074e80289c
- re-added future support - fix for exists of category - new footer - footnote display fixes |
||
---|---|---|
.venv | ||
templates | ||
.gitignore | ||
__init__.py | ||
exiftool.py | ||
LICENSE | ||
nasg.py | ||
README.md | ||
requirements.txt | ||
run | ||
settings.py |
NASG - not another static generator...
Near full circle: from static to full dynamic to semi-static
Nearly 20 years ago I did my very first website with a thing called Microsoft FrontPage. I was static, but assembled from footer, nav, etc. html parts by FrontPage, then uploaded to a free webhost, and I was very happy with it: it was extremely simple to edit and to maintain.
Years passed and first I wrote a CMS that used text files as storage, based on a PHP library that stored serialized objects in text files - basically JSON, before JSON even existed. Then I moved to MySQL, then dropped the whole thing for WordPress, which I loved, up until Gutenberg was announced, at which point I realized how nastily I tinkered and altered my WordPress already to use it with Markdown, to make image and handling a tiny bit better, etc.
So I dropped it and make a static generator, only to realize, there are things I can't make static. At that point - 2017 -, these were:
- search
- proper redirect and gone entry handling (you can't set HTTP headers from a HTML file)
- receiving webmentions
Because I wanted to learn Python, the static generator is coded in Python, so I decided to run a Python web service with Sanic. It took me 3 iterations to realize, I'm doing it wrong, because the one and only thing that is available on nearly any webhost - think of plain old Apache - is PHP.
So the abomination I'm doing right now is to generate some near-static PHP files from the Python code which handles:
- search still with SQLite, but due to PHP versions, with FTS4 instead of FTS5; populated from Python, read by PHP
- gone (HTTP 410) and redirect (HTTP 301)
As for webmentions, as much as I try avoiding external dependencies, I came to realize a very simple fact: webmentions are external as well. So I started using webmention.io to receive them and query it on build time.
Now, about that the generator itself
The content is StriclYAML + MultiMarkdown. Except exiftool, there are no non-python dependencies any more, but exiftool
is the only thing that parses lends data for photos.
Python libraries used:
- arrow
- bleach
- emoji
- feedgen
- Jinja2
- langdetect
- Markdown
- markdown-urlize
- Pygments
- python-frontmatter
- requests
- unicode-slugify
- Wand
- ... and their dependencies
Most of the processing relies on the structuring of my data:
- whatever is not a directory in the root folder of the contents will be copied as is
- directories mean category
- 1 sub-directory per entry within the category, named as the post slug
- index.md as main file
- timestamp-sanitizedurl.md for webmentions and comments
- if there is a .jpg, named the same as the post directory name, the post is a photo
- all markdown image entries are replaced with
<figure>
with added visible exif data if they math the criterias that they are my photos, namely they match a regular expression in their exif Copyright or Artist field - this is produced by my camera - all images will be downsized and, if matched as photo, watermarked on build
/
├── about.html
├── category-1
│ ├── article-1
│ │ └── index.md
│ │ └── extra-file.mp4
│ │ └── 1509233601-domaincomentrytitle.md
│ ├── fancy-photo
│ │ └── index.md
│ │ └── fancy-photo.jpg
Mostly that's all.