HTML parsers benchmark

Simple HTML DOM parser benchmark.

Competitors

Erlang

Mochiweb html parser. https://github.com/mochi/mochiweb/blob/master/src/mochiweb_html.erl The only real-world HTML parser for Erlang. Written on pure Erlang.

CPython

lxml.etree.HTML http://lxml.de/ libxml2 binding. Cython
BeautifulSoup 3 http://www.crummy.com/software/BeautifulSoup/bs3/documentation.html Pure python HTML DOM parser v3.
BeautifulSoup 4 http://www.crummy.com/software/BeautifulSoup/bs4/doc/ HTML DOM parser with pluggable backends.
html5lib http://code.google.com/p/html5lib/ Pure python DOM parser oriented to HTML5.

PyPi

BeautifulSoup 3
BeautifulSoup 4
html5lib

Node.JS

cheerio https://npmjs.org/package/cheerio Pure JS HTML DOM parser with jQuery API
htmlparser https://npmjs.org/package/htmlparser Pure JS HTML DOM parser
jsdom https://npmjs.org/package/jsdom Pure JS HTML DOM parser with rich browser-like API

Ruby

nokogiri http://nokogiri.org/ HTML/XML parser based on libxml2

C

libxml2 http://xmlsoft.org/html/libxml-HTMLparser.html Written on C non-strict HTML parser.

Perl

Mojo::DOM http://search.cpan.org/~sri/Mojolicious-3.70/lib/Mojo/DOM.pm Written on Perl DOM parser, part of Mojolicious framework

Google Go

gokogiri https://github.com/moovweb/gokogiri
go-html-transform https://code.google.com/p/go-html-transform
exp/html http://tip.golang.org/pkg/exp/html

PHP

tidy + simplexml http://php.net/manual/ru/book.tidy.php http://php.net/manual/ru/book.simplexml.php Tidy - PHP extension for repearing broken HTML, simplexml - built-in libxml2 binding

Haskell

fast-tagsoup http://hackage.haskell.org/package/fast-tagsoup

Java

JSoup http://jsoup.org/

Dart

html5lib.dart https://github.com/dart-lang/html5lib Written on pure dart port of python's html5lib.py

Mono

Html Agility Pack http://htmlagilitypack.codeplex.com
SGMLReader https://github.com/MindTouch/SGMLReader

Preparation

Install OS dependencies python-virtualenv, erlang, pypy, C compiler and libxml2 dev packages

sudo apt-get install ...
    libxml2-dev libxslt1-dev build-essential  # common
    python-virtualenv python-lxml             # python
    erlang-base                               # erlang
    pypy                                      # python PyPy
    nodejs npm                                # NodeJS
    cabal-install libicu-dev                  # Haskell
    php5-cli php5-tidy                        # PHP
    golang                                    # Go
    ruby1.9.1 ruby1.9.1-dev rubygems1.9.1     # Ruby
    maven2 default-jdk                        # Java
    mono-runtime mono-dmcs                    # Mono

Then run (it will prepare virtual environments, fetch dependencies, compile sources etc)

./prepare.sh

In case of errors, I recommended to install also cython, python-dev and retry.

To prepare only some of the platforms, define PLATFORMS environment variable:

PLATFORMS="pypy python" ./prepare.sh

RUN

Just run

./run.sh <number of parser iterations>

./run.sh 5000

To run tests only for some of the platforms, define PLATFORMS envifonment variable:

PLATFORMS="pypy python" ./run.sh 5000

To run series of tests use snippets like

for C in $(echo "10 50 100 400 600 1000"); do ./run.sh $C | tee output_$C.txt; done

Results

To convert results to CSV file, use to_csv.py

./run.sh 5000 | ./to_csv.py

or smth like

./run.sh 5000 | tee output.txt
./to_csv.py < output.txt

or, for series

for C in $(echo "10 50 100 400 600 1000"); do ./to_csv.py < output_$C.txt > results-$C.csv; done

There is also R - script that can build some pretty graphs: stats/main.r.

How to add my %platformname% to benchmark set?

Create directory %platformname%

mkdir %platformname%

Create run.sh and prepare.sh scripts:

run.sh - called every time when benchmark starts. Must use print_header() and timeit() functions from lib.sh to format output for each test. It must accept 2 arguments: HTML file path and number of iterations and pass them unchanged to benchmark scripts.
prepare.sh - called only once, before runing any benchmarks. It can download dependencies, compile sources etc.

Create your benchmark scripts. Requirements:

Must accept 2 arguments: path to HTML file and number of iterations
Must read HTML file once, then perform "number of iterations" parse cycles
Must print parser-loop runtime in seconds, calculated like start = time(); do_n_iterations(N); print time() - start
On each iteration must build full DOM tree in memory

Add %platformname% to platforms.txt file.

How to add new HTML to benchmark?

Just create HTML file named page_<some_page_name>.html.

seriyps/html-parsers-benchmark