- Lib/pickle.py
- Modules/_pickle.c
- Modules/clinic/_pickle.c.h
We use pickle module to serialize and unserialize our python object, There're several protocol version for pickle, The current version is 4
The pickle module will use the faster _pickle implemented in C(Modules/_pickle.c) if possible, if the _pickle module not found, the python implemented pickle(Lib/pickle.py) will be used
| Type | Implementation |
|---|---|
| None | save_none |
| bool | save_bool |
| int | save_long |
| float | save_float |
| bytes | save_bytes |
| str | save_str |
| tuple | save_tuple |
| list | save_list |
| dict | save_dict |
| set | save_set |
| frozenset | save_frozenset |
| FunctionType | save_global |
| save_reduce |
Whenever you call dump, some extra information will be added to the result
The first byte is an identifier indicate that the following binary content is encoded in "pickle protocol"
The second byte is the protocol version
The final byte is a stop symbol indicate that it's the end of the binary content
NONE = b'N' # push None
def save_none(self, obj):
self.write(NONE)The data is N here, with the aforementioned information added to it
>>> import pickle
>>> pickle.dumps(None)
b'\x80\x04N.'bool is simiiar to None
NEWTRUE = b'\x88' # push True
NEWFALSE = b'\x89' # push False
def save_bool(self, obj):
if self.proto >= 2:
self.write(NEWTRUE if obj else NEWFALSE)The data here is b'\x88'(True) and b'\x89'(False)
>>> import pickle
>>> pickle.dumps(True)
b'\x80\x04\x88.'
>>> pickle.dumps(False)
b'\x80\x04\x89.'The integer will be saved in various format according to it's value
The float is saved in IEEE_754 standard
bytes object is save directly as the data part below
The head part various according to the data size
str is similar to bytes, except that str is encoded in utf-8 format before dump
tuple is more complicated than other basic type
If the tuple is empty
Let's see an example
dumps(("a", "b", (2, )))
b'\x80\x04\x95\x0f\x00\x00\x00\x00\x00\x00\x00\x8c\x01a\x94\x8c\x01b\x94K\x02\x85\x94\x87\x94.'\x80\x04 is pickle protocol and pickle version
\x95\x0f\x00\x00\x00\x00\x00\x00\x00 is frame symbol(\x95) and frame size(8 bytes) in little endian
. in last byte is the STOP symbol
Besides are the data
I find that dumps does not support self reference tuples(how to Build Self-Referencing Tuples)
Let's see an exmple again
dumps(["a", "b", (2, )])
b'\x80\x04\x95\x11\x00\x00\x00\x00\x00\x00\x00]\x94(\x8c\x01a\x94\x8c\x01b\x94K\x02\x85\x94e.'The first several bytes are pickle protocol, pickle version and frame size
The last byte is STOP symbol
The data can be described as (]\x94(\x8c\x01a\x94\x8c\x01b\x94K\x02\x85\x94e)
list will be dumped batch by batch(default batch size 1000)
dict and set are similar to list and tuple, begin and end with type symbol inidicate the type, and iter through each object and recursive call dump for each object
If what's to be saved is a type
class A(object):
a = "a"
b = "b"
def run(self):
print(self.a, self.b)
pickle.dumps(A)
b'\x80\x04\x95\x12\x00\x00\x00\x00\x00\x00\x00\x8c\x08__main__\x94\x8c\x01A\x94\x93\x94.'The data part is \x8c\x08__main__\x94\x8c\x01A\x94\x93\x94
dumps(A) saves the module_name (__main__) and the object name(A) in str formart
If what's to be saved is an instance
a = A()
pickle.dumps(a)
b'\x80\x04\x95\x15\x00\x00\x00\x00\x00\x00\x00\x8c\x08__main__\x94\x8c\x01A\x94\x93\x94)\x81\x94.'The data part is \x8c\x08__main__\x94\x8c\x01A\x94\x93\x94)\x81\x94
The only difference is that there're some extra information appended after the previous dumped result
A TUPLE indicate the args needed for instance call, in the current case a = A() the args is empty, so it's an EMPTY_TUPLE
A NEWOBJ symbol indicate that it needs to call cls.__new__(cls, *args) after load the dumped result
