Yi Kong | 8328301 | 2023-12-13 12:57:00 +0900 | [diff] [blame^] | 1 | """Helper to provide extensibility for pickle. |
| 2 | |
| 3 | This is only useful to add pickle support for extension types defined in |
| 4 | C, not for instances of user-defined classes. |
| 5 | """ |
| 6 | |
| 7 | __all__ = ["pickle", "constructor", |
| 8 | "add_extension", "remove_extension", "clear_extension_cache"] |
| 9 | |
| 10 | dispatch_table = {} |
| 11 | |
| 12 | def pickle(ob_type, pickle_function, constructor_ob=None): |
| 13 | if not callable(pickle_function): |
| 14 | raise TypeError("reduction functions must be callable") |
| 15 | dispatch_table[ob_type] = pickle_function |
| 16 | |
| 17 | # The constructor_ob function is a vestige of safe for unpickling. |
| 18 | # There is no reason for the caller to pass it anymore. |
| 19 | if constructor_ob is not None: |
| 20 | constructor(constructor_ob) |
| 21 | |
| 22 | def constructor(object): |
| 23 | if not callable(object): |
| 24 | raise TypeError("constructors must be callable") |
| 25 | |
| 26 | # Example: provide pickling support for complex numbers. |
| 27 | |
| 28 | try: |
| 29 | complex |
| 30 | except NameError: |
| 31 | pass |
| 32 | else: |
| 33 | |
| 34 | def pickle_complex(c): |
| 35 | return complex, (c.real, c.imag) |
| 36 | |
| 37 | pickle(complex, pickle_complex, complex) |
| 38 | |
| 39 | def pickle_union(obj): |
| 40 | import functools, operator |
| 41 | return functools.reduce, (operator.or_, obj.__args__) |
| 42 | |
| 43 | pickle(type(int | str), pickle_union) |
| 44 | |
| 45 | # Support for pickling new-style objects |
| 46 | |
| 47 | def _reconstructor(cls, base, state): |
| 48 | if base is object: |
| 49 | obj = object.__new__(cls) |
| 50 | else: |
| 51 | obj = base.__new__(cls, state) |
| 52 | if base.__init__ != object.__init__: |
| 53 | base.__init__(obj, state) |
| 54 | return obj |
| 55 | |
| 56 | _HEAPTYPE = 1<<9 |
| 57 | _new_type = type(int.__new__) |
| 58 | |
| 59 | # Python code for object.__reduce_ex__ for protocols 0 and 1 |
| 60 | |
| 61 | def _reduce_ex(self, proto): |
| 62 | assert proto < 2 |
| 63 | cls = self.__class__ |
| 64 | for base in cls.__mro__: |
| 65 | if hasattr(base, '__flags__') and not base.__flags__ & _HEAPTYPE: |
| 66 | break |
| 67 | new = base.__new__ |
| 68 | if isinstance(new, _new_type) and new.__self__ is base: |
| 69 | break |
| 70 | else: |
| 71 | base = object # not really reachable |
| 72 | if base is object: |
| 73 | state = None |
| 74 | else: |
| 75 | if base is cls: |
| 76 | raise TypeError(f"cannot pickle {cls.__name__!r} object") |
| 77 | state = base(self) |
| 78 | args = (cls, base, state) |
| 79 | try: |
| 80 | getstate = self.__getstate__ |
| 81 | except AttributeError: |
| 82 | if getattr(self, "__slots__", None): |
| 83 | raise TypeError(f"cannot pickle {cls.__name__!r} object: " |
| 84 | f"a class that defines __slots__ without " |
| 85 | f"defining __getstate__ cannot be pickled " |
| 86 | f"with protocol {proto}") from None |
| 87 | try: |
| 88 | dict = self.__dict__ |
| 89 | except AttributeError: |
| 90 | dict = None |
| 91 | else: |
| 92 | dict = getstate() |
| 93 | if dict: |
| 94 | return _reconstructor, args, dict |
| 95 | else: |
| 96 | return _reconstructor, args |
| 97 | |
| 98 | # Helper for __reduce_ex__ protocol 2 |
| 99 | |
| 100 | def __newobj__(cls, *args): |
| 101 | return cls.__new__(cls, *args) |
| 102 | |
| 103 | def __newobj_ex__(cls, args, kwargs): |
| 104 | """Used by pickle protocol 4, instead of __newobj__ to allow classes with |
| 105 | keyword-only arguments to be pickled correctly. |
| 106 | """ |
| 107 | return cls.__new__(cls, *args, **kwargs) |
| 108 | |
| 109 | def _slotnames(cls): |
| 110 | """Return a list of slot names for a given class. |
| 111 | |
| 112 | This needs to find slots defined by the class and its bases, so we |
| 113 | can't simply return the __slots__ attribute. We must walk down |
| 114 | the Method Resolution Order and concatenate the __slots__ of each |
| 115 | class found there. (This assumes classes don't modify their |
| 116 | __slots__ attribute to misrepresent their slots after the class is |
| 117 | defined.) |
| 118 | """ |
| 119 | |
| 120 | # Get the value from a cache in the class if possible |
| 121 | names = cls.__dict__.get("__slotnames__") |
| 122 | if names is not None: |
| 123 | return names |
| 124 | |
| 125 | # Not cached -- calculate the value |
| 126 | names = [] |
| 127 | if not hasattr(cls, "__slots__"): |
| 128 | # This class has no slots |
| 129 | pass |
| 130 | else: |
| 131 | # Slots found -- gather slot names from all base classes |
| 132 | for c in cls.__mro__: |
| 133 | if "__slots__" in c.__dict__: |
| 134 | slots = c.__dict__['__slots__'] |
| 135 | # if class has a single slot, it can be given as a string |
| 136 | if isinstance(slots, str): |
| 137 | slots = (slots,) |
| 138 | for name in slots: |
| 139 | # special descriptors |
| 140 | if name in ("__dict__", "__weakref__"): |
| 141 | continue |
| 142 | # mangled names |
| 143 | elif name.startswith('__') and not name.endswith('__'): |
| 144 | stripped = c.__name__.lstrip('_') |
| 145 | if stripped: |
| 146 | names.append('_%s%s' % (stripped, name)) |
| 147 | else: |
| 148 | names.append(name) |
| 149 | else: |
| 150 | names.append(name) |
| 151 | |
| 152 | # Cache the outcome in the class if at all possible |
| 153 | try: |
| 154 | cls.__slotnames__ = names |
| 155 | except: |
| 156 | pass # But don't die if we can't |
| 157 | |
| 158 | return names |
| 159 | |
| 160 | # A registry of extension codes. This is an ad-hoc compression |
| 161 | # mechanism. Whenever a global reference to <module>, <name> is about |
| 162 | # to be pickled, the (<module>, <name>) tuple is looked up here to see |
| 163 | # if it is a registered extension code for it. Extension codes are |
| 164 | # universal, so that the meaning of a pickle does not depend on |
| 165 | # context. (There are also some codes reserved for local use that |
| 166 | # don't have this restriction.) Codes are positive ints; 0 is |
| 167 | # reserved. |
| 168 | |
| 169 | _extension_registry = {} # key -> code |
| 170 | _inverted_registry = {} # code -> key |
| 171 | _extension_cache = {} # code -> object |
| 172 | # Don't ever rebind those names: pickling grabs a reference to them when |
| 173 | # it's initialized, and won't see a rebinding. |
| 174 | |
| 175 | def add_extension(module, name, code): |
| 176 | """Register an extension code.""" |
| 177 | code = int(code) |
| 178 | if not 1 <= code <= 0x7fffffff: |
| 179 | raise ValueError("code out of range") |
| 180 | key = (module, name) |
| 181 | if (_extension_registry.get(key) == code and |
| 182 | _inverted_registry.get(code) == key): |
| 183 | return # Redundant registrations are benign |
| 184 | if key in _extension_registry: |
| 185 | raise ValueError("key %s is already registered with code %s" % |
| 186 | (key, _extension_registry[key])) |
| 187 | if code in _inverted_registry: |
| 188 | raise ValueError("code %s is already in use for key %s" % |
| 189 | (code, _inverted_registry[code])) |
| 190 | _extension_registry[key] = code |
| 191 | _inverted_registry[code] = key |
| 192 | |
| 193 | def remove_extension(module, name, code): |
| 194 | """Unregister an extension code. For testing only.""" |
| 195 | key = (module, name) |
| 196 | if (_extension_registry.get(key) != code or |
| 197 | _inverted_registry.get(code) != key): |
| 198 | raise ValueError("key %s is not registered with code %s" % |
| 199 | (key, code)) |
| 200 | del _extension_registry[key] |
| 201 | del _inverted_registry[code] |
| 202 | if code in _extension_cache: |
| 203 | del _extension_cache[code] |
| 204 | |
| 205 | def clear_extension_cache(): |
| 206 | _extension_cache.clear() |
| 207 | |
| 208 | # Standard extension code assignments |
| 209 | |
| 210 | # Reserved ranges |
| 211 | |
| 212 | # First Last Count Purpose |
| 213 | # 1 127 127 Reserved for Python standard library |
| 214 | # 128 191 64 Reserved for Zope |
| 215 | # 192 239 48 Reserved for 3rd parties |
| 216 | # 240 255 16 Reserved for private use (will never be assigned) |
| 217 | # 256 Inf Inf Reserved for future assignment |
| 218 | |
| 219 | # Extension codes are assigned by the Python Software Foundation. |