear__doc__, "od.clear() -> None. Remove all items from od."); static PyObject * odict_clear(register PyODictObject *od, PyObject *Py_UNUSED(ignored)) { … } /* copy() */ /* forward */ static int _PyODict_SetItem_KnownHash(PyObject *, PyObject *, PyObject *, Py_hash_t); PyDoc_STRVAR(odict_copy__doc__, "od.copy() -> a shallow copy of od"); static PyObject * odict_copy(register PyODictObject *od, PyObject *Py_UNUSED(ignored)) { … } /* __reversed__() */ PyDoc_STRVAR(odict_reversed__doc__, "od.__reversed__() <==> reversed(od)"); #define _odict_ITER_REVERSED … #define _odict_ITER_KEYS … #define _odict_ITER_VALUES … #define _odict_ITER_ITEMS … /* forward */ static PyObject * odictiter_new(PyODictObject *, int); static PyObject * odict_reversed(PyODictObject *od, PyObject *Py_UNUSED(ignored)) { … } /* move_to_end() */ /*[clinic input] OrderedDict.move_to_end key: object last: bool = True Move an existing element to the end (or beginning if last is false). Raise KeyError if the element does not exist. [clinic start generated code]*/ static PyObject * OrderedDict_move_to_end_impl(PyODictObject *self, PyObject *key, int last) /*[clinic end generated code: output=fafa4c5cc9b92f20 input=d6ceff7132a2fcd7]*/ { … } /* tp_methods */ static PyMethodDef odict_methods[] = …; /* ---------------------------------------------- * OrderedDict members */ /* tp_getset */ static PyGetSetDef odict_getset[] = …; /* ---------------------------------------------- * OrderedDict type slot methods */ /* tp_dealloc */ static void odict_dealloc(PyODictObject *self) { … } /* tp_repr */ static PyObject * odict_repr(PyODictObject *self) { … } /* tp_doc */ PyDoc_STRVAR(odict_doc, "Dictionary that remembers insertion order"); /* tp_traverse */ static int odict_traverse(PyODictObject *od, visitproc visit, void *arg) { … } /* tp_clear */ static int odict_tp_clear(PyODictObject *od) { … } /* tp_richcompare */ static PyObject * odict_richcompare(PyObject *v, PyObject *w, int op) { … } /* tp_iter */ static PyObject * odict_iter(PyODictObject *od) { … } /* tp_init */ static int odict_init(PyObject *self, PyObject *args, PyObject *kwds) { … } /* PyODict_Type */ PyTypeObject PyODict_Type = …; /* ---------------------------------------------- * the public OrderedDict API */ PyObject * PyODict_New(void) { … } static int _PyODict_SetItem_KnownHash(PyObject *od, PyObject *key, PyObject *value, Py_hash_t hash) { … } int PyODict_SetItem(PyObject *od, PyObject *key, PyObject *value) { … } int PyODict_DelItem(PyObject *od, PyObject *key) { … } /* ------------------------------------------- * The OrderedDict views (keys/values/items) */ odictiterobject; static void odictiter_dealloc(odictiterobject *di) { … } static int odictiter_traverse(odictiterobject *di, visitproc visit, void *arg) { … } /* In order to protect against modifications during iteration, we track * the current key instead of the current node. */ static PyObject * odictiter_nextkey(odictiterobject *di) { … } static PyObject * odictiter_iternext(odictiterobject *di) { … } /* No need for tp_clear because odictiterobject is not mutable. */ PyDoc_STRVAR(reduce_doc, "Return state information for pickling"); static PyObject * odictiter_reduce(odictiterobject *di, PyObject *Py_UNUSED(ignored)) { … } static PyMethodDef odictiter_methods[] = …; PyTypeObject PyODictIter_Type = …; static PyObject * odictiter_new(PyODictObject *od, int kind) { … } /* keys() */ static PyObject * odictkeys_iter(_PyDictViewObject *dv) { … } static PyObject * odictkeys_reversed(_PyDictViewObject *dv, PyObject *Py_UNUSED(ignored)) { … } static PyMethodDef odictkeys_methods[] = …; PyTypeObject PyODictKeys_Type = …; static PyObject * odictkeys_new(PyObject *od, PyObject *Py_UNUSED(ignored)) { … } /* items() */ static PyObject * odictitems_iter(_PyDictViewObject *dv) { … } static PyObject * odictitems_reversed(_PyDictViewObject *dv, PyObject *Py_UNUSED(ignored)) { … } static PyMethodDef odictitems_methods[] = …; PyTypeObject PyODictItems_Type = …; static PyObject * odictitems_new(PyObject *od, PyObject *Py_UNUSED(ignored)) { … } /* values() */ static PyObject * odictvalues_iter(_PyDictViewObject *dv) { … } static PyObject * odictvalues_reversed(_PyDictViewObject *dv, PyObject *Py_UNUSED(ignored)) { … } static PyMethodDef odictvalues_methods[] = …; PyTypeObject PyODictValues_Type = …; static PyObject * odictvalues_new(PyObject *od, PyObject *Py_UNUSED(ignored)) { … } /* ---------------------------------------------- MutableMapping implementations Mapping: ============ =========== method uses ============ =========== __contains__ __getitem__ __eq__ items __getitem__ + __iter__ + __len__ + __ne__ __eq__ get __getitem__ items ItemsView keys KeysView values ValuesView ============ =========== ItemsView uses __len__, __iter__, and __getitem__. KeysView uses __len__, __iter__, and __contains__. ValuesView uses __len__, __iter__, and __getitem__. MutableMapping: ============ =========== method uses ============ =========== __delitem__ + __setitem__ + clear popitem pop __getitem__ __delitem__ popitem __iter__ _getitem__ __delitem__ setdefault __getitem__ __setitem__ update __setitem__ ============ =========== */ static int mutablemapping_add_pairs(PyObject *self, PyObject *pairs) { … } static int mutablemapping_update_arg(PyObject *self, PyObject *arg) { … } static PyObject * mutablemapping_update(PyObject *self, PyObject *args, PyObject *kwargs) { … }
Codebase Browser
cpython