Life is Long, Use Python

Random Python benchmarks. Is Python fast yet 🐍?

Keyword Args of Partial Functions

def f(a1, a2, a3, a4, a5, a6, a7):
    return sum([a1, a2, a3, a4, a5, a6, a7])

# Raw
f(1, a2=2, a3=3, a4=4, a5=5, a6=6, a7=7)

# Partial
f_partial = functools.partial(f, a2=2, a3=3, a4=4, a5=5, a6=6, a7=7)
f_partial(1)

# Spread
kwargs = {"a2": 2, "a3": 3, "a4": 4, "a5": 5, "a6": 6, "a7": 7}
f(1, **kwargs)

In Python, keyword arguments **kwargs are potentially mutable within the function body.

def why_you_have_to_do_this(a, **kwargs):
    kwargs["a"] += a

Therefore, each call to the function has to create a new dict to store all keyword arguments. This copy also happens when using functools.partial. There is no performance gain compared to simply spreading a dict of arguments in the call, yet you have to pay for the overhead of merging arguments when using partial.

Python Dictionary Literals and Constructors

One difference between JavaScript and Python is that JavaScript Object literals do not require quoting their keys, but Python dicts do.

let d = { a: 1, b: 2, [some_var]: 3 }

d = { "a": 1, "b": 2, some_var: 3 }

This requires more typing when using dicts as structs. However, Python also supports defining dictionaries using the dict constructor if no keys are variables:

a = dict(a=1, b=2)

What about their speed? Benchmarking these two ways by creating 10^7^ dicts with the same 4 key-value pairs.

Sadly the constructor is slightly slower. If you look at the constructor definition at dict_new of CPython source, the keyword arguments of the constructor are gathered as a dict first, and then merged to an empty dict which is the one we are constructing. This creates a small overhead.