Commit | Line | Data |
---|---|---|
920dae64 AT |
1 | """Create portable serialized representations of Python objects. |
2 | ||
3 | See module cPickle for a (much) faster implementation. | |
4 | See module copy_reg for a mechanism for registering custom picklers. | |
5 | See module pickletools source for extensive comments. | |
6 | ||
7 | Classes: | |
8 | ||
9 | Pickler | |
10 | Unpickler | |
11 | ||
12 | Functions: | |
13 | ||
14 | dump(object, file) | |
15 | dumps(object) -> string | |
16 | load(file) -> object | |
17 | loads(string) -> object | |
18 | ||
19 | Misc variables: | |
20 | ||
21 | __version__ | |
22 | format_version | |
23 | compatible_formats | |
24 | ||
25 | """ | |
26 | ||
27 | __version__ = "$Revision: 1.158 $" # Code version | |
28 | ||
29 | from types import * | |
30 | from copy_reg import dispatch_table | |
31 | from copy_reg import _extension_registry, _inverted_registry, _extension_cache | |
32 | import marshal | |
33 | import sys | |
34 | import struct | |
35 | import re | |
36 | import warnings | |
37 | ||
38 | __all__ = ["PickleError", "PicklingError", "UnpicklingError", "Pickler", | |
39 | "Unpickler", "dump", "dumps", "load", "loads"] | |
40 | ||
41 | # These are purely informational; no code uses these. | |
42 | format_version = "2.0" # File format version we write | |
43 | compatible_formats = ["1.0", # Original protocol 0 | |
44 | "1.1", # Protocol 0 with INST added | |
45 | "1.2", # Original protocol 1 | |
46 | "1.3", # Protocol 1 with BINFLOAT added | |
47 | "2.0", # Protocol 2 | |
48 | ] # Old format versions we can read | |
49 | ||
50 | # Keep in synch with cPickle. This is the highest protocol number we | |
51 | # know how to read. | |
52 | HIGHEST_PROTOCOL = 2 | |
53 | ||
54 | # Why use struct.pack() for pickling but marshal.loads() for | |
55 | # unpickling? struct.pack() is 40% faster than marshal.dumps(), but | |
56 | # marshal.loads() is twice as fast as struct.unpack()! | |
57 | mloads = marshal.loads | |
58 | ||
59 | class PickleError(Exception): | |
60 | """A common base class for the other pickling exceptions.""" | |
61 | pass | |
62 | ||
63 | class PicklingError(PickleError): | |
64 | """This exception is raised when an unpicklable object is passed to the | |
65 | dump() method. | |
66 | ||
67 | """ | |
68 | pass | |
69 | ||
70 | class UnpicklingError(PickleError): | |
71 | """This exception is raised when there is a problem unpickling an object, | |
72 | such as a security violation. | |
73 | ||
74 | Note that other exceptions may also be raised during unpickling, including | |
75 | (but not necessarily limited to) AttributeError, EOFError, ImportError, | |
76 | and IndexError. | |
77 | ||
78 | """ | |
79 | pass | |
80 | ||
81 | # An instance of _Stop is raised by Unpickler.load_stop() in response to | |
82 | # the STOP opcode, passing the object that is the result of unpickling. | |
83 | class _Stop(Exception): | |
84 | def __init__(self, value): | |
85 | self.value = value | |
86 | ||
87 | # Jython has PyStringMap; it's a dict subclass with string keys | |
88 | try: | |
89 | from org.python.core import PyStringMap | |
90 | except ImportError: | |
91 | PyStringMap = None | |
92 | ||
93 | # UnicodeType may or may not be exported (normally imported from types) | |
94 | try: | |
95 | UnicodeType | |
96 | except NameError: | |
97 | UnicodeType = None | |
98 | ||
99 | # Pickle opcodes. See pickletools.py for extensive docs. The listing | |
100 | # here is in kind-of alphabetical order of 1-character pickle code. | |
101 | # pickletools groups them by purpose. | |
102 | ||
103 | MARK = '(' # push special markobject on stack | |
104 | STOP = '.' # every pickle ends with STOP | |
105 | POP = '0' # discard topmost stack item | |
106 | POP_MARK = '1' # discard stack top through topmost markobject | |
107 | DUP = '2' # duplicate top stack item | |
108 | FLOAT = 'F' # push float object; decimal string argument | |
109 | INT = 'I' # push integer or bool; decimal string argument | |
110 | BININT = 'J' # push four-byte signed int | |
111 | BININT1 = 'K' # push 1-byte unsigned int | |
112 | LONG = 'L' # push long; decimal string argument | |
113 | BININT2 = 'M' # push 2-byte unsigned int | |
114 | NONE = 'N' # push None | |
115 | PERSID = 'P' # push persistent object; id is taken from string arg | |
116 | BINPERSID = 'Q' # " " " ; " " " " stack | |
117 | REDUCE = 'R' # apply callable to argtuple, both on stack | |
118 | STRING = 'S' # push string; NL-terminated string argument | |
119 | BINSTRING = 'T' # push string; counted binary string argument | |
120 | SHORT_BINSTRING = 'U' # " " ; " " " " < 256 bytes | |
121 | UNICODE = 'V' # push Unicode string; raw-unicode-escaped'd argument | |
122 | BINUNICODE = 'X' # " " " ; counted UTF-8 string argument | |
123 | APPEND = 'a' # append stack top to list below it | |
124 | BUILD = 'b' # call __setstate__ or __dict__.update() | |
125 | GLOBAL = 'c' # push self.find_class(modname, name); 2 string args | |
126 | DICT = 'd' # build a dict from stack items | |
127 | EMPTY_DICT = '}' # push empty dict | |
128 | APPENDS = 'e' # extend list on stack by topmost stack slice | |
129 | GET = 'g' # push item from memo on stack; index is string arg | |
130 | BINGET = 'h' # " " " " " " ; " " 1-byte arg | |
131 | INST = 'i' # build & push class instance | |
132 | LONG_BINGET = 'j' # push item from memo on stack; index is 4-byte arg | |
133 | LIST = 'l' # build list from topmost stack items | |
134 | EMPTY_LIST = ']' # push empty list | |
135 | OBJ = 'o' # build & push class instance | |
136 | PUT = 'p' # store stack top in memo; index is string arg | |
137 | BINPUT = 'q' # " " " " " ; " " 1-byte arg | |
138 | LONG_BINPUT = 'r' # " " " " " ; " " 4-byte arg | |
139 | SETITEM = 's' # add key+value pair to dict | |
140 | TUPLE = 't' # build tuple from topmost stack items | |
141 | EMPTY_TUPLE = ')' # push empty tuple | |
142 | SETITEMS = 'u' # modify dict by adding topmost key+value pairs | |
143 | BINFLOAT = 'G' # push float; arg is 8-byte float encoding | |
144 | ||
145 | TRUE = 'I01\n' # not an opcode; see INT docs in pickletools.py | |
146 | FALSE = 'I00\n' # not an opcode; see INT docs in pickletools.py | |
147 | ||
148 | # Protocol 2 | |
149 | ||
150 | PROTO = '\x80' # identify pickle protocol | |
151 | NEWOBJ = '\x81' # build object by applying cls.__new__ to argtuple | |
152 | EXT1 = '\x82' # push object from extension registry; 1-byte index | |
153 | EXT2 = '\x83' # ditto, but 2-byte index | |
154 | EXT4 = '\x84' # ditto, but 4-byte index | |
155 | TUPLE1 = '\x85' # build 1-tuple from stack top | |
156 | TUPLE2 = '\x86' # build 2-tuple from two topmost stack items | |
157 | TUPLE3 = '\x87' # build 3-tuple from three topmost stack items | |
158 | NEWTRUE = '\x88' # push True | |
159 | NEWFALSE = '\x89' # push False | |
160 | LONG1 = '\x8a' # push long from < 256 bytes | |
161 | LONG4 = '\x8b' # push really big long | |
162 | ||
163 | _tuplesize2code = [EMPTY_TUPLE, TUPLE1, TUPLE2, TUPLE3] | |
164 | ||
165 | ||
166 | __all__.extend([x for x in dir() if re.match("[A-Z][A-Z0-9_]+$",x)]) | |
167 | del x | |
168 | ||
169 | ||
170 | # Pickling machinery | |
171 | ||
172 | class Pickler: | |
173 | ||
174 | def __init__(self, file, protocol=None, bin=None): | |
175 | """This takes a file-like object for writing a pickle data stream. | |
176 | ||
177 | The optional protocol argument tells the pickler to use the | |
178 | given protocol; supported protocols are 0, 1, 2. The default | |
179 | protocol is 0, to be backwards compatible. (Protocol 0 is the | |
180 | only protocol that can be written to a file opened in text | |
181 | mode and read back successfully. When using a protocol higher | |
182 | than 0, make sure the file is opened in binary mode, both when | |
183 | pickling and unpickling.) | |
184 | ||
185 | Protocol 1 is more efficient than protocol 0; protocol 2 is | |
186 | more efficient than protocol 1. | |
187 | ||
188 | Specifying a negative protocol version selects the highest | |
189 | protocol version supported. The higher the protocol used, the | |
190 | more recent the version of Python needed to read the pickle | |
191 | produced. | |
192 | ||
193 | The file parameter must have a write() method that accepts a single | |
194 | string argument. It can thus be an open file object, a StringIO | |
195 | object, or any other custom object that meets this interface. | |
196 | ||
197 | """ | |
198 | if protocol is not None and bin is not None: | |
199 | raise ValueError, "can't specify both 'protocol' and 'bin'" | |
200 | if bin is not None: | |
201 | warnings.warn("The 'bin' argument to Pickler() is deprecated", | |
202 | DeprecationWarning) | |
203 | protocol = bin | |
204 | if protocol is None: | |
205 | protocol = 0 | |
206 | if protocol < 0: | |
207 | protocol = HIGHEST_PROTOCOL | |
208 | elif not 0 <= protocol <= HIGHEST_PROTOCOL: | |
209 | raise ValueError("pickle protocol must be <= %d" % HIGHEST_PROTOCOL) | |
210 | self.write = file.write | |
211 | self.memo = {} | |
212 | self.proto = int(protocol) | |
213 | self.bin = protocol >= 1 | |
214 | self.fast = 0 | |
215 | ||
216 | def clear_memo(self): | |
217 | """Clears the pickler's "memo". | |
218 | ||
219 | The memo is the data structure that remembers which objects the | |
220 | pickler has already seen, so that shared or recursive objects are | |
221 | pickled by reference and not by value. This method is useful when | |
222 | re-using picklers. | |
223 | ||
224 | """ | |
225 | self.memo.clear() | |
226 | ||
227 | def dump(self, obj): | |
228 | """Write a pickled representation of obj to the open file.""" | |
229 | if self.proto >= 2: | |
230 | self.write(PROTO + chr(self.proto)) | |
231 | self.save(obj) | |
232 | self.write(STOP) | |
233 | ||
234 | def memoize(self, obj): | |
235 | """Store an object in the memo.""" | |
236 | ||
237 | # The Pickler memo is a dictionary mapping object ids to 2-tuples | |
238 | # that contain the Unpickler memo key and the object being memoized. | |
239 | # The memo key is written to the pickle and will become | |
240 | # the key in the Unpickler's memo. The object is stored in the | |
241 | # Pickler memo so that transient objects are kept alive during | |
242 | # pickling. | |
243 | ||
244 | # The use of the Unpickler memo length as the memo key is just a | |
245 | # convention. The only requirement is that the memo values be unique. | |
246 | # But there appears no advantage to any other scheme, and this | |
247 | # scheme allows the Unpickler memo to be implemented as a plain (but | |
248 | # growable) array, indexed by memo key. | |
249 | if self.fast: | |
250 | return | |
251 | assert id(obj) not in self.memo | |
252 | memo_len = len(self.memo) | |
253 | self.write(self.put(memo_len)) | |
254 | self.memo[id(obj)] = memo_len, obj | |
255 | ||
256 | # Return a PUT (BINPUT, LONG_BINPUT) opcode string, with argument i. | |
257 | def put(self, i, pack=struct.pack): | |
258 | if self.bin: | |
259 | if i < 256: | |
260 | return BINPUT + chr(i) | |
261 | else: | |
262 | return LONG_BINPUT + pack("<i", i) | |
263 | ||
264 | return PUT + repr(i) + '\n' | |
265 | ||
266 | # Return a GET (BINGET, LONG_BINGET) opcode string, with argument i. | |
267 | def get(self, i, pack=struct.pack): | |
268 | if self.bin: | |
269 | if i < 256: | |
270 | return BINGET + chr(i) | |
271 | else: | |
272 | return LONG_BINGET + pack("<i", i) | |
273 | ||
274 | return GET + repr(i) + '\n' | |
275 | ||
276 | def save(self, obj): | |
277 | # Check for persistent id (defined by a subclass) | |
278 | pid = self.persistent_id(obj) | |
279 | if pid: | |
280 | self.save_pers(pid) | |
281 | return | |
282 | ||
283 | # Check the memo | |
284 | x = self.memo.get(id(obj)) | |
285 | if x: | |
286 | self.write(self.get(x[0])) | |
287 | return | |
288 | ||
289 | # Check the type dispatch table | |
290 | t = type(obj) | |
291 | f = self.dispatch.get(t) | |
292 | if f: | |
293 | f(self, obj) # Call unbound method with explicit self | |
294 | return | |
295 | ||
296 | # Check for a class with a custom metaclass; treat as regular class | |
297 | try: | |
298 | issc = issubclass(t, TypeType) | |
299 | except TypeError: # t is not a class (old Boost; see SF #502085) | |
300 | issc = 0 | |
301 | if issc: | |
302 | self.save_global(obj) | |
303 | return | |
304 | ||
305 | # Check copy_reg.dispatch_table | |
306 | reduce = dispatch_table.get(t) | |
307 | if reduce: | |
308 | rv = reduce(obj) | |
309 | else: | |
310 | # Check for a __reduce_ex__ method, fall back to __reduce__ | |
311 | reduce = getattr(obj, "__reduce_ex__", None) | |
312 | if reduce: | |
313 | rv = reduce(self.proto) | |
314 | else: | |
315 | reduce = getattr(obj, "__reduce__", None) | |
316 | if reduce: | |
317 | rv = reduce() | |
318 | else: | |
319 | raise PicklingError("Can't pickle %r object: %r" % | |
320 | (t.__name__, obj)) | |
321 | ||
322 | # Check for string returned by reduce(), meaning "save as global" | |
323 | if type(rv) is StringType: | |
324 | self.save_global(obj, rv) | |
325 | return | |
326 | ||
327 | # Assert that reduce() returned a tuple | |
328 | if type(rv) is not TupleType: | |
329 | raise PicklingError("%s must return string or tuple" % reduce) | |
330 | ||
331 | # Assert that it returned an appropriately sized tuple | |
332 | l = len(rv) | |
333 | if not (2 <= l <= 5): | |
334 | raise PicklingError("Tuple returned by %s must have " | |
335 | "two to five elements" % reduce) | |
336 | ||
337 | # Save the reduce() output and finally memoize the object | |
338 | self.save_reduce(obj=obj, *rv) | |
339 | ||
340 | def persistent_id(self, obj): | |
341 | # This exists so a subclass can override it | |
342 | return None | |
343 | ||
344 | def save_pers(self, pid): | |
345 | # Save a persistent id reference | |
346 | if self.bin: | |
347 | self.save(pid) | |
348 | self.write(BINPERSID) | |
349 | else: | |
350 | self.write(PERSID + str(pid) + '\n') | |
351 | ||
352 | def save_reduce(self, func, args, state=None, | |
353 | listitems=None, dictitems=None, obj=None): | |
354 | # This API is called by some subclasses | |
355 | ||
356 | # Assert that args is a tuple or None | |
357 | if not isinstance(args, TupleType): | |
358 | if args is None: | |
359 | # A hack for Jim Fulton's ExtensionClass, now deprecated. | |
360 | # See load_reduce() | |
361 | warnings.warn("__basicnew__ special case is deprecated", | |
362 | DeprecationWarning) | |
363 | else: | |
364 | raise PicklingError( | |
365 | "args from reduce() should be a tuple") | |
366 | ||
367 | # Assert that func is callable | |
368 | if not callable(func): | |
369 | raise PicklingError("func from reduce should be callable") | |
370 | ||
371 | save = self.save | |
372 | write = self.write | |
373 | ||
374 | # Protocol 2 special case: if func's name is __newobj__, use NEWOBJ | |
375 | if self.proto >= 2 and getattr(func, "__name__", "") == "__newobj__": | |
376 | # A __reduce__ implementation can direct protocol 2 to | |
377 | # use the more efficient NEWOBJ opcode, while still | |
378 | # allowing protocol 0 and 1 to work normally. For this to | |
379 | # work, the function returned by __reduce__ should be | |
380 | # called __newobj__, and its first argument should be a | |
381 | # new-style class. The implementation for __newobj__ | |
382 | # should be as follows, although pickle has no way to | |
383 | # verify this: | |
384 | # | |
385 | # def __newobj__(cls, *args): | |
386 | # return cls.__new__(cls, *args) | |
387 | # | |
388 | # Protocols 0 and 1 will pickle a reference to __newobj__, | |
389 | # while protocol 2 (and above) will pickle a reference to | |
390 | # cls, the remaining args tuple, and the NEWOBJ code, | |
391 | # which calls cls.__new__(cls, *args) at unpickling time | |
392 | # (see load_newobj below). If __reduce__ returns a | |
393 | # three-tuple, the state from the third tuple item will be | |
394 | # pickled regardless of the protocol, calling __setstate__ | |
395 | # at unpickling time (see load_build below). | |
396 | # | |
397 | # Note that no standard __newobj__ implementation exists; | |
398 | # you have to provide your own. This is to enforce | |
399 | # compatibility with Python 2.2 (pickles written using | |
400 | # protocol 0 or 1 in Python 2.3 should be unpicklable by | |
401 | # Python 2.2). | |
402 | cls = args[0] | |
403 | if not hasattr(cls, "__new__"): | |
404 | raise PicklingError( | |
405 | "args[0] from __newobj__ args has no __new__") | |
406 | if obj is not None and cls is not obj.__class__: | |
407 | raise PicklingError( | |
408 | "args[0] from __newobj__ args has the wrong class") | |
409 | args = args[1:] | |
410 | save(cls) | |
411 | save(args) | |
412 | write(NEWOBJ) | |
413 | else: | |
414 | save(func) | |
415 | save(args) | |
416 | write(REDUCE) | |
417 | ||
418 | if obj is not None: | |
419 | self.memoize(obj) | |
420 | ||
421 | # More new special cases (that work with older protocols as | |
422 | # well): when __reduce__ returns a tuple with 4 or 5 items, | |
423 | # the 4th and 5th item should be iterators that provide list | |
424 | # items and dict items (as (key, value) tuples), or None. | |
425 | ||
426 | if listitems is not None: | |
427 | self._batch_appends(listitems) | |
428 | ||
429 | if dictitems is not None: | |
430 | self._batch_setitems(dictitems) | |
431 | ||
432 | if state is not None: | |
433 | save(state) | |
434 | write(BUILD) | |
435 | ||
436 | # Methods below this point are dispatched through the dispatch table | |
437 | ||
438 | dispatch = {} | |
439 | ||
440 | def save_none(self, obj): | |
441 | self.write(NONE) | |
442 | dispatch[NoneType] = save_none | |
443 | ||
444 | def save_bool(self, obj): | |
445 | if self.proto >= 2: | |
446 | self.write(obj and NEWTRUE or NEWFALSE) | |
447 | else: | |
448 | self.write(obj and TRUE or FALSE) | |
449 | dispatch[bool] = save_bool | |
450 | ||
451 | def save_int(self, obj, pack=struct.pack): | |
452 | if self.bin: | |
453 | # If the int is small enough to fit in a signed 4-byte 2's-comp | |
454 | # format, we can store it more efficiently than the general | |
455 | # case. | |
456 | # First one- and two-byte unsigned ints: | |
457 | if obj >= 0: | |
458 | if obj <= 0xff: | |
459 | self.write(BININT1 + chr(obj)) | |
460 | return | |
461 | if obj <= 0xffff: | |
462 | self.write("%c%c%c" % (BININT2, obj&0xff, obj>>8)) | |
463 | return | |
464 | # Next check for 4-byte signed ints: | |
465 | high_bits = obj >> 31 # note that Python shift sign-extends | |
466 | if high_bits == 0 or high_bits == -1: | |
467 | # All high bits are copies of bit 2**31, so the value | |
468 | # fits in a 4-byte signed int. | |
469 | self.write(BININT + pack("<i", obj)) | |
470 | return | |
471 | # Text pickle, or int too big to fit in signed 4-byte format. | |
472 | self.write(INT + repr(obj) + '\n') | |
473 | dispatch[IntType] = save_int | |
474 | ||
475 | def save_long(self, obj, pack=struct.pack): | |
476 | if self.proto >= 2: | |
477 | bytes = encode_long(obj) | |
478 | n = len(bytes) | |
479 | if n < 256: | |
480 | self.write(LONG1 + chr(n) + bytes) | |
481 | else: | |
482 | self.write(LONG4 + pack("<i", n) + bytes) | |
483 | return | |
484 | self.write(LONG + repr(obj) + '\n') | |
485 | dispatch[LongType] = save_long | |
486 | ||
487 | def save_float(self, obj, pack=struct.pack): | |
488 | if self.bin: | |
489 | self.write(BINFLOAT + pack('>d', obj)) | |
490 | else: | |
491 | self.write(FLOAT + repr(obj) + '\n') | |
492 | dispatch[FloatType] = save_float | |
493 | ||
494 | def save_string(self, obj, pack=struct.pack): | |
495 | if self.bin: | |
496 | n = len(obj) | |
497 | if n < 256: | |
498 | self.write(SHORT_BINSTRING + chr(n) + obj) | |
499 | else: | |
500 | self.write(BINSTRING + pack("<i", n) + obj) | |
501 | else: | |
502 | self.write(STRING + repr(obj) + '\n') | |
503 | self.memoize(obj) | |
504 | dispatch[StringType] = save_string | |
505 | ||
506 | def save_unicode(self, obj, pack=struct.pack): | |
507 | if self.bin: | |
508 | encoding = obj.encode('utf-8') | |
509 | n = len(encoding) | |
510 | self.write(BINUNICODE + pack("<i", n) + encoding) | |
511 | else: | |
512 | obj = obj.replace("\\", "\\u005c") | |
513 | obj = obj.replace("\n", "\\u000a") | |
514 | self.write(UNICODE + obj.encode('raw-unicode-escape') + '\n') | |
515 | self.memoize(obj) | |
516 | dispatch[UnicodeType] = save_unicode | |
517 | ||
518 | if StringType == UnicodeType: | |
519 | # This is true for Jython | |
520 | def save_string(self, obj, pack=struct.pack): | |
521 | unicode = obj.isunicode() | |
522 | ||
523 | if self.bin: | |
524 | if unicode: | |
525 | obj = obj.encode("utf-8") | |
526 | l = len(obj) | |
527 | if l < 256 and not unicode: | |
528 | self.write(SHORT_BINSTRING + chr(l) + obj) | |
529 | else: | |
530 | s = pack("<i", l) | |
531 | if unicode: | |
532 | self.write(BINUNICODE + s + obj) | |
533 | else: | |
534 | self.write(BINSTRING + s + obj) | |
535 | else: | |
536 | if unicode: | |
537 | obj = obj.replace("\\", "\\u005c") | |
538 | obj = obj.replace("\n", "\\u000a") | |
539 | obj = obj.encode('raw-unicode-escape') | |
540 | self.write(UNICODE + obj + '\n') | |
541 | else: | |
542 | self.write(STRING + repr(obj) + '\n') | |
543 | self.memoize(obj) | |
544 | dispatch[StringType] = save_string | |
545 | ||
546 | def save_tuple(self, obj): | |
547 | write = self.write | |
548 | proto = self.proto | |
549 | ||
550 | n = len(obj) | |
551 | if n == 0: | |
552 | if proto: | |
553 | write(EMPTY_TUPLE) | |
554 | else: | |
555 | write(MARK + TUPLE) | |
556 | return | |
557 | ||
558 | save = self.save | |
559 | memo = self.memo | |
560 | if n <= 3 and proto >= 2: | |
561 | for element in obj: | |
562 | save(element) | |
563 | # Subtle. Same as in the big comment below. | |
564 | if id(obj) in memo: | |
565 | get = self.get(memo[id(obj)][0]) | |
566 | write(POP * n + get) | |
567 | else: | |
568 | write(_tuplesize2code[n]) | |
569 | self.memoize(obj) | |
570 | return | |
571 | ||
572 | # proto 0 or proto 1 and tuple isn't empty, or proto > 1 and tuple | |
573 | # has more than 3 elements. | |
574 | write(MARK) | |
575 | for element in obj: | |
576 | save(element) | |
577 | ||
578 | if id(obj) in memo: | |
579 | # Subtle. d was not in memo when we entered save_tuple(), so | |
580 | # the process of saving the tuple's elements must have saved | |
581 | # the tuple itself: the tuple is recursive. The proper action | |
582 | # now is to throw away everything we put on the stack, and | |
583 | # simply GET the tuple (it's already constructed). This check | |
584 | # could have been done in the "for element" loop instead, but | |
585 | # recursive tuples are a rare thing. | |
586 | get = self.get(memo[id(obj)][0]) | |
587 | if proto: | |
588 | write(POP_MARK + get) | |
589 | else: # proto 0 -- POP_MARK not available | |
590 | write(POP * (n+1) + get) | |
591 | return | |
592 | ||
593 | # No recursion. | |
594 | self.write(TUPLE) | |
595 | self.memoize(obj) | |
596 | ||
597 | dispatch[TupleType] = save_tuple | |
598 | ||
599 | # save_empty_tuple() isn't used by anything in Python 2.3. However, I | |
600 | # found a Pickler subclass in Zope3 that calls it, so it's not harmless | |
601 | # to remove it. | |
602 | def save_empty_tuple(self, obj): | |
603 | self.write(EMPTY_TUPLE) | |
604 | ||
605 | def save_list(self, obj): | |
606 | write = self.write | |
607 | ||
608 | if self.bin: | |
609 | write(EMPTY_LIST) | |
610 | else: # proto 0 -- can't use EMPTY_LIST | |
611 | write(MARK + LIST) | |
612 | ||
613 | self.memoize(obj) | |
614 | self._batch_appends(iter(obj)) | |
615 | ||
616 | dispatch[ListType] = save_list | |
617 | ||
618 | # Keep in synch with cPickle's BATCHSIZE. Nothing will break if it gets | |
619 | # out of synch, though. | |
620 | _BATCHSIZE = 1000 | |
621 | ||
622 | def _batch_appends(self, items): | |
623 | # Helper to batch up APPENDS sequences | |
624 | save = self.save | |
625 | write = self.write | |
626 | ||
627 | if not self.bin: | |
628 | for x in items: | |
629 | save(x) | |
630 | write(APPEND) | |
631 | return | |
632 | ||
633 | r = xrange(self._BATCHSIZE) | |
634 | while items is not None: | |
635 | tmp = [] | |
636 | for i in r: | |
637 | try: | |
638 | x = items.next() | |
639 | tmp.append(x) | |
640 | except StopIteration: | |
641 | items = None | |
642 | break | |
643 | n = len(tmp) | |
644 | if n > 1: | |
645 | write(MARK) | |
646 | for x in tmp: | |
647 | save(x) | |
648 | write(APPENDS) | |
649 | elif n: | |
650 | save(tmp[0]) | |
651 | write(APPEND) | |
652 | # else tmp is empty, and we're done | |
653 | ||
654 | def save_dict(self, obj): | |
655 | write = self.write | |
656 | ||
657 | if self.bin: | |
658 | write(EMPTY_DICT) | |
659 | else: # proto 0 -- can't use EMPTY_DICT | |
660 | write(MARK + DICT) | |
661 | ||
662 | self.memoize(obj) | |
663 | self._batch_setitems(obj.iteritems()) | |
664 | ||
665 | dispatch[DictionaryType] = save_dict | |
666 | if not PyStringMap is None: | |
667 | dispatch[PyStringMap] = save_dict | |
668 | ||
669 | def _batch_setitems(self, items): | |
670 | # Helper to batch up SETITEMS sequences; proto >= 1 only | |
671 | save = self.save | |
672 | write = self.write | |
673 | ||
674 | if not self.bin: | |
675 | for k, v in items: | |
676 | save(k) | |
677 | save(v) | |
678 | write(SETITEM) | |
679 | return | |
680 | ||
681 | r = xrange(self._BATCHSIZE) | |
682 | while items is not None: | |
683 | tmp = [] | |
684 | for i in r: | |
685 | try: | |
686 | tmp.append(items.next()) | |
687 | except StopIteration: | |
688 | items = None | |
689 | break | |
690 | n = len(tmp) | |
691 | if n > 1: | |
692 | write(MARK) | |
693 | for k, v in tmp: | |
694 | save(k) | |
695 | save(v) | |
696 | write(SETITEMS) | |
697 | elif n: | |
698 | k, v = tmp[0] | |
699 | save(k) | |
700 | save(v) | |
701 | write(SETITEM) | |
702 | # else tmp is empty, and we're done | |
703 | ||
704 | def save_inst(self, obj): | |
705 | cls = obj.__class__ | |
706 | ||
707 | memo = self.memo | |
708 | write = self.write | |
709 | save = self.save | |
710 | ||
711 | if hasattr(obj, '__getinitargs__'): | |
712 | args = obj.__getinitargs__() | |
713 | len(args) # XXX Assert it's a sequence | |
714 | _keep_alive(args, memo) | |
715 | else: | |
716 | args = () | |
717 | ||
718 | write(MARK) | |
719 | ||
720 | if self.bin: | |
721 | save(cls) | |
722 | for arg in args: | |
723 | save(arg) | |
724 | write(OBJ) | |
725 | else: | |
726 | for arg in args: | |
727 | save(arg) | |
728 | write(INST + cls.__module__ + '\n' + cls.__name__ + '\n') | |
729 | ||
730 | self.memoize(obj) | |
731 | ||
732 | try: | |
733 | getstate = obj.__getstate__ | |
734 | except AttributeError: | |
735 | stuff = obj.__dict__ | |
736 | else: | |
737 | stuff = getstate() | |
738 | _keep_alive(stuff, memo) | |
739 | save(stuff) | |
740 | write(BUILD) | |
741 | ||
742 | dispatch[InstanceType] = save_inst | |
743 | ||
744 | def save_global(self, obj, name=None, pack=struct.pack): | |
745 | write = self.write | |
746 | memo = self.memo | |
747 | ||
748 | if name is None: | |
749 | name = obj.__name__ | |
750 | ||
751 | module = getattr(obj, "__module__", None) | |
752 | if module is None: | |
753 | module = whichmodule(obj, name) | |
754 | ||
755 | try: | |
756 | __import__(module) | |
757 | mod = sys.modules[module] | |
758 | klass = getattr(mod, name) | |
759 | except (ImportError, KeyError, AttributeError): | |
760 | raise PicklingError( | |
761 | "Can't pickle %r: it's not found as %s.%s" % | |
762 | (obj, module, name)) | |
763 | else: | |
764 | if klass is not obj: | |
765 | raise PicklingError( | |
766 | "Can't pickle %r: it's not the same object as %s.%s" % | |
767 | (obj, module, name)) | |
768 | ||
769 | if self.proto >= 2: | |
770 | code = _extension_registry.get((module, name)) | |
771 | if code: | |
772 | assert code > 0 | |
773 | if code <= 0xff: | |
774 | write(EXT1 + chr(code)) | |
775 | elif code <= 0xffff: | |
776 | write("%c%c%c" % (EXT2, code&0xff, code>>8)) | |
777 | else: | |
778 | write(EXT4 + pack("<i", code)) | |
779 | return | |
780 | ||
781 | write(GLOBAL + module + '\n' + name + '\n') | |
782 | self.memoize(obj) | |
783 | ||
784 | dispatch[ClassType] = save_global | |
785 | dispatch[FunctionType] = save_global | |
786 | dispatch[BuiltinFunctionType] = save_global | |
787 | dispatch[TypeType] = save_global | |
788 | ||
789 | # Pickling helpers | |
790 | ||
791 | def _keep_alive(x, memo): | |
792 | """Keeps a reference to the object x in the memo. | |
793 | ||
794 | Because we remember objects by their id, we have | |
795 | to assure that possibly temporary objects are kept | |
796 | alive by referencing them. | |
797 | We store a reference at the id of the memo, which should | |
798 | normally not be used unless someone tries to deepcopy | |
799 | the memo itself... | |
800 | """ | |
801 | try: | |
802 | memo[id(memo)].append(x) | |
803 | except KeyError: | |
804 | # aha, this is the first one :-) | |
805 | memo[id(memo)]=[x] | |
806 | ||
807 | ||
808 | # A cache for whichmodule(), mapping a function object to the name of | |
809 | # the module in which the function was found. | |
810 | ||
811 | classmap = {} # called classmap for backwards compatibility | |
812 | ||
813 | def whichmodule(func, funcname): | |
814 | """Figure out the module in which a function occurs. | |
815 | ||
816 | Search sys.modules for the module. | |
817 | Cache in classmap. | |
818 | Return a module name. | |
819 | If the function cannot be found, return "__main__". | |
820 | """ | |
821 | # Python functions should always get an __module__ from their globals. | |
822 | mod = getattr(func, "__module__", None) | |
823 | if mod is not None: | |
824 | return mod | |
825 | if func in classmap: | |
826 | return classmap[func] | |
827 | ||
828 | for name, module in sys.modules.items(): | |
829 | if module is None: | |
830 | continue # skip dummy package entries | |
831 | if name != '__main__' and getattr(module, funcname, None) is func: | |
832 | break | |
833 | else: | |
834 | name = '__main__' | |
835 | classmap[func] = name | |
836 | return name | |
837 | ||
838 | ||
839 | # Unpickling machinery | |
840 | ||
841 | class Unpickler: | |
842 | ||
843 | def __init__(self, file): | |
844 | """This takes a file-like object for reading a pickle data stream. | |
845 | ||
846 | The protocol version of the pickle is detected automatically, so no | |
847 | proto argument is needed. | |
848 | ||
849 | The file-like object must have two methods, a read() method that | |
850 | takes an integer argument, and a readline() method that requires no | |
851 | arguments. Both methods should return a string. Thus file-like | |
852 | object can be a file object opened for reading, a StringIO object, | |
853 | or any other custom object that meets this interface. | |
854 | """ | |
855 | self.readline = file.readline | |
856 | self.read = file.read | |
857 | self.memo = {} | |
858 | ||
859 | def load(self): | |
860 | """Read a pickled object representation from the open file. | |
861 | ||
862 | Return the reconstituted object hierarchy specified in the file. | |
863 | """ | |
864 | self.mark = object() # any new unique object | |
865 | self.stack = [] | |
866 | self.append = self.stack.append | |
867 | read = self.read | |
868 | dispatch = self.dispatch | |
869 | try: | |
870 | while 1: | |
871 | key = read(1) | |
872 | dispatch[key](self) | |
873 | except _Stop, stopinst: | |
874 | return stopinst.value | |
875 | ||
876 | # Return largest index k such that self.stack[k] is self.mark. | |
877 | # If the stack doesn't contain a mark, eventually raises IndexError. | |
878 | # This could be sped by maintaining another stack, of indices at which | |
879 | # the mark appears. For that matter, the latter stack would suffice, | |
880 | # and we wouldn't need to push mark objects on self.stack at all. | |
881 | # Doing so is probably a good thing, though, since if the pickle is | |
882 | # corrupt (or hostile) we may get a clue from finding self.mark embedded | |
883 | # in unpickled objects. | |
884 | def marker(self): | |
885 | stack = self.stack | |
886 | mark = self.mark | |
887 | k = len(stack)-1 | |
888 | while stack[k] is not mark: k = k-1 | |
889 | return k | |
890 | ||
891 | dispatch = {} | |
892 | ||
893 | def load_eof(self): | |
894 | raise EOFError | |
895 | dispatch[''] = load_eof | |
896 | ||
897 | def load_proto(self): | |
898 | proto = ord(self.read(1)) | |
899 | if not 0 <= proto <= 2: | |
900 | raise ValueError, "unsupported pickle protocol: %d" % proto | |
901 | dispatch[PROTO] = load_proto | |
902 | ||
903 | def load_persid(self): | |
904 | pid = self.readline()[:-1] | |
905 | self.append(self.persistent_load(pid)) | |
906 | dispatch[PERSID] = load_persid | |
907 | ||
908 | def load_binpersid(self): | |
909 | pid = self.stack.pop() | |
910 | self.append(self.persistent_load(pid)) | |
911 | dispatch[BINPERSID] = load_binpersid | |
912 | ||
913 | def load_none(self): | |
914 | self.append(None) | |
915 | dispatch[NONE] = load_none | |
916 | ||
917 | def load_false(self): | |
918 | self.append(False) | |
919 | dispatch[NEWFALSE] = load_false | |
920 | ||
921 | def load_true(self): | |
922 | self.append(True) | |
923 | dispatch[NEWTRUE] = load_true | |
924 | ||
925 | def load_int(self): | |
926 | data = self.readline() | |
927 | if data == FALSE[1:]: | |
928 | val = False | |
929 | elif data == TRUE[1:]: | |
930 | val = True | |
931 | else: | |
932 | try: | |
933 | val = int(data) | |
934 | except ValueError: | |
935 | val = long(data) | |
936 | self.append(val) | |
937 | dispatch[INT] = load_int | |
938 | ||
939 | def load_binint(self): | |
940 | self.append(mloads('i' + self.read(4))) | |
941 | dispatch[BININT] = load_binint | |
942 | ||
943 | def load_binint1(self): | |
944 | self.append(ord(self.read(1))) | |
945 | dispatch[BININT1] = load_binint1 | |
946 | ||
947 | def load_binint2(self): | |
948 | self.append(mloads('i' + self.read(2) + '\000\000')) | |
949 | dispatch[BININT2] = load_binint2 | |
950 | ||
951 | def load_long(self): | |
952 | self.append(long(self.readline()[:-1], 0)) | |
953 | dispatch[LONG] = load_long | |
954 | ||
955 | def load_long1(self): | |
956 | n = ord(self.read(1)) | |
957 | bytes = self.read(n) | |
958 | self.append(decode_long(bytes)) | |
959 | dispatch[LONG1] = load_long1 | |
960 | ||
961 | def load_long4(self): | |
962 | n = mloads('i' + self.read(4)) | |
963 | bytes = self.read(n) | |
964 | self.append(decode_long(bytes)) | |
965 | dispatch[LONG4] = load_long4 | |
966 | ||
967 | def load_float(self): | |
968 | self.append(float(self.readline()[:-1])) | |
969 | dispatch[FLOAT] = load_float | |
970 | ||
971 | def load_binfloat(self, unpack=struct.unpack): | |
972 | self.append(unpack('>d', self.read(8))[0]) | |
973 | dispatch[BINFLOAT] = load_binfloat | |
974 | ||
975 | def load_string(self): | |
976 | rep = self.readline()[:-1] | |
977 | for q in "\"'": # double or single quote | |
978 | if rep.startswith(q): | |
979 | if not rep.endswith(q): | |
980 | raise ValueError, "insecure string pickle" | |
981 | rep = rep[len(q):-len(q)] | |
982 | break | |
983 | else: | |
984 | raise ValueError, "insecure string pickle" | |
985 | self.append(rep.decode("string-escape")) | |
986 | dispatch[STRING] = load_string | |
987 | ||
988 | def load_binstring(self): | |
989 | len = mloads('i' + self.read(4)) | |
990 | self.append(self.read(len)) | |
991 | dispatch[BINSTRING] = load_binstring | |
992 | ||
993 | def load_unicode(self): | |
994 | self.append(unicode(self.readline()[:-1],'raw-unicode-escape')) | |
995 | dispatch[UNICODE] = load_unicode | |
996 | ||
997 | def load_binunicode(self): | |
998 | len = mloads('i' + self.read(4)) | |
999 | self.append(unicode(self.read(len),'utf-8')) | |
1000 | dispatch[BINUNICODE] = load_binunicode | |
1001 | ||
1002 | def load_short_binstring(self): | |
1003 | len = ord(self.read(1)) | |
1004 | self.append(self.read(len)) | |
1005 | dispatch[SHORT_BINSTRING] = load_short_binstring | |
1006 | ||
1007 | def load_tuple(self): | |
1008 | k = self.marker() | |
1009 | self.stack[k:] = [tuple(self.stack[k+1:])] | |
1010 | dispatch[TUPLE] = load_tuple | |
1011 | ||
1012 | def load_empty_tuple(self): | |
1013 | self.stack.append(()) | |
1014 | dispatch[EMPTY_TUPLE] = load_empty_tuple | |
1015 | ||
1016 | def load_tuple1(self): | |
1017 | self.stack[-1] = (self.stack[-1],) | |
1018 | dispatch[TUPLE1] = load_tuple1 | |
1019 | ||
1020 | def load_tuple2(self): | |
1021 | self.stack[-2:] = [(self.stack[-2], self.stack[-1])] | |
1022 | dispatch[TUPLE2] = load_tuple2 | |
1023 | ||
1024 | def load_tuple3(self): | |
1025 | self.stack[-3:] = [(self.stack[-3], self.stack[-2], self.stack[-1])] | |
1026 | dispatch[TUPLE3] = load_tuple3 | |
1027 | ||
1028 | def load_empty_list(self): | |
1029 | self.stack.append([]) | |
1030 | dispatch[EMPTY_LIST] = load_empty_list | |
1031 | ||
1032 | def load_empty_dictionary(self): | |
1033 | self.stack.append({}) | |
1034 | dispatch[EMPTY_DICT] = load_empty_dictionary | |
1035 | ||
1036 | def load_list(self): | |
1037 | k = self.marker() | |
1038 | self.stack[k:] = [self.stack[k+1:]] | |
1039 | dispatch[LIST] = load_list | |
1040 | ||
1041 | def load_dict(self): | |
1042 | k = self.marker() | |
1043 | d = {} | |
1044 | items = self.stack[k+1:] | |
1045 | for i in range(0, len(items), 2): | |
1046 | key = items[i] | |
1047 | value = items[i+1] | |
1048 | d[key] = value | |
1049 | self.stack[k:] = [d] | |
1050 | dispatch[DICT] = load_dict | |
1051 | ||
1052 | # INST and OBJ differ only in how they get a class object. It's not | |
1053 | # only sensible to do the rest in a common routine, the two routines | |
1054 | # previously diverged and grew different bugs. | |
1055 | # klass is the class to instantiate, and k points to the topmost mark | |
1056 | # object, following which are the arguments for klass.__init__. | |
1057 | def _instantiate(self, klass, k): | |
1058 | args = tuple(self.stack[k+1:]) | |
1059 | del self.stack[k:] | |
1060 | instantiated = 0 | |
1061 | if (not args and | |
1062 | type(klass) is ClassType and | |
1063 | not hasattr(klass, "__getinitargs__")): | |
1064 | try: | |
1065 | value = _EmptyClass() | |
1066 | value.__class__ = klass | |
1067 | instantiated = 1 | |
1068 | except RuntimeError: | |
1069 | # In restricted execution, assignment to inst.__class__ is | |
1070 | # prohibited | |
1071 | pass | |
1072 | if not instantiated: | |
1073 | try: | |
1074 | value = klass(*args) | |
1075 | except TypeError, err: | |
1076 | raise TypeError, "in constructor for %s: %s" % ( | |
1077 | klass.__name__, str(err)), sys.exc_info()[2] | |
1078 | self.append(value) | |
1079 | ||
1080 | def load_inst(self): | |
1081 | module = self.readline()[:-1] | |
1082 | name = self.readline()[:-1] | |
1083 | klass = self.find_class(module, name) | |
1084 | self._instantiate(klass, self.marker()) | |
1085 | dispatch[INST] = load_inst | |
1086 | ||
1087 | def load_obj(self): | |
1088 | # Stack is ... markobject classobject arg1 arg2 ... | |
1089 | k = self.marker() | |
1090 | klass = self.stack.pop(k+1) | |
1091 | self._instantiate(klass, k) | |
1092 | dispatch[OBJ] = load_obj | |
1093 | ||
1094 | def load_newobj(self): | |
1095 | args = self.stack.pop() | |
1096 | cls = self.stack[-1] | |
1097 | obj = cls.__new__(cls, *args) | |
1098 | self.stack[-1] = obj | |
1099 | dispatch[NEWOBJ] = load_newobj | |
1100 | ||
1101 | def load_global(self): | |
1102 | module = self.readline()[:-1] | |
1103 | name = self.readline()[:-1] | |
1104 | klass = self.find_class(module, name) | |
1105 | self.append(klass) | |
1106 | dispatch[GLOBAL] = load_global | |
1107 | ||
1108 | def load_ext1(self): | |
1109 | code = ord(self.read(1)) | |
1110 | self.get_extension(code) | |
1111 | dispatch[EXT1] = load_ext1 | |
1112 | ||
1113 | def load_ext2(self): | |
1114 | code = mloads('i' + self.read(2) + '\000\000') | |
1115 | self.get_extension(code) | |
1116 | dispatch[EXT2] = load_ext2 | |
1117 | ||
1118 | def load_ext4(self): | |
1119 | code = mloads('i' + self.read(4)) | |
1120 | self.get_extension(code) | |
1121 | dispatch[EXT4] = load_ext4 | |
1122 | ||
1123 | def get_extension(self, code): | |
1124 | nil = [] | |
1125 | obj = _extension_cache.get(code, nil) | |
1126 | if obj is not nil: | |
1127 | self.append(obj) | |
1128 | return | |
1129 | key = _inverted_registry.get(code) | |
1130 | if not key: | |
1131 | raise ValueError("unregistered extension code %d" % code) | |
1132 | obj = self.find_class(*key) | |
1133 | _extension_cache[code] = obj | |
1134 | self.append(obj) | |
1135 | ||
1136 | def find_class(self, module, name): | |
1137 | # Subclasses may override this | |
1138 | __import__(module) | |
1139 | mod = sys.modules[module] | |
1140 | klass = getattr(mod, name) | |
1141 | return klass | |
1142 | ||
1143 | def load_reduce(self): | |
1144 | stack = self.stack | |
1145 | args = stack.pop() | |
1146 | func = stack[-1] | |
1147 | if args is None: | |
1148 | # A hack for Jim Fulton's ExtensionClass, now deprecated | |
1149 | warnings.warn("__basicnew__ special case is deprecated", | |
1150 | DeprecationWarning) | |
1151 | value = func.__basicnew__() | |
1152 | else: | |
1153 | value = func(*args) | |
1154 | stack[-1] = value | |
1155 | dispatch[REDUCE] = load_reduce | |
1156 | ||
1157 | def load_pop(self): | |
1158 | del self.stack[-1] | |
1159 | dispatch[POP] = load_pop | |
1160 | ||
1161 | def load_pop_mark(self): | |
1162 | k = self.marker() | |
1163 | del self.stack[k:] | |
1164 | dispatch[POP_MARK] = load_pop_mark | |
1165 | ||
1166 | def load_dup(self): | |
1167 | self.append(self.stack[-1]) | |
1168 | dispatch[DUP] = load_dup | |
1169 | ||
1170 | def load_get(self): | |
1171 | self.append(self.memo[self.readline()[:-1]]) | |
1172 | dispatch[GET] = load_get | |
1173 | ||
1174 | def load_binget(self): | |
1175 | i = ord(self.read(1)) | |
1176 | self.append(self.memo[repr(i)]) | |
1177 | dispatch[BINGET] = load_binget | |
1178 | ||
1179 | def load_long_binget(self): | |
1180 | i = mloads('i' + self.read(4)) | |
1181 | self.append(self.memo[repr(i)]) | |
1182 | dispatch[LONG_BINGET] = load_long_binget | |
1183 | ||
1184 | def load_put(self): | |
1185 | self.memo[self.readline()[:-1]] = self.stack[-1] | |
1186 | dispatch[PUT] = load_put | |
1187 | ||
1188 | def load_binput(self): | |
1189 | i = ord(self.read(1)) | |
1190 | self.memo[repr(i)] = self.stack[-1] | |
1191 | dispatch[BINPUT] = load_binput | |
1192 | ||
1193 | def load_long_binput(self): | |
1194 | i = mloads('i' + self.read(4)) | |
1195 | self.memo[repr(i)] = self.stack[-1] | |
1196 | dispatch[LONG_BINPUT] = load_long_binput | |
1197 | ||
1198 | def load_append(self): | |
1199 | stack = self.stack | |
1200 | value = stack.pop() | |
1201 | list = stack[-1] | |
1202 | list.append(value) | |
1203 | dispatch[APPEND] = load_append | |
1204 | ||
1205 | def load_appends(self): | |
1206 | stack = self.stack | |
1207 | mark = self.marker() | |
1208 | list = stack[mark - 1] | |
1209 | list.extend(stack[mark + 1:]) | |
1210 | del stack[mark:] | |
1211 | dispatch[APPENDS] = load_appends | |
1212 | ||
1213 | def load_setitem(self): | |
1214 | stack = self.stack | |
1215 | value = stack.pop() | |
1216 | key = stack.pop() | |
1217 | dict = stack[-1] | |
1218 | dict[key] = value | |
1219 | dispatch[SETITEM] = load_setitem | |
1220 | ||
1221 | def load_setitems(self): | |
1222 | stack = self.stack | |
1223 | mark = self.marker() | |
1224 | dict = stack[mark - 1] | |
1225 | for i in range(mark + 1, len(stack), 2): | |
1226 | dict[stack[i]] = stack[i + 1] | |
1227 | ||
1228 | del stack[mark:] | |
1229 | dispatch[SETITEMS] = load_setitems | |
1230 | ||
1231 | def load_build(self): | |
1232 | stack = self.stack | |
1233 | state = stack.pop() | |
1234 | inst = stack[-1] | |
1235 | setstate = getattr(inst, "__setstate__", None) | |
1236 | if setstate: | |
1237 | setstate(state) | |
1238 | return | |
1239 | slotstate = None | |
1240 | if isinstance(state, tuple) and len(state) == 2: | |
1241 | state, slotstate = state | |
1242 | if state: | |
1243 | try: | |
1244 | inst.__dict__.update(state) | |
1245 | except RuntimeError: | |
1246 | # XXX In restricted execution, the instance's __dict__ | |
1247 | # is not accessible. Use the old way of unpickling | |
1248 | # the instance variables. This is a semantic | |
1249 | # difference when unpickling in restricted | |
1250 | # vs. unrestricted modes. | |
1251 | # Note, however, that cPickle has never tried to do the | |
1252 | # .update() business, and always uses | |
1253 | # PyObject_SetItem(inst.__dict__, key, value) in a | |
1254 | # loop over state.items(). | |
1255 | for k, v in state.items(): | |
1256 | setattr(inst, k, v) | |
1257 | if slotstate: | |
1258 | for k, v in slotstate.items(): | |
1259 | setattr(inst, k, v) | |
1260 | dispatch[BUILD] = load_build | |
1261 | ||
1262 | def load_mark(self): | |
1263 | self.append(self.mark) | |
1264 | dispatch[MARK] = load_mark | |
1265 | ||
1266 | def load_stop(self): | |
1267 | value = self.stack.pop() | |
1268 | raise _Stop(value) | |
1269 | dispatch[STOP] = load_stop | |
1270 | ||
1271 | # Helper class for load_inst/load_obj | |
1272 | ||
1273 | class _EmptyClass: | |
1274 | pass | |
1275 | ||
1276 | # Encode/decode longs in linear time. | |
1277 | ||
1278 | import binascii as _binascii | |
1279 | ||
1280 | def encode_long(x): | |
1281 | r"""Encode a long to a two's complement little-endian binary string. | |
1282 | Note that 0L is a special case, returning an empty string, to save a | |
1283 | byte in the LONG1 pickling context. | |
1284 | ||
1285 | >>> encode_long(0L) | |
1286 | '' | |
1287 | >>> encode_long(255L) | |
1288 | '\xff\x00' | |
1289 | >>> encode_long(32767L) | |
1290 | '\xff\x7f' | |
1291 | >>> encode_long(-256L) | |
1292 | '\x00\xff' | |
1293 | >>> encode_long(-32768L) | |
1294 | '\x00\x80' | |
1295 | >>> encode_long(-128L) | |
1296 | '\x80' | |
1297 | >>> encode_long(127L) | |
1298 | '\x7f' | |
1299 | >>> | |
1300 | """ | |
1301 | ||
1302 | if x == 0: | |
1303 | return '' | |
1304 | if x > 0: | |
1305 | ashex = hex(x) | |
1306 | assert ashex.startswith("0x") | |
1307 | njunkchars = 2 + ashex.endswith('L') | |
1308 | nibbles = len(ashex) - njunkchars | |
1309 | if nibbles & 1: | |
1310 | # need an even # of nibbles for unhexlify | |
1311 | ashex = "0x0" + ashex[2:] | |
1312 | elif int(ashex[2], 16) >= 8: | |
1313 | # "looks negative", so need a byte of sign bits | |
1314 | ashex = "0x00" + ashex[2:] | |
1315 | else: | |
1316 | # Build the 256's-complement: (1L << nbytes) + x. The trick is | |
1317 | # to find the number of bytes in linear time (although that should | |
1318 | # really be a constant-time task). | |
1319 | ashex = hex(-x) | |
1320 | assert ashex.startswith("0x") | |
1321 | njunkchars = 2 + ashex.endswith('L') | |
1322 | nibbles = len(ashex) - njunkchars | |
1323 | if nibbles & 1: | |
1324 | # Extend to a full byte. | |
1325 | nibbles += 1 | |
1326 | nbits = nibbles * 4 | |
1327 | x += 1L << nbits | |
1328 | assert x > 0 | |
1329 | ashex = hex(x) | |
1330 | njunkchars = 2 + ashex.endswith('L') | |
1331 | newnibbles = len(ashex) - njunkchars | |
1332 | if newnibbles < nibbles: | |
1333 | ashex = "0x" + "0" * (nibbles - newnibbles) + ashex[2:] | |
1334 | if int(ashex[2], 16) < 8: | |
1335 | # "looks positive", so need a byte of sign bits | |
1336 | ashex = "0xff" + ashex[2:] | |
1337 | ||
1338 | if ashex.endswith('L'): | |
1339 | ashex = ashex[2:-1] | |
1340 | else: | |
1341 | ashex = ashex[2:] | |
1342 | assert len(ashex) & 1 == 0, (x, ashex) | |
1343 | binary = _binascii.unhexlify(ashex) | |
1344 | return binary[::-1] | |
1345 | ||
1346 | def decode_long(data): | |
1347 | r"""Decode a long from a two's complement little-endian binary string. | |
1348 | ||
1349 | >>> decode_long('') | |
1350 | 0L | |
1351 | >>> decode_long("\xff\x00") | |
1352 | 255L | |
1353 | >>> decode_long("\xff\x7f") | |
1354 | 32767L | |
1355 | >>> decode_long("\x00\xff") | |
1356 | -256L | |
1357 | >>> decode_long("\x00\x80") | |
1358 | -32768L | |
1359 | >>> decode_long("\x80") | |
1360 | -128L | |
1361 | >>> decode_long("\x7f") | |
1362 | 127L | |
1363 | """ | |
1364 | ||
1365 | nbytes = len(data) | |
1366 | if nbytes == 0: | |
1367 | return 0L | |
1368 | ashex = _binascii.hexlify(data[::-1]) | |
1369 | n = long(ashex, 16) # quadratic time before Python 2.3; linear now | |
1370 | if data[-1] >= '\x80': | |
1371 | n -= 1L << (nbytes * 8) | |
1372 | return n | |
1373 | ||
1374 | # Shorthands | |
1375 | ||
1376 | try: | |
1377 | from cStringIO import StringIO | |
1378 | except ImportError: | |
1379 | from StringIO import StringIO | |
1380 | ||
1381 | def dump(obj, file, protocol=None, bin=None): | |
1382 | Pickler(file, protocol, bin).dump(obj) | |
1383 | ||
1384 | def dumps(obj, protocol=None, bin=None): | |
1385 | file = StringIO() | |
1386 | Pickler(file, protocol, bin).dump(obj) | |
1387 | return file.getvalue() | |
1388 | ||
1389 | def load(file): | |
1390 | return Unpickler(file).load() | |
1391 | ||
1392 | def loads(str): | |
1393 | file = StringIO(str) | |
1394 | return Unpickler(file).load() | |
1395 | ||
1396 | # Doctest | |
1397 | ||
1398 | def _test(): | |
1399 | import doctest | |
1400 | return doctest.testmod() | |
1401 | ||
1402 | if __name__ == "__main__": | |
1403 | _test() |