How I built this website (Part 3)

Published on 2018-07-13

Tags: flask markdown web development


This series is a step-by-step tutorial for making a website like the one you're currently browsing. It includes an introduction to the Flask web framework, an introduction to Markdown (the markup language used to write this post), and also an introduction to hosting and deploying websites using Amazon Web Services.

Posts in this series:

  1. Introduction to Flask
  2. Introduction to Markdown
  3. Configuring a Markdown blog (you are here)
  4. Deploying to Amazon Web Services

In this post we will build a more sophisticated Flask project that has a Markdown blog component. This post builds on Part 1 of the series. Before starting, clone my GitHub repository for this project in order to follow along (this is different than the repository we used in Part 1). As with last time, remember to install packages for both the Python virtual environment and the Yarn web environment prior to starting.

Updated website structure

In Part 1 we generated a minimal Flask website that had a simple landing page with two views -- one for the main page and one page with a customizable route. In this part, we'll add a few views and templates for the blogging component and also the Markdown files for blog entries. We will also expand the static/ directory in order to handle custom CSS files, images and downloads.

The structure of this improved Flask project is as follows:

├── venv/
├── requirements.txt
├── website/
    └── static/
        └── package.json
        └── node_modules/
        └── extra_modules/
        └── css/
        └── downloads/
        └── img/
    └── templates/
        └── blog_home.html
        └── blog_post.html
        └── home_page.html
        └── layout.html
    └── content/
        └── Markdown files for blog entries

Markdown blogging

In designing this website, I wanted the blogging component to be as lightweight as possible. I want to be able to write a post in a text editor like Sublime Text and then publish it in a limited number of keystrokes. This means sacrificing fancy GUIs for writing and editing posts in some admin panel of the website. I also want my website to be "static". This means not having to rely on a database for persistence of content -- all content is derived directly from the file system. Static blogs are more secure (there are no concerns about database vulnerabilities) and more simple to configure.

The downside is that I am sacrificing add-ons like commenting systems, user accounts, etc. but I can live with that1. The only add-on I'm indulging in this blog is a system for "tagging" entries with various keywords that can serve as links for returning all posts with the same tag.

Structure of Markdown blog posts

Here's an example of what the raw Markdown of a blog post will look like:

title: An example post
date: 7/11/18
tags: markdown, flask
summary: In this post we show what a post looks like.

Blog post content begings here. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor incididunt ut labore et dolore magna aliqua...

Before we translate Markdown to HTML, we are going to parse these posts in order to pull out key content pieces, namely the title, date, summary and tags. Without doing this, everything would get rendered as normal text in <p> tags2. We will handle the additional formatting of these special elements in the blog post template using Jinja.

Serving blog posts

For the sake of simplicity, the URL for each blog post will be identical to the title, which is in turn identical to the file name. So if the title of a blog post is An example post, the URL for the post will be and the Markdown file will be an-example-post.md3. All Markdown files will live in the content/ directory.

By structuring our files and routes this way, our view for blog posts looks rather simple:

def blog_post(post_title):
    md_path  = os.path.join(app.root_path, 'content', '' % post_title)
    post = parse_markdown_post(md_path)
    return render_template('blog_post.html', post=post)

The post title is gathered from the route and then converted into the absolute path of the Markdown file on the file system by leveraging the app.root_path feature that Flask provides. We then read the file, parse out the metadata and convert the rest to HTML.

Pre-processing of Markdown file

Before doing the conversion to HTML, we want to parse out the metadata in the file's header. We will do this using a regular expression. More specifically, we will lean heavily on match groups to pull out specific parts of the raw HTML and store them as variables for later use.

re_pat = re.compile(r'title: (?P<title>[^\n]*)\sdate: (?P<date>\d{4}-\d{2}-\d{2})\s'
                    r'tags: (?P<tags>[^\n]*)\ssummary: (?P<summary>[^\n]*)')

This regular expression is pretty beefy. Here are the important pieces:

  • (?P<title>[^\n]*) - Pulls out all of the text starting after the substring title: and until the next newline character. This matching string (the title) is stored with the key title.
  • (?P<date>\d{4}-\d{2}-\d{2}) - Pulls out the YYYY-MM-DD date and stores this with the key date.
  • (?P<tags>[^\n^<]*) - Pulls out all of the text starting after the substring tags: and until the next newline character. This matching string (the comma-separated list of tags) is stored with the key tags.
  • (?P<summary>[^\n^<]*) - Pulls out all of the text after the substring summary: and until the next newline character. This matching string is stored with the key summary.

After using the re.match() function, we can then access each individual match group with a command like'title'). I also call re.split() in order to split the overall Markdown string using the regex pattern seen above. The last element of the list resulting from the split is all of the true Markdown content after the bits of metadata I've parsed out. All of these bits of metadata and the post content are stuffed into a custom Post object that I've defined in order to make organization of the metadata easier.

The Post object looks like this:

class Post:
    def __init__(self, title, date, tags, summary, href, content_md):
        self.title = title = date
        self.tags = tags
        self.summary = summary
        self.href = href
        self.content_md = content_md
        self.content_html = md_to_html(content_md)

Importantly, the HTML content of the post is generated automatically upon instantiation of a Post object. I'll cover the md_to_html() function below.

Converting Markdown to HTML

To accomplish Markdown-to-HTML conversion I am using the Python package mistune. The function in that leverages the markdown2 module is short:

def md_to_html(md_string):
    Convert a Markdown string to HTML.
    markdown_formatter = mistune.Markdown(
    html = markdown_formatter(md_string)
    return html

Here, we are enhancing the mistune module's native Markdown rendering by adding syntax highlighting in fenced code blocks. We do this by leveraging the Pygments module. The HighlightRenderer class is structured as follows:

from pygments import highlight
from pygments.formatters import html
from pygments.lexers import get_lexer_by_name

class HighlightRenderer(mistune.Renderer):
    Extend renderer built into mistune module. This object unables code
    highlighting during Markdown-to-HTML conversions.
    def block_code(self, code, lang):
        Get the language indicated in each fenced code block. Get the
        appropriate Pygments lexer based on this language and parse code 
        accordingly into HTML format. If not language is detected, use vanilla
        <code> blocks.
        if not lang:
            return '\n<pre><code>%s</code></pre>\n' % \
        lexer = get_lexer_by_name(lang, stripall=True)
        formatter = html.HtmlFormatter()
        return highlight(code, lexer, formatter)

Blog home page

The landing page for the /blog view will have three purposes. First, it will show an archive of all previous blog posts, showing the title, tags and summary for each post4. Second, it will provide a list of all the tags used on published posts, each tag will be a link to see all posts with a given tag. Third, the blog landing page will double as the view for browsing posts with a specific tag. We'll cover this in the next section.

In order to show an archive of every post, the function associated with the /blog view will need to loop over every Markdown post in the /content directory and then parse each post's header information. Along the way, we'll want to keep running totals of the number of posts associated with each tag. This is accomplished by using a dictionary. We'll sort the tags in alphabetical order (built-in Python dictionaries are intrinsically unordered so we're using an OrderedDict from the collections library to allow for sorting) and sort the posts by the date they were published. Below you can find the blog_home() function for the /blog view.

def blog_home():
    tag_dict = dict()
    posts = []
    content_path = os.path.join(app.root_path, 'content')
    for file in os.listdir(content_path):
        if not file.endswith('.md'):
        full_path = os.path.join(content_path, file)
        post_obj = parse_markdown_post(full_path)
        for tag in post_obj.tags:
            if tag not in tag_dict.keys():
                tag_dict[tag] = 0
            tag_dict[tag] += 1
    sorted_tag_dict = collections.OrderedDict()
    for key in sorted(tag_dict.keys()):
        sorted_tag_dict[key] = tag_dict[key]
    sorted_posts = sorted(posts, 
        key=lambda x: datetime.datetime.strptime(, '%Y-%m-%d'), reverse=True)
    return render_template('blog_home.html', posts=sorted_posts,

When this information is passed to the front end, we handle the posts and tag_dict objects using Jinja. In the blog_home.html template we create two column divs. On the left hand side we loop over the posts list and create a jumbotron element for each post5.

{% for post in posts %}
    <div class="jumbotron jumbotron-posts">
        <h3 class="tight-top"><a href="{{ post.href }}">{{ post.title }}</a></h3>
        <h5>{{ }}</h5>
        <p>{{ post.summary }}</p>
        {% for tag in post.tags %}
            <a class="btn btn-default btn-xs tag-link" href="{{ }}">
                <span class="glyphicon glyphicon-tag"></span> {{ tag }}
        {% endfor %}
{% endfor %}

On the right hand side we iterate over the tag_dict tag and create links for each tag while showing how many posts have each tag.

{% for tag in tag_dict %}
        <a class="btn btn-default btn-xs tag-link" href="{{ tag }}">
            <span class="glyphicon glyphicon-tag"></span> {{ tag }}
        {{ tag_dict[tag] }}
{% endfor %}

The final product looks like this:


In the absence of a search feature, I wanted some low-overhead means of organizing posts into broad categories. To accomplish this, I'm using tags. As we saw above, the tags for a post exist in the post's header section. These tags then appear as links at the top of a blog post page. These tag links also appear on the blog home page that we just covered. When someone clicks one of these links they should see all of the posts that have that tag. To accomplish this, I created a view that looks like this:

def get_tagged_posts(queried_tag):
    tag_dict = dict()
    matching_posts = []
    content_path = os.path.join(app.root_path, 'content')
    for file in os.listdir(content_path):
        if not file.endswith('.md'):
        full_path = os.path.join(content_path, file)
        post_obj = parse_markdown_post(full_path)
        if queried_tag in post_obj.tags:
        for tag in post_obj.tags:
            if tag not in tag_dict.keys():
                tag_dict[tag] = 0
            tag_dict[tag] += 1
    sorted_tag_dict = collections.OrderedDict()
    for key in sorted(tag_dict.keys()):
        sorted_tag_dict[key] = tag_dict[key]
    sorted_posts = sorted(matching_posts,
        key=lambda x: datetime.datetime.strptime(, '%Y-%m-%d'), reverse=True)
    return render_template('blog_home.html', posts=sorted_posts,
        tag_dict=sorted_tag_dict, queried_tag=queried_tag)

Note that the queried_tag argument for the function get_tagged_posts() comes from the URL. Also note that this function is very similar to the blog_home() function we just covered. In fact, we are going to use the same blog home page template. The key difference is that we are only passing posts back to the front end that include the tag we are querying for. In fact, I could have only used one function for both routes and merely decorated the function twice like this:

def get_tagged_posts(queried_tag=None):
    return render_template('blog_home.html', posts=sorted_posts,
        tag_dict=sorted_tag_dict, queried_tag=queried_tag)

In this case, when the /blog route is accessed, the queried_tag argument defaults to None and our function implementation would be such that all posts would ultimately end up in sorted_posts.

For either implementation, though, the template we wish to render is the blog_home.html template. Because this page will handle the cases of "all posts" and "only specifically tagged posts", we need to incorporate some logic on the front end using Jinja:

{% if queried_tag %}
    <h2 class="text-center">Posts with tag: "{{ queried_tag }}"</h2>
{% else %}
    <h2 class="text-center">All posts</h2>
{% endif %}

In the case of showing "all posts", the blog_home() function does not pass the variable queried_tag to the front end and so it defaults to None. For either case, the format Post objects being used are identical, it is only the length of the list of Post objects that changes so the rest of the template can be used for both cases.


In this post, we built out the rest of our Markdown blog. This involved creating a few new views and templates that allowed us to see individual blog posts as well as post archives. Because we are not using a database, we devised a way to include post metadata in the Markdown files without actually having to convert this information directly to HTML. We also built a system for categorizing posts using tags.

Now let's proceed to Part 4 where we'll deploy this finalized Flask project to Amazon Web Services.


  1. For an depth look at a Flask blogging set up with user accounts, comments, fancy GUIs, etc. check out Miguel Grinberg's excellent Flask Mega-Tutorial.

  2. This is a self-inflicted challenge due to my desire not to involve a database. Because there is no convenient means to store post metadata else where, it must live in the Markdown file for that post. Including a header in the post that is not directly translated to Markdown is perhaps a little kludgy, but I'm the only one who will be writing posts and I am comfortable keeping to this artificial standardized format. Perhaps most importantly, the end-user will never know the difference.

  3. Again, this is a convention that I am holding myself to.

  4. I do not include the summary anywhere in the post. I plan to use the summary only on the landing page for the blog.

  5. We are handling the same Post objects that used when creating the page for an individual post.