Is there a way to conveniently define a C-like structure in Python? I'm tired of writing stuff like:
class MyStruct():
def __init__(self, field1, field2, field3):
self.field1 = field1
self.field2 = field2
self.field3 = field3
Use a named tuple, which was added to the collections module in the standard library in Python 2.6. It's also possible to use Raymond Hettinger's named tuple recipe if you need to support Python 2.4.
It's nice for your basic example, but also covers a bunch of edge cases you might run into later as well. Your fragment above would be written as:
from collections import namedtuple
MyStruct = namedtuple("MyStruct", "field1 field2 field3")
The newly created type can be used like this:
m = MyStruct("foo", "bar", "baz")
You can also use named arguments:
m = MyStruct(field1="foo", field2="bar", field3="baz")
You can use a tuple for a lot of things where you would use a struct in C (something like x,y coordinates or RGB colors for example).
For everything else you can use dictionary, or a utility class like this one:
>>> class Bunch:
... def __init__(self, **kwds):
... self.__dict__.update(kwds)
...
>>> mystruct = Bunch(field1=value1, field2=value2)
I think the "definitive" discussion is here, in the published version of the Python Cookbook.
How about a dictionary?
Something like this:
myStruct = {'field1': 'some val', 'field2': 'some val'}
Then you can use this to manipulate values:
print myStruct['field1']
myStruct['field2'] = 'some other values'
And the values don't have to be strings. They can be pretty much any other object.
dF: that's pretty cool... I didn't know that I could access the fields in a class using dict.
Mark: the situations that I wish I had this are precisely when I want a tuple but nothing as "heavy" as a dictionary.
You can access the fields of a class using a dictionary because the fields of a class, its methods and all its properties are stored internally using dicts (at least in CPython).
...Which leads us to your second comment. Believing that Python dicts are "heavy" is an extremely non-pythonistic concept. And reading such comments kills my Python Zen. That's not good.
You see, when you declare a class you are actually creating a pretty complex wrapper around a dictionary - so, if anything, you are adding more overhead than by using a simple dictionary. An overhead which, by the way, is meaningless in any case. If you are working on performance critical applications, use C or something.
You can also pass the init parameters to the instance variables by position
# Abstract struct class
class Struct:
def __init__ (self, *argv, **argd):
if len(argd):
# Update by dictionary
self.__dict__.update (argd)
else:
# Update by position
attrs = filter (lambda x: x[0:2] != "__", dir(self))
for n in range(len(argv)):
setattr(self, attrs[n], argv[n])
# Specific class
class Point3dStruct (Struct):
x = 0
y = 0
z = 0
pt1 = Point3dStruct()
pt1.x = 10
print pt1.x
print "-"*10
pt2 = Point3dStruct(5, 6)
print pt2.x, pt2.y
print "-"*10
pt3 = Point3dStruct (x=1, y=2, z=3)
print pt3.x, pt3.y, pt3.z
print "-"*10
Here is a solution which uses a class (never instantiated) to hold data. I like that this way involves very little typing and does not require any additional packages etc.
class myStruct:
field1 = "one"
field2 = "2"
You can add more fields later, as needed:
myStruct.field3 = 3
To get the values, the fields are accessed as usual:
>>> myStruct.field1
'one'
Personally, I like this variant too. It extends @dF's answer.
class struct:
def __init__(self, *sequential, **named):
fields = dict(zip(sequential, [None]*len(sequential)), **named)
self.__dict__.update(fields)
def __repr__(self):
return str(self.__dict__)
It supports two modes of initialization (that can be blended):
# Struct with field1, field2, field3 that are initialized to None.
mystruct1 = struct("field1", "field2", "field3")
# Struct with field1, field2, field3 that are initialized according to arguments.
mystruct2 = struct(field1=1, field2=2, field3=3)
Also, it prints nicer:
print(mystruct2)
# Prints: {'field3': 3, 'field1': 1, 'field2': 2}
There is a python package exactly for this purpose. see cstruct2py
cstruct2py is a pure python library for generate python classes from C code and use them to pack and unpack data. The library can parse C headres (structs, unions, enums, and arrays declarations) and emulate them in python. The generated pythonic classes can parse and pack the data.
For example:
typedef struct {
int x;
int y;
} Point;
after generating pythonic class...
p = Point(x=0x1234, y=0x5678)
p.packed == "\x34\x12\x00\x00\x78\x56\x00\x00"
How to use
First we need to generate the pythonic structs:
import cstruct2py
parser = cstruct2py.c2py.Parser()
parser.parse_file('examples/example.h')
Now we can import all names from the C code:
parser.update_globals(globals())
We can also do that directly:
A = parser.parse_string('struct A { int x; int y;};')
Using types and defines from the C code
a = A()
a.x = 45
print a
buf = a.packed
b = A(buf)
print b
c = A('aaaa11112222', 2)
print c
print repr(c)
The output will be:
{'x':0x2d, 'y':0x0}
{'x':0x2d, 'y':0x0}
{'x':0x31316161, 'y':0x32323131}
A('aa111122', x=0x31316161, y=0x32323131)
Clone
For clone cstruct2py run:
git clone https://github.com/st0ky/cstruct2py.git --recursive
Here is a quick and dirty trick:
>>> ms = Warning()
>>> ms.foo = 123
>>> ms.bar = 'akafrit'
How does it works? It just re-use the builtin class Warning (derived from Exception) and use it as it was you own defined class.
The good points are that you do not need to import or define anything first, that "Warning" is a short name, and that it also makes clear you are doing something dirty which should not be used elsewhere than a small script of yours.
By the way, I tried to find something even simpler like ms = object() but could not (this last exemple is not working). If you have one, I am interested.
NamedTuple is comfortable. but there no one shares the performance and storage.
from typing import NamedTuple
import guppy # pip install guppy
import timeit
class User:
def __init__(self, name: str, uid: int):
self.name = name
self.uid = uid
class UserSlot:
__slots__ = ('name', 'uid')
def __init__(self, name: str, uid: int):
self.name = name
self.uid = uid
class UserTuple(NamedTuple):
# __slots__ = () # AttributeError: Cannot overwrite NamedTuple attribute __slots__
name: str
uid: int
def get_fn(obj, attr_name: str):
def get():
getattr(obj, attr_name)
return get
if 'memory test':
obj = [User('Carson', 1) for _ in range(1000000)] # Cumulative: 189138883
obj_slot = [UserSlot('Carson', 1) for _ in range(1000000)] # 77718299 <-- winner
obj_namedtuple = [UserTuple('Carson', 1) for _ in range(1000000)] # 85718297
print(guppy.hpy().heap()) # Run this function individually.
"""
Index Count % Size % Cumulative % Kind (class / dict of class)
0 1000000 24 112000000 34 112000000 34 dict of __main__.User
1 1000000 24 64000000 19 176000000 53 __main__.UserTuple
2 1000000 24 56000000 17 232000000 70 __main__.User
3 1000000 24 56000000 17 288000000 87 __main__.UserSlot
...
"""
if 'performance test':
obj = User('Carson', 1)
obj_slot = UserSlot('Carson', 1)
obj_tuple = UserTuple('Carson', 1)
time_normal = min(timeit.repeat(get_fn(obj, 'name'), repeat=20))
print(time_normal) # 0.12550550000000005
time_slot = min(timeit.repeat(get_fn(obj_slot, 'name'), repeat=20))
print(time_slot) # 0.1368690000000008
time_tuple = min(timeit.repeat(get_fn(obj_tuple, 'name'), repeat=20))
print(time_tuple) # 0.16006120000000124
print(time_tuple/time_slot) # 1.1694481584580898 # The slot is almost 17% faster than NamedTuple on Windows. (Python 3.7.7)
If your __dict__ is not using, please choose between __slots__ (higher performance and storage) and NamedTuple (clear for reading and use)
You can review this link(Usage of slots
) to get more __slots__ information.
https://stackoverflow.com/a/32448434/159695 does not work in Python3.
https://stackoverflow.com/a/35993/159695 works in Python3.
And I extends it to add default values.
class myStruct:
def __init__(self, **kwds):
self.x=0
self.__dict__.update(kwds) # Must be last to accept assigned member variable.
def __repr__(self):
args = ['%s=%s' % (k, repr(v)) for (k,v) in vars(self).items()]
return '%s(%s)' % ( self.__class__.__qualname__, ', '.join(args) )
a=myStruct()
b=myStruct(x=3,y='test')
c=myStruct(x='str')
>>> a
myStruct(x=0)
>>> b
myStruct(x=3, y='test')
>>> c
myStruct(x='str')
If you don't have a 3.7 for @dataclass and need mutability, the following code might work for you. It's quite self-documenting and IDE-friendly (auto-complete), prevents writing things twice, is easily extendable and it is very simple to test that all instance variables are completely initialized:
class Params():
def __init__(self):
self.var1 : int = None
self.var2 : str = None
def are_all_defined(self):
for key, value in self.__dict__.items():
assert (value is not None), "instance variable {} is still None".format(key)
return True
params = Params()
params.var1 = 2
params.var2 = 'hello'
assert(params.are_all_defined)
The best way I found to do this was to use a custom dictionary class as explained in this post: https://stackoverflow.com/a/14620633/8484485
If iPython autocompletion support is needed, simply define the dir() function like this:
class AttrDict(dict):
def __init__(self, *args, **kwargs):
super(AttrDict, self).__init__(*args, **kwargs)
self.__dict__ = self
def __dir__(self):
return self.keys()
You then define your pseudo struct like so: (this one is nested)
my_struct=AttrDict ({
'com1':AttrDict ({
'inst':[0x05],
'numbytes':2,
'canpayload':False,
'payload':None
})
})
You can then access the values inside my_struct like this:
print(my_struct.com1.inst)
=>[5]