import functools import time import timeit from statistics import mean import requests from autoscraper import AutoScraper from bs4 import BeautifulSoup from lxml import etree, html from mechanicalsoup import StatefulBrowser from parsel import Selector from pyquery import PyQuery as pq from selectolax.parser import HTMLParser from scrapling import Selector as ScraplingSelector large_html = ( "" + '

' * 5000 + "

" * 5000 + "" ) def benchmark(func): @functools.wraps(func) def wrapper(*args, **kwargs): benchmark_name = func.__name__.replace("test_", "").replace("_", " ") print(f"-> {benchmark_name}", end=" ", flush=True) # Warm-up phase timeit.repeat( lambda: func(*args, **kwargs), number=2, repeat=2, globals=globals() ) # Measure time (1 run, repeat 100 times, take average) times = timeit.repeat( lambda: func(*args, **kwargs), number=1, repeat=100, globals=globals(), timer=time.process_time, ) min_time = round(mean(times) * 1000, 2) # Convert to milliseconds print(f"average execution time: {min_time} ms") return min_time return wrapper @benchmark def test_lxml(): return [ e.text for e in etree.fromstring( large_html, # Scrapling and Parsel use the same parser inside so this is just to make it fair parser=html.HTMLParser(recover=True, huge_tree=True), ).cssselect(".item") ] @benchmark def test_bs4_lxml(): return [e.text for e in BeautifulSoup(large_html, "lxml").select(".item")] @benchmark def test_bs4_html5lib(): return [e.text for e in BeautifulSoup(large_html, "html5lib").select(".item")] @benchmark def test_pyquery(): return [e.text() for e in pq(large_html)(".item").items()] @benchmark def test_scrapling(): # No need to do `.extract()` like parsel to extract text # Also, this is faster than `[t.text for t in Selector(large_html, adaptive=False).css('.item')]` # for obvious reasons, of course. return ScraplingSelector(large_html, adaptive=False).css(".item::text") @benchmark def test_parsel(): return Selector(content=large_html).css(".item::text").extract() @benchmark def test_mechanicalsoup(): browser = StatefulBrowser() browser.open_fake_page(large_html) return [e.text for e in browser.page.select(".item")] @benchmark def test_selectolax(): return [node.text() for node in HTMLParser(large_html).css(".item")] def display(results): # Sort and display results sorted_results = sorted(results.items(), key=lambda x: x[1]) # Sort by time scrapling_time = results["Scrapling"] print("\nRanked Results (fastest to slowest):") print(f" i. {'Library tested':<18} | {'avg. time (ms)':<15} | vs Scrapling") print("-" * 50) for i, (test_name, test_time) in enumerate(sorted_results, 1): compare = round(test_time / scrapling_time, 3) print(f" {i}. {test_name:<18} | {str(test_time):<15} | {compare}") @benchmark def test_scrapling_text(request_html): # Will loop over resulted elements to get text too to make comparison even more fair otherwise Scrapling will be even faster return [ element.text for element in ScraplingSelector(request_html, adaptive=False) .find_by_text("Tipping the Velvet", first_match=True) .find_similar(ignore_attributes=["title"]) ] @benchmark def test_autoscraper(request_html): # autoscraper by default returns elements text return AutoScraper().build(html=request_html, wanted_list=["Tipping the Velvet"]) if __name__ == "__main__": print( " Benchmark: Speed of parsing and retrieving the text content of 5000 nested elements \n" ) results1 = { "Raw Lxml": test_lxml(), "Parsel/Scrapy": test_parsel(), "Scrapling": test_scrapling(), "Selectolax": test_selectolax(), "PyQuery": test_pyquery(), "BS4 with Lxml": test_bs4_lxml(), "MechanicalSoup": test_mechanicalsoup(), "BS4 with html5lib": test_bs4_html5lib(), } display(results1) print("\n" + "=" * 25) req = requests.get("https://books.toscrape.com/index.html") print( " Benchmark: Speed of searching for an element by text content, and retrieving the text of similar elements\n" ) results2 = { "Scrapling": test_scrapling_text(req.text), "AutoScraper": test_autoscraper(req.text), } display(results2)