I've recently decided to start writing again.

And like most engineers I've immediately got down to... updating my website to
contain a blog, and automating things like posting to other sites like dev.to.

So to at least get some writing done and share a bit of my workflow, here's
how I've automated posting to dev.to using Github Actions.

The Idea

My personal website is setup as a GitHub Page which allows
me to focus on content by turning text checked in to a GitHub repo
into a static website using jekyll.

Given that my blogposts are already written in markdown and dev.to offers an API that accepts
markdown, publishing new posts to my personal website as well as dev.to seems easy.

Simply take any new blog posts pushed to the GitHub repo, and send them to the dev.to API in addition to rendering my
website.

image created with Excalidraw and the Dev Ops Icons library by Mark Sharpley

And because it seems so easy and I like things simple I've decided to make this work on my own using just Unix tools, git and curl instead of using an existing GitHub Action.

Finding new Blog Posts

For the initial version of this I'm only interested in finding newly added blog posts on the main
branch of my GitHub repo.

Let's say the latest commit on the main branch was aaaaaaa and I've just added two new posts and modified an old one in bbbbbbb.

Just a git diff

A simple git diff aaaaaaa bbbbbbb will show us the changes the between these two commits.

But that's way to much, because all we care about are the type of modification and filename.

Which diff can luckily give us with a flag:

git diff --name-status aaaaaaa bbbbbbb
M  _posts/2019-06-02-modified.md
A  _posts/2021-09-30-added.md
A  _posts/2021-10-17-added.md

Great! One Modified file and two Added files in _posts/.

But we still need to get this down to just the paths to new posts.

sed to the Rescue

To filter this down to just the filepaths of added posts, we'll use one of the more daunting Unix tools, the stream editor sed.

We'll use it with a simple regex to transform the output of git diff into a list of newly added blog posts:

sed -n -e 's/A\s*\(_posts\/.*.md\)/\1/p'

Simple right? Let's quickly run through that line:

-n tells sed to be 'silent'.

sed operates per line, reading a line, performing operations and then printing/returning the edited line - unless told to be silent, which makes it print nothing unless told otherwise.

The p at the end of our commands tells it to print when the pattern was matched.

-e simply tells sed explicitly that the next part is a command. We could omit this.

The substitution command that follows is the interesting part. It follows this format:

s/[pattern to replace]/[what to replace it with]/[options]

Let's dissect how ours turns the input into just outputting the paths to any added files:

s/A\s*\(_posts\/.*.md\)/\1/p`

> s/

Substitution command

> A\s*\(_posts\/.*.md\)

A         - lines starting with 'A'
\s*       - followed by zero or more whitespaces
_posts\/  - followed by '_posts/'
.*        - followed by zero or more characters
.md       - ending in .md
\(...\)   - captures what is between the braces to use again later

> \1

Simply references the content of the first capture group - the content of the braces above, which is our filepath.

> /p

Print the output of the substitution command.

Putting git and sed together with a pipe (|), we get what we need:

git diff --name-status aaaaaaa bbbbbbb | sed -n -e 's/A\s*\(_posts\/.*.md\)/\1/p'
_posts/2021-09-30-added.md
_posts/2021-10-17-added.md

Posting to dev.to

Now that we have the markdown files of new posts, it's time to send them to dev.to.

Forem API

dev.to uses forem which offers a straight forward articles API that accepts markdown with typical frontmatter directly.

So in theory we can POST the contents of our markdown files straight to dev.to.

Wiremock for testing API integrations

To test this - or any API - locally, wiremock is a great tool.

In standalone mode it can be run as a jar directly or using the docker container, and be configured using JSON to offer the endpoints you need and return what you want for testing.

To test the basics of integrating with the forem API we'll do the following:

Start the wiremock docker container in verbose mode, mapping it's default 8080 port to the same port on our system.

docker run -p 8080:8080 wiremock/wiremock:latest -v

And then define a simple version of the articles API by sending it to the wiremock instance.

For this we use curl to call wiremock's API with JSON data modelling the API endpoint we want to mock:

curl localhost:8080/__admin/mappings/new -d\
'{
    "request": {
        "method": "POST",
        "url": "/api/articles"
    },
    "response": {
        "status": 200
    }
}'

Now localhost:8080/api/articles will accept POST requests and always return a HTTP 200 response.

Not really realistic, but a good start for testing, and seeing what payload arrives at the API, thanks to wiremock printing all calls it receives in verbose mode.

POSTing Blog Posts

To try out the local mock API we can send it a test payload:

curl localhost:8080/api/articles -d '{"article": { "published": false, "body_markdown": "--- \n layout: post \n title: An automation test\n subtitle: a test subtitle\n---\n\n# Test content\n"} }'

This will POST an article object - which could be completely defined with just it's body_markdown as that already contains required information like the title, etc. in it's frontmatter.

But as I want to be able to give posts a final look, the payload also includes "published": false so posts will be created as a draft that can be published manually.

Putting it all together

git + sed + curl = profit

Time to put it all together, which I've done in a simple shell script:

#!/bin/sh

url=$URL
api_key=$API_KEY

echo "Detecting files added between commits $1 and $2..."

new_files=`git diff --name-status $1 $2 | sed -n -e 's/A\s*\(_posts\/.*.md\)/\1/p'`

for f in $new_files; do
    echo "Creating dev.to article for $f..."

    echo '{"article": { "published": false, "body_markdown": "' > api_payload
    sed -e ':a;N;$!ba;s/\n/\\n/g' $f >> api_payload
    echo '" }}' >> api_payload

    curl $url \
        -H 'Content-Type: application/json' \
        -H "api-key: $api_key" \
        -d @api_payload \
        -v

    rm api_payload
done

There's a few more steps than just git, sed and curl in there, so let's quickly run through them.

The scripts will get the URL and dev.to API_KEY from environment variables, so they can be passed in from a GitHub Action.

url=$URL
api_key=$API_KEY

Then it will find any new_files between two commits that are passed in as first and second input arguments ($1 and $2), and then post the content to the given URL for each of the new_files.

new_files=`git diff --name-status $1 $2 | sed -n -e 's/A\s*\(_posts\/.*.md\)/\1/p'`

for f in $new_files; do

The body of the HTTP POST is placed in a temporary api_payload file - but could also be built in place, or one long command piping into curl.

echo -n '{"article": { "published": false, "body_markdown": "' > api_payload
sed -e ':a;N;$!ba;s/\(["\\]\)/\\\1/g;s/\n/\\n/g;s/\t/  /g' $f >> api_payload
echo -n '" }}' >> api_payload

The only really interesting part of creating the payload is:

sed -e ':a;N;$!ba;s/\(["\\]\)/\\\1/g;s/\n/\\n/g;s/\t/  /g' $f >> api_payload`

sed is used here to escape or replace special characters in the JSON payload.

Double quotes and backslashes will be escaped with a backslash by the first substitution.

The second substitution will replace newline in the markdown file with an escaped newline \\n, so that the payload will actually contain \n characters for each line break. Without this the API call would swallow all newlines and the blog post on dev.to would miss all linebreaks.

The third substitution replaces tabs with two spaces rather than escaping them.

How exactly that sed expression replaces characters in the whole file is nicely described in this stakeoverflow post, so I will not repeat it here.

curl simply sends the json content of the temporary api_payload file to the given URL using the API_KEY.

curl $url \
    -H 'Content-Type: application/json' \
    -H "api-key: $api_key" \
    -d @api_payload \
    -v

GitHub Action

Now to have this script run whenever I add a new blog post to my website repo, we're still missing a GitHub action that calls it:

name: post-to-dev-to

on:
  push:
    branches: [ master ]
    paths:
      - "_posts/**"

jobs:
  post-to-dev:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
        with:
        fetch-depth: 0
      - name: post
        run: .github/workflows/dev-post.sh ${{ github.event.before}} ${{github.event.after}}
        env:
          URL: https://dev.to/api/articles
          API_KEY: ${{ secrets.DEV_TO_KEY }}

Because posting to dev.to should only happen for new posts I've pushed to the main branch of the repo, the Action is configured to only trigger if the _posts/ folder changes after a push to master.

on:
  push:
    branches: [ master ]
    paths:
      - "_posts/**"

Then the Action executes one job on an Ubuntu container, in which it first makes sure the repo is checked out, and then executes our script from above.

jobs:
  post-to-dev:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v2
      - name: post
        run: .github/workflows/dev-post.sh ${{ github.event.before}} ${{github.event.after}}
        env:
          URL: https://dev.to/api/articles
          API_KEY: ${{ secrets.DEV_TO_KEY }}

To execute the script is uses information from the github.event that triggered the Action to get the latest commit before and after the push happened.

What's next?

You're hopefully reading this on dev.to, proving this first part of blog automation has worked.

If this is actually interesting to anyone I'll continue the series with a write-up on how I use markdown and GitHub actions to also create and publish slide-decks.

Else I'll just continue my personal mission of only ever having to write markdown and wasting time on automation quietly and go back to writing about other things.