>>> A = [{'name':'john','age':45},
{'name':'andi','age':23},
{'name':'john','age':22},
{'name':'paul','age':35},
{'name':'john','age':21}]
>>> sorted(A, key = lambda user: (user['name'], user['age']))
[{'age': 23, 'name': 'andi'}, {'age': 21, 'name': 'john'}, {'age': 22, 'name': 'john'}, {'age': 45, 'name': 'john'}, {'age': 35, 'name': 'paul'}]
This sorts by a tuple of the two attributes, the following is equivalent and much faster/cleaner:
>>> from operator import itemgetter
>>> sorted(A, key=itemgetter('name', 'age'))
[{'age': 23, 'name': 'andi'}, {'age': 21, 'name': 'john'}, {'age': 22, 'name': 'john'}, {'age': 45, 'name': 'john'}, {'age': 35, 'name': 'paul'}]
From the comments: @Bakuriu
I bet there is not a big difference between the two, but
itemgetter
avoids a bit of overhead because it extracts the keys and make thetuple
during a single opcode(CALL_FUNCTION
), while calling thelambda
will have to call the function, load the various constants(which are other bytecodes) finally call the subscript (BINARY_SUBSCR
), build thetuple
and return it… that’s a lot more work for the interpreter.
To summarize: itemgetter
keeps the execution fully on the C
level, so it’s as fast as possible.