llllllllll

Initial Observations

The provided benchmark script is using Python's timeit module for benchmarking. I noticed that I was getting very different results between runs even with the same compiler, so I first switched to pyperf to attempt to get more stable results.

I am not sure if there is a Python API for pyperf, so I started by writing a small bash wrapper for benchmark:

llllllllll

Comments on optimizations around string concatenation.

Note: The code links are to CPython 3.8.5, the most recent release when this was written.

I was recently asked about a performance optimization in CPython around using += and + for string objects. As some people may already know, if you use += or + a string, it can sometimes be just as fast as ''.join. The question was to explain when that optimization couldn't be performed.

We will be going through the following example scenarios:

llllllllll

#include <cstdlib>
#include <compare>
#include <ranges>

namespace jj {
class iota : public std::ranges::view_base {
public:
    iota(long start, long end) : m_start(start), m_end(end) {}
    iota(long end) : iota(0, end) {}
    iota() : iota(0, 0) {}

llllllllll

from collections import UserString


class ResolvedDocString(UserString):
    __class__ = str


class DeferDocString(UserString):
    def __init__(self, f):
        self._f = f

llllllllll

class WithLazyDoc:                                                                   
    def __init__(self, get_doc, f): 
        self._get_doc = get_doc 
        self._f = f 
      
    @property 
    def __wrapped__(self): 
        return self._f 
 
    @property 

llllllllll

""" Example:

In [5]: class Class:
   ...:     def __init__(self, a, b):
   ...:         self.a = a
   ...:         self.b = b
   ...: 
   ...:     @self_scope
   ...:     def f(self):
   ...:         return a + b

llllllllll

In [1]: class Descr:
   ...:     def __init__(self):
   ...:         self.get_counter = 0
   ...: 
   ...:     def __get__(self, instance, owner):
   ...:         if instance is None:
   ...:             return self
   ...:         self.get_counter += 1
   ...:         return 1
   ...:     

llllllllll

"""
$ qipython -i ~/projects/tmp/lazy_talk_code.py 
Python 3.4.6 (default, Feb 16 2017, 15:43:12) 
Type "copyright", "credits" or "license" for more information.

IPython 5.3.0 -- An enhanced Interactive Python.
?         -> Introduction and overview of IPython's features.
%quickref -> Quick reference.
help      -> Python's own help system.
object?   -> Details about 'object', use 'object??' for extra details.

llllllllll

import pandas as pd
import networkx
import toolz


df = pd.read_csv('/home/joe/downloads/debug.csv', escapechar='\\')
df.columns = ['stack', 'module', 'function', 'lineno', 'id', 'type', 'repr']

g = networkx.DiGraph()

llllllllll

"""Example usage:

In [1]: !cat ayy.s
        .text
        .global _start
        .type	_start, @function
_start:
        movq    $10, %rdx

        pushw   $10