Title: Build-a-blog
Date: 2024-06-17T14:46:36-04:00
---
I want to share my thought process for how to go about building a static blog generator from scratch.
There will be nothing ground breaking here - in fact this software will not be good. So turn back now if you're expecting the new [Hugo][hugo].
Actually you should probably stop reading and just use [Hugo][Hugo].
In case you are still interested, the goal is to take 1 afternoon + caffeine + some DIY spirit → _something_ resembling a static site/blog generator.
And I hope by the end of this post you might be inspired to build your own generation scripts, maybe in a new language you always wanted to try.
Lets see how hard this will be.
Here are the requirements for this blog:
* Generate an index with recent list of posts.
* Generate each individual post written in markdown -> html
* Support some metadata in each post
* A post title should have a slug
* Generate RSS
That boils down to:
1. Read some files
2. Parse markdown, maybe parse a header with some key/values.
3. Template strings
So there is 1 "exotic" feature in parsing/rendering Markdown as HTML that will need some thought.
The rest is just file and string manipulation.
Lets get it on.
## Picking the tool for the job
Most scripting languages would be fine tools for this task. But how to handle Markdown?
I've had [Crystal][1] in the back of my mind for this task. It is a nice general purpose language that included Markdown in the stdlib! But unfortunately Markdown was removed in [0.31.0][2]. Other than that, I'm not sure any other languages include a well rounded Markdown implementation out of the box.
I'll likely end up building the site in docker with an alpine image down the road, so just a quick search in alpines repos to see what could be useful:
```shell
❯ docker run --rm -it alpine
/ # apk update
fetch https://dl-cdn.alpinelinux.org/alpine/v3.18/main/x86_64/APKINDEX.tar.gz
fetch https://dl-cdn.alpinelinux.org/alpine/v3.18/community/x86_64/APKINDEX.tar.gz
v3.18.6-263-g77db018514d [https://dl-cdn.alpinelinux.org/alpine/v3.18/main]
v3.18.6-263-g77db018514d [https://dl-cdn.alpinelinux.org/alpine/v3.18/community]
OK: 20079 distinct packages available
/ # apk search markdown
discount-2.2.7c-r1
discount-dev-2.2.7c-r1
discount-libs-2.2.7c-r1
kdepim-addons-23.04.3-r0
markdown-1.0.1-r3
markdown-doc-1.0.1-r3
py3-docstring-to-markdown-0.12-r1
py3-docstring-to-markdown-pyc-0.12-r1
py3-html2markdown-0.1.7-r3
py3-html2markdown-pyc-0.1.7-r3
py3-markdown-3.4.3-r1
py3-markdown-it-py-2.2.0-r1
py3-markdown-it-py-pyc-2.2.0-r1
py3-markdown-pyc-3.4.3-r1
```
[`py3-markdown` in alpine][3] is the popular [`python-markdown`][4]. It's mature and available as a package in my [home distro][5].
Incredible.
## Let's build
First, lets read 1 post file and render some html.
I'll store posts in `posts/` like `posts/build_a_blog.md`.
And we'll store the HTML output in the same directory: `posts/build_a_blog.html`.
```python
import re
import logging
import markdown
destpath_re = re.compile(r'\.md$')
logging.basicConfig(encoding='utf-8', level=logging.INFO)
def render_post(fpath):
destpath = destpath_re.sub('.html', fpath)
logging.info("opening %s for parsing, dest %s", fpath, destpath)
# from: https://python-markdown.github.io/reference/
with open(fpath, "r", encoding="utf-8") as input_file:
logging.info("reading %s", fpath)
text = input_file.read()
logging.info("parsing %s", fpath)
out = markdown.markdown(text)
with open(destpath, "w", encoding="utf-8", errors="xmlcharrefreplace") as output_file:
logging.info("writing to %s", destpath)
output_file.write(out)
if __name__ == '__main__':
render_post('posts/build_a_blog.md')
```
And if we run it.
```shell
❯ python3 ./main.py
INFO:root:opening posts/build_a_blog.md for parsing, dest posts/build_a_blog.html
INFO:root:reading posts/build_a_blog.md
INFO:root:parsing posts/build_a_blog.md
INFO:root:writing to posts/build_a_blog.html
❯ head posts/build_a_blog.html
Build-a-blog
I want to share my thought process for how one would go about building a static blog generator from scratch.
Generate an index with recent list of posts.
Generate each individual post written in markdown -> html
Support some metadata in each post
A post title should have a slug
Generate RSS
```
Looking pretty good.
Now lets do this for all `.md` files in `posts/`
```python
import glob
...
def render_posts():
files = glob.glob('posts/*.md')
logging.info('found post files %s', files)
for fname in files:
render_post(fname)
if __name__ == '__main__':
render_posts()
```
And add another simple test post
```shell
❯ echo '# A new post' > ./posts/a_new_post.md
❯ python3 ./main.py
INFO:root:found post files ['posts/a_new_post.md', 'posts/build_a_blog.md']
INFO:root:opening posts/a_new_post.md for parsing, dest posts/a_new_post.html
INFO:root:reading posts/a_new_post.md
INFO:root:parsing posts/a_new_post.md
INFO:root:writing to posts/a_new_post.html
INFO:root:opening posts/build_a_blog.md for parsing, dest posts/build_a_blog.html
INFO:root:reading posts/build_a_blog.md
INFO:root:parsing posts/build_a_blog.md
INFO:root:writing to posts/build_a_blog.html
❯ head ./posts/a_new_post.html
A new post
```
Basically at this point, it's a blog generator!
But I want a few more features:
* Want the posts listed in the index sorted by date.
* Want each post to be templated in some html wrapper.
## Post ordering and templating
`python-markdown` supports metadata embedded in posts:
I thought I'd need to build something here, but turns out it's exactly what I need to assign a few extra attributes to a post.
I'll adjust our "spec" for posts such that each post must include the following metadata at the top of the file:
```txt
Title: Build-a-blog
Date: 2024-06-17T14:46:36-04:00
---
```
And I'd like to insert the `Title` automatically as a `
` tag in each post so I don't have to write it again in the markdown.
So first, lets test the metadata and adjust the test blog post.
```shell
❯ head -n4 ./posts/build_a_blog.md
Title: Build-a-blog
Date: 2024-06-17T14:46:36-04:00
---
```
And pop open a python repl to see how this works.
```python
>>> md = markdown.Markdown(extensions = ['meta']); f = open('posts/build_a_blog.md', 'r'); txt = f.read(); out = md.convert(txt); md.Meta
{'title': ['Build-a-blog'], 'date': ['2024-06-17T14:46:36-04:00']}
```
Looks pretty nice!
So first I will adjust the rendering function to prepend a line:
```markdown
# {title}
```
This has to be done after the full document renders because the `meta` `python-markdown` extension extracts metadata and converts to html in one call.
```python
def render_post(fpath):
...
md = markdown.Markdown(extensions = ['meta'])
logging.info("parsing %s", fpath)
out = md.convert(text)
title = md.Meta.get('title')[0]
date = md.Meta.get('date')[0]
out = markdown.markdown('# ' + title) + out
```
Finally, lets return a structure that will make other parts of the program aware of the filename that was rendered and the metadata (title, date)
```python
def render_post(fpath):
...
out = markdown.markdown('# ' + title) + out
with open(destpath, "w", encoding="utf-8", errors="xmlcharrefreplace") as output_file:
logging.info("writing to %s", destpath)
output_file.write(out)
return {
'title': title,
'date': date,
'fpath': fpath,
'destpath': destpath,
}
```
Now we have what we need to generate a complete index.
### Index templating
Lets start by defining what our index template file will be.
I'll choose `index.html.tmpl` and after rendering we will write to `index.html`.
So lets make a function that will take a list of our post structure above and render it in a `
'
for post in posts:
disp_date = datetime.datetime.fromisoformat(post.get('date')).strftime('%Y-%m-%d')
out += post_tpl.format(href=post.get('destpath'),
title=post.get('title'),
date=post.get('date'),
disp_date=disp_date)
return out + '
'
def render_index(posts):
fname = 'index.html.tmpl'
outname = 'index.html'
with open(fname, 'r', encoding='utf-8') as inf:
tmpl = Template(inf.read())
posts_html = posts_html(posts)
html = tmpl.substitute(posts=posts_html)
with open(outname, 'w', encoding='utf-8') as outf:
outf.write(html)
```
Make sure that `index.html.tmpl` contains a template variable for `${posts}`
```shell
❯ grep -C2 '\${posts}' ./index.html.tmpl
Welcome. Something will go here eventually.
${posts}
```
And we now need to connect `render_posts()` which returns each post that was processed to `render_index()`
```python
def render_posts():
files = glob.glob('posts/*.md')
logging.info('found post files %s', files)
posts = []
for fname in files:
p = render_post(fname)
posts.append(p)
logging.info('rendered post: %s', p)
return posts
if __name__ == '__main__':
posts = render_posts()
logging.info('rendered posts: %s', posts)
render_index(posts)
```
And lets run it!
```shell
❯ python3 ./main.py
INFO:root:found post files ['posts/a_new_post.md', 'posts/build_a_blog.md']
INFO:root:opening posts/a_new_post.md for parsing, dest posts/a_new_post.html
INFO:root:reading posts/a_new_post.md
INFO:root:parsing posts/a_new_post.md
INFO:root:writing to posts/a_new_post.html
INFO:root:rendered post: {'title': 'A new post', 'date': '2024-06-17T15:09:26-04:00', 'fpath': 'posts/a_new_post.md', 'destpath': 'posts/a_new_post.html'}
INFO:root:opening posts/build_a_blog.md for parsing, dest posts/build_a_blog.html
INFO:root:reading posts/build_a_blog.md
INFO:root:parsing posts/build_a_blog.md
INFO:root:writing to posts/build_a_blog.html
INFO:root:rendered post: {'title': 'Build-a-blog', 'date': '2024-06-17T14:46:36-04:00', 'fpath': 'posts/build_a_blog.md', 'destpath': 'posts/build_a_blog.html'}
INFO:root:rendered posts: [{'title': 'A new post', 'date': '2024-06-17T15:09:26-04:00', 'fpath': 'posts/a_new_post.md', 'destpath': 'posts/a_new_post.html'}, {'title': 'Build-a-blog', 'date': '2024-06-17T14:46:36-04:00', 'fpath': 'posts/build_a_blog.md', 'destpath': 'posts/build_a_blog.html'}]
```
And check how the output looks:
```shell
❯ grep -C4 'blog-posts-list' ./index.html
Build-a-blog
```
Not bad!
### Post templating
I think I want my blog to just maintain the overall layout from the index page and just render the post body where the main post list is.
So lets make that template rendering a bit more general.
I'll redefine the content area template variable to replace as `${content}` too.
```python
def render_template(tpl_fname, out_fname, content_html):
with open(tpl_fname, 'r', encoding='utf-8') as inf:
tmpl = Template(inf.read())
html = tmpl.substitute(content=content_html)
with open(out_fname, 'w', encoding='utf-8') as outf:
outf.write(html)
def render_index(posts):
content_html = posts_list_html(posts)
render_template('index.html.tmpl', 'index.html', content_html)
outf.write(out)
```
And now adjust where posts are written out.
```python
def render_post(fpath):
...
out = markdown.markdown('# ' + title) + out
logging.info("writing to %s", destpath)
render_template('index.html.tmpl', destpath, html)
```
After running you should see the each `post/*.html` file where each post file uses the full index template and includes each generated post HTML.
### Post sorting
With everything wired up now just need to sort the posts lists by the date metadata.
Lets do a bit of python repl sort testing because I never remember `datetime` usage.
Lets generate a few nicely formatted ISO date strings for testing.
```shell
❯ date -d'2023-01-01' -Is
2023-01-01T00:00:00-05:00
❯ date -Is
2024-06-17T16:30:35-04:00
```
And make a test array
```python
>>> posts = [{'date': '2023-01-01T00:00:00-05:00'}, {'date': '2024-06-17T16:30:35-04:00'}]
```
With our current script, the older post would be listed first. So lets try a sort.
```
# Double checking datetime parsing
>>> import datetime
>>> newer = datetime.datetime.fromisoformat('2024-06-17T16:30:35-04:00')
datetime.datetime(2024, 6, 17, 16, 30, 35, tzinfo=datetime.timezone(datetime.timedelta(days=-1, seconds=72000)))
>>> older = datetime.datetime.fromisoformat('2024-06-17T16:30:35-04:00')
datetime.datetime(2024, 6, 17, 16, 30, 35, tzinfo=datetime.timezone(datetime.timedelta(days=-1, seconds=72000)))
# Checking python sorting methods work as expected
>>> newer.__gt__(older)
True
>>> newer.__lt__(older)
False
>>> older.__gt__(newer)
False
>>> older.__lt__(newer)
True
# Doing the sort
>>> sorted(posts, key=lambda x: datetime.datetime.fromisoformat(x['date']), reverse=True)
[{'date': '2024-06-17T16:30:35-04:00'}, {'date': '2023-01-01T00:00:00-05:00'}]
```
Now lets apply this to our posts.
```python
if __name__ == '__main__':
posts = render_posts()
logging.info('rendered posts: %s', posts)
sorted_posts = sorted(posts,
key=lambda p: datetime.datetime.fromisoformat(p['date']), reverse=True)
render_index(sorted_posts)
```
### `` Templating
The last bit of templating is to make each post `` different.
I'll try something like `<title>cfebs.com - ${title}`
So `index.html.tmpl`
```html
cfebs.com${more_title}
```
And where using the title template `more_title` will default to empty string.
```python
def render_index(posts):
content_html = posts_list_html(posts)
render_template('index.html.tmpl', 'index.html', {'content': content_html, 'more_title': ''})
```
But for a post:
```python
def render_post(fpath):
...
title = md.Meta.get('title')[0]
date = md.Meta.get('date')[0]
out = markdown.markdown('# ' + title) + out
logging.info("writing to %s", destpath)
render_template('index.html.tmpl', destpath, {'content': out, 'more_title': ' - ' + title})
```
At this point we have functioning blog post generation with templating.
## RSS
This should be pretty easy as RSS is just reformatting our blog index list into different XML.
The `render_template` function will be useful here with a few more tweaks. So I'll make another template file (based off a reference )
```shell
# Grab the reference
❯ curl -sL 'https://drewdevault.com/blog/index.xml' > index.xml.example
# After a bit of editing
❯ cat ./index.xml.tmpl
${site_title}
${site_link}
${description}en${last_build_date}
${items}
```
`render_template` now gets even more generic and passes a `dict` to `Template.substitute()`
```python
def render_template(tpl_fname, out_fname, subs):
with open(tpl_fname, 'r', encoding='utf-8') as inf:
tmpl = Template(inf.read())
out = tmpl.substitute(subs)
with open(out_fname, 'w', encoding='utf-8') as outf:
outf.write(out)
```
And make sure to adjust any usages of `render_template` that exist.
```python
def render_index(posts):
content_html = posts_list_html(posts)
render_template('index.html.tmpl', 'index.html', {'content': content_html})
def render_post(fname):
...
render_template('index.html.tmpl', destpath, {'content': out, 'more_title': ' - ' + title})
```
And now can hack away at RSS generation:
```
def render_rss_index(posts):
subs = {
'site_title': 'cfebs.com',
'site_link': 'https://cfebs.com',
'self_full_link': 'https://cfebs.com/index.xml',
'description': 'Recent content from cfebs.com',
'last_build_date': 'TODO',
'items': 'TODO',
}
render_template('index.xml.tmpl', 'index.xml', subs)
```
After this initial test and a `python3 ./main.py` run, should see xml filled out.
```
❯ cat ./index.xml
cfebs.com
https://cfebs.com
Recent content from cfebs.comenTODO
TODO
```
Now lets finish up by generating each item entry and collecting them to be replaced in the template.
And this adds another `md.convert()` call, so why not add a util to reuse a single Markdown instance.
```python
# add a module var and helper method to reuse Markdown instance
md = markdown.Markdown(extensions=['extra', 'meta', TocExtension(anchorlink=True)])
def convert(text):
md.reset()
return md.convert()
def render_post(fpath):
...
out = convert(text)
title = md.Meta.get('title')[0]
date = md.Meta.get('date')[0]
out = convert('# ' + title) + out
...
def rss_post_xml(post):
tpl = """
{title}
{link}
{pubdate}{link}{description}
"""
with open(post['fpath'], 'r') as inf:
text = inf.read()
converted = convert(text)
link = "https://cfebs.com/" + post['destpath']
pubdate = email.utils.format_datetime(datetime.datetime.fromisoformat(post['date']))
subs = dict(title=post['title'], link=link,
pubdate=pubdate,
description=converted)
for k,v in subs.items():
subs[k] = html.escape(v)
return tpl.format(**subs)
def render_rss_index(posts):
items = ''
for post in posts[:5]:
items += rss_post_xml(post)
subs = {
'site_title': 'cfebs.com',
'site_link': 'https://cfebs.com',
'self_full_link': 'https://cfebs.com/index.xml',
'description': 'Recent content from cfebs.com',
'last_build_date': email.utils.format_datetime(datetime.datetime.now()),
}
for k,v in subs.items():
subs[k] = html.escape(v)
subs['items'] = items
render_template('index.xml.tmpl', 'index.xml', subs)
```
* Need to use `html.escape` anywhere there could be HTML tags in output.
* `posts[:5]` should always take the most recent 5 posts to add to the RSS feed.
## Wrapping up
Reached the end of the afternoon, so this is where I'll leave it.
It's not great software.
* No tests, no docs
* No validation of input or function arguments
* Hard coding values like the domain
* Using adhoc dicts for generic structures
* Relies on system python version and packages.
* Does not offer anything a tool like [hugo][hugo] does not already offer.
But, it's ~150 lines of python with 1 external dependency.
If python or `python-markdown` drastically changes, it'll probably take <10 minutes to debug.
And - it was fun to write and write about.
View the complete source for generating this blog:
* [main.py](https://git.sr.ht/~cfebs/cfebs.srht.site/tree/main/item/main.py)
* [index.html.tmpl](https://git.sr.ht/~cfebs/cfebs.srht.site/tree/main/item/index.html.tmpl)
* [index.xml.tmpl](https://git.sr.ht/~cfebs/cfebs.srht.site/tree/main/item/index.xml.tmpl)
Or the full repo tree:
## EDIT
Few additional things that will be added will go here.
### 2024-06-19, adding draft state
It might be nice to work on a rough draft, generate it for previewing, track it in git, but skip including it in the posts index.
So I'll add a piece of post metadata called `Draft` and use `filter()` before the posts are sorted or applied to the `index.html` or RSS `index.xml`.
This is the result.
```diff
diff --git a/main.py b/main.py
index bfd9382..52ce57b 100644
--- a/main.py
+++ b/main.py
@@ -31,6 +31,9 @@ def render_post(fpath):
title = md.Meta.get('title')[0]
date = md.Meta.get('date')[0]
+ draft = False
+ if md.Meta.get('draft'):
+ draft = True
out = convert('# ' + title) + out
@@ -42,6 +45,7 @@ def render_post(fpath):
'date': date,
'fpath': fpath,
'destpath': destpath,
+ 'draft': draft,
}
def render_posts():
@@ -134,6 +138,7 @@ def render_rss_index(posts):
def main():
posts = render_posts()
logging.info('rendered posts: %s', posts)
+ posts = filter(lambda p: not p['draft'], posts)
sorted_posts = sorted(posts,
key=lambda p: datetime.datetime.fromisoformat(p['date']), reverse=True)
render_index(sorted_posts)
```
And testing it:
```
❯ cat ./posts/test_thing.md
Title: Test thing
Date: 2024-06-19T13:38:34-04:00
Draft: 1
❯ python main.py
...
```
html should get generated, but not in the index or xml
```
❯ grep 'Test thing' ./posts/test_thing.html
cfebs.com - Test thing