My first personal project

Hi, today I wanted to share my thoughts about project that I made called histmag to kindle.


I really enjoy reading historical articles over the internet. Very good source of such writings in polish is histmag. One problem is that articles are divided into sub-pages which don’t work well while I save them to read later. Then when I’m onto reading I send my read it later articles to kindle. To bypass such inconvenience I started adding every subpage manually to my queue in pocket. As a step into programming path I decided to code my own crawler to crawl through the website and then send the neatly prepared mobi file to my kindle.

What I learned?

After I made my mind I have to choose between many crawling tools available in python like: scrapy, beautifulsoup or lxml.

I have chosen lxml because it was new to me and I really enjoy working with new things. So after reading documentation about lxml I found out that most of scraping is done via XPATH.

Syntax and how to use XPATH is for another post, but I end up with very long queries like:

for elem in parsed_page.xpath('{root}//child::p[not(contains(@class, "article-tags")) '
                                   'and not(contains(@class, "article-info"))] '
                                   '| {root}//a[contains(@href, "author")]'
                                   '| {root}//em '
                                   '| {root}//img '
                                   '| {root}//span'.format(root=self.xpath_root)):

As you can see I take elements that are not paragraphs with class article-tags and article-info. Moreover I pull out the author and em and img tags with contents.

What did I learn about lxml? I really enjoyed using XPATH which is easily provided by this library. I also was able to get python representation of html object retrieved by XPATH. What I don’t like is that sometimes installing lxml can cause memory problems as well as you need to install some packages first (system wide).

Another thing that I try was TDD (Test Driven Development). I have used in on code dojo before but no in real project. The first thing that I learned was to use TDD only for production code- I was trying to test code that supposes to be playground and I wasn’t able to do this. But after designing how it suppose to work, TDD actually helps. Thanks to that method I have to create such small functions or classes that they will be easily testable. Moreover, I have a lot html in test because it is a web crawler so I created a mocked web pages to be the same as real one. What is more test allows me to change my code and see if there is any place that my program is broken.

After development, I use my test in tox. For using tox I change a bit imports and some pieces of code to let the test be run in a proper way. Learning how to organize imports to let package be installable from was another new experience.

After parsing html I wanted to generate my own html document with all subpages into. At first, I used dominate library but I then I switched to pyxml. Mainly because I need something simple to generate html. Pyxml API is very simple:

doc = html.html(
        # python don't allow keyword args with hypens
        html.meta(content="text/html; charset=utf-8", **{"http-equiv": "Content-Type"}),
        _get_tile(html, pages)
        html.div(_generate_body(html, pages, tempdir), id='article'),

This piece of code actually will generate the whole html.

Then thanks to amazon I can send html to my kindle and it will be automatically converted to mobi. The one problem with this approach is that I cannot send the whole html with images so I have to generate mobi locally and send it to the ebook reader. I did this by calling kindlegen. It works really good but the license stays that I can’t redistribute it so enduser have to download it manually. Whole html file with pictures was in the same directory, then I called kindlegen on html file:

proc = subprocess.Popen(
        os.path.join(html_dir_path, 'histmag.html'),
    stdout=subprocess.PIPE, stderr=subprocess.STDOUT

To call external binay I used subprocess library.

Lastly, I send a simple email via mailgun. It could be done in better way with my own domain, because right now end user has to provide mailgun_api_key and email_server for sending messages to kindle.

What can be done differently?

I believe that whole crawling mechanism should be redesigned due to being synchronous. Moreover, next time I will try beautifulsoup for getting particular elements from html tree. I would also like to try generating epub file instead of mobi.