MangaCMS

Comic/Manga/Light-Novel Download tool and reader.

Plugin scrapers for:

• Manga Sites:

  • Batoto.com
  • download.Japanzai
  • MangaCow
  • MangaJoy
  • Starkana.com
  • CXC Scans
  • Webtoons.com
  • WebtoonsReader.com
  • LoneManga.com
  • EasyGoing Scans
  • Dynasty Scans
  • KissManga
  • Manga.Madokami
  • Crunchyroll
  • MangaBox
  • Comic Zenon english site
  • Many FoolSlide based sites:
    • Vortex Scans
    • RedHawk Scans
    • Roselia Scans
    • Sense Scans
    • ShoujoSense Scans
    • TwistedHel Scans
    • Casanova Scans
    • MangaTopia
  • Numerous IRC bots.
    • ViScans: Bort
    • StupidCommotion: Suzume, Torako
    • iMangaScans: Icebox
    • EgScans: Easybot
    • Illuminati-Manga: Illum, Illuminati|Mad
    • A-Team: Azrael, Boink, Death, Hannibal
    • BentoScans: [Bento]
    • ForTheHalibut: `FTH`
  • IRC Channel Triggers:
    • Cat Scans: #CATscans
    • Renzokusei Scans: #renzokusei
    • Anything in any channel MOTD on Irchighway.

• Hentai Sites:

  • Fakku
  • Doujin Moe
  • Pururin
  • Exhen.Madokami
  • HBrowse
  • Exhentai/SadPanda

• Metadata:

  • MangaUpdates (metadata only).

Under consideration:

• http://tapastic.com/ (Lots of western stuff. Add a separate western table system, and start grabbing webcomics?)

This is a "firehose-style" scraper, in that it doesn't really provide any built in facilities for filtering the scraped content. It basically just scrapes everything, all the time. This is because my own interests are more along the lines of archiving everything, rather then just selective downloading (there are other tools for that).

I prefer to focus on scraping sites that offer archive downloads, but those are fairly rare, so image-hosting sites are also scraped. Scraped images are automatically packed into per-chapter archives. My only absolute criteria for inclusion at this time is that the site not resize or watermark files, since that prevents the deduplication system from working properly (note: I'm more relaxed about this for the H sites, primarily out of necessity).

Automatic ad-removal and banner-removal. Currently, Starkana inserts an irritating self-aggrandizing image at a random position in each archive downloaded from their site. This is currently automatically and silently removed from each zip-file on download.
Any download system for sites that add source banners should crop the banners automatically, or remove any intersitial cruft.

Light novels content is pre-processed with a python port of the readability tool, together with some carefully crafted content selectors to extract just the actual content, with the goal of providing a better reading experience. Baka-Tsuki, in particular, is fairly terrible from a UI perspective on their actual site.

---

Streaming archive decompression so there are no temporary files or unzipped archives for the reader.

Manga reading is through @balaclark's HTML5-Comic-Book-Reader, specifically my fork, which is similar to the original, except being heavily tweaked for usage on tablets. The reader is HTML5/javascript based, and features extremely aggressive pre-caching to provide the best reading experience possible. It actually downloads the entire manga/comic in the background as soon as it's opened, so page-changing is near-instantaneous.

Has lots of dependencies:

• Mako
• CherryPy
• Pyramid
• Psycopg2 (And a PostgreSQL install!)
• Beautifulsoup4
• FeedParser
• colorama
• pyinotify
• python-dateutil
• apscheduler
• rarfile
• python-magic
• babel
• cython
• irc
• python-sql
• readability
• Probably more

Installing:

Add Postgresql 9.5 repository if not present (see here).

Run the installDeps.sh script in the setuptools directory (note: must be run with sudo or as root. Please review it before running for security reasons).
The dependency installation cannot be done using only pip, unfortunately, as a number of the changes made during the setup process are outside of what pip can change. In particular, properly installing PIL requires installing a number of system packages (libtiff4-dev libjpeg-turbo8-dev zlib1g-dev liblcms2-dev libwebp-dev libxml2 libxslt1-dev). Numpy/scipy also requires some system packages, and the setup script also does some setup for postgres, such as installing it, and enabling the citext extension. It also increases the number of inotify watches.

Once you have installed the dependencies, you have to configure the various options. Copy settings.base.py to settings.py, and then edit it:

Note: All paths MUST be absolute. Some of the threading somewhere is eating the environment information. Rather then try to track down where in the complex dependency system the environment variables are being modified, it's easier to just use absolute paths. (Note: this means that paths like ~/blah will also not work. You have to specify /home/{username}/blah).

• pickedDir = Directories in this folder are preferentially used, and if a download goes in this directory, it is highlighted in the web interface.
• baseDir = This is where the bulk manga downloads go, if a folder is not found in the "picked" directory..
• djMoeDir = And DoujinMoe files
• puRinDir = Pururin Files
• ExhenMadokamiDir = ExHen.madokami Files
• fkDir = Fakku Files Files
• fufuDir = Where Fufufuu.net files go
• webCtntPath = Path to the ctnt directory in the repo.

If you want to pull down MangaUpdate metadata, you need to specify your username and password in the buSettings fields (mangaupdate = Baka-Updates). Eventually the starkana scraper will pull down information about your watched series as well, which is why there are fields for starkana username+pass as well. This is currently not implemented.

Finally, the actual scraper has two halves. mainWeb.py is the web-interface, and mainScrape.py is the scraper. Currently, if you run mainWeb.py having never run mainScrape.py, it will probably fail, because the scraper is responsible for setting up the database. Simply run mainScrape.py first. (All are run python3 mainScrape.py or python3 mainWeb.py).

The tools are currently not daemonized at all, and must be manually run after restarting. I normally just leave them running in a screen session on my server.

---

The database must be somewhat manually set up at this time.

postgres=# CREATE USER mangacmsuser;
CREATE ROLE
postgres=# \password mangacmsuser
Enter new password:   { Type your Password }
Enter it again:   { Type your Password }
postgres=# CREATE DATABASE mangacms;
CREATE DATABASE
postgres=# GRANT ALL PRIVILEGES ON DATABASE mangacms to mangacmsuser;
GRANT
postgres=# CREATE EXTENSION citext;
CREATE EXTENSION
postgres=# \q

Note: using capitals in usernames/tablenames/columnames in postgres is rather difficult, just don't do it.

You must then specify the Database server's IP, the username, databasename, and password in the settings.py file.

You also have to add the line

local mangacms all md5

to /etc/postgresql/9.5/main/pg_hba.conf, to allow local connections to the postgres database with password auth.

Once you have the above complete, the various tables should be bootstrapped upon the execution of mainScrape.py. This should be done /before/ any attempt to execute mainWeb.py, or the latter will fail due to the lack of the required tables.

You can manually bootstrap the required tables by running firstRun.py. This file should be idempotent, so running it repeatedly shouldn't have any harmful side-effects.

If you don't want all the plugins to run, you can disable them by commenting them out in activePlugins.py. The plugins that will run are in the dictionary scrapePlugins in that file. Just comment out any plugin you don't want running.

---

Preliminary deduplication support is currently present, (using my IntraArchiveDeduplicator) tool. This is intended to allow collation of files from many sources while having as few local duplicate files actually stored locally as possible.

Archives downloaded are automatically added to the deduplication database, and if a downloaded archive is found to contain no new files, it is automatically deleted, which prevents the fact that the scraper fetches files from multiple sources from resulting in duplicate downloads.

There is some support for fuzzy image matching using perceptual-hashing mechanisms that are already in-place in perceptual-hashing system.

---

Because I'm sure no one wants to just read about what MangaCMS does, here are some screenshots of the web-interface:

---

Tests for all the various plugins are in the /tests/ directory. Because of how they hook into the rest of the system, they must be run as modules: python3 -m tests.test-{name}. This is required to avoid having to dick about with the contents of sys.path.

---

Q&A

Q: Is this hard to use?
A: It requires at minimum a decent familiarity with the Linux command line. Python and/or SQL knowledge is also helpful.
I generally idle in #madokami on rizon, so you can ask me questions if you need there, though I'm not actually present behind my client a lot of the time. I'll help with whatever, though I can't exactly give a complete lesson on "how to linux" or so forth.

Q: You're scraping my site! Don't do that!
A: Your web-site. That you posted publically. You don't really understand how this whole "internet" thing works, do you?
TL;DR No.

---

This was written as much for programming excercise as for practical use, so there may be some NIH-y things going on. For example, I wrote my own database abstraction layer (see MonitorDbBase.py, RetreivalDbBase.py), primarily as an opportunity to teach myself SQL. Some of the interitence structures are for a similar purpose (I wanted to play with abstract classes).

The light-novel scraper uses SqlAlchemy, so it's not all NIH. Dumped SqlAlchemy. The documentation is too poor for it to be useable.

Currently looking at python-sql for dynamic SQL where I have more control over the generated SQL. Most of the database interfacing now uses python-sql for dynamic query generation, and all query variables are parameterized, so SQL injection should be fairly challenging.
There are a few unparameterized variables, primarily because there are some things that cannot be parameterized, such as table names. However, the only unparameterized variable (tablename) can only be altered by modifying local plugin python files, so I think it's at least an acceptable risk.
AFICT, the primary issue I had with SQLAlchemy is that my current page-generation mechanisms are very procedural in nature, which meshes extremely poorly with a database API that is basically only designed to s upport OOP-y ORM-style code structures.

Realistically, the ENTIRE web interface is something of an afterthought. This project was initially only a scraper, and the web interface was initially more of a "oh look, I can run a web server with twistd (back when it was a Py2K+Twistd project!)" thing. It kind of grew from there, but the database schema was always designed from the get-go to be scraper-centric.

This is basically:

• The first web application I'd ever written.
• The first >5KLOC project I've ever done.
• I am absolutely terrible at producing nice-looking web pages.

So take that into consideration when considering the codebase. I've learnt a /LOT/ about designing a system for maintainability and modularity, but I'm stuck with some original design-choices that are less then ideal.

As with about everything I do, the first run through teaches me all the things I /wish/ I had done differently.