Coffee Space


Listen:

Site Creation V2

Preview Image

Preview Image

Why This Again?

So, here we are again. Designing a new system for automatic static page creation. Ah well, it had to happen some time.

The old system was a nice little technical feat, with a javascript file parsing the entire page from markdown to HTML in the client's browser - reliably and without crashing the browser. At the time the thinking was:

In reality, what happened is:

The point is, I started off with the best of intentions and ended up making my website less accessible. My page download time (request to render) was 0.7 seconds, which I thought was good... I had a feeling I could do better though.

So, What's Different?

The new process is in my opinion better. Let's start with some comparisons from what was previously spoken about:

How this is achieved is the files are now stored as pure markdown, with a custom script being run over the files to produce the HTML files using pandoc.

Below is the script used (for the purpose of completeness):

0001 #!/bin/bash
0002 
0003 # build()
0004 #
0005 # Build in the current directory.
0006 #
0007 # @param $1 The location of the style sheet.
0008 # @param $2 The location of the navigation HTML data.
0009 function build {
0010   for filename in *.md; do
0011     pandoc --section-divs -c $1 -B $2 -o ${filename%.*}.html $filename
0012     echo -n "."
0013   done
0014 }
0015 
0016 # clean()
0017 #
0018 # Clean the working directory.
0019 function clean {
0020   rm *.html
0021 }
0022 
0023 # help()
0024 #
0025 # Display the help for this script.
0026 function help {
0027   echo "bash run.sh <CMD>"
0028   echo ""
0029   echo "  ComManD"
0030   echo ""
0031   echo "    build    Build the different directories"
0032   echo "    clean    Clean the directory of build files"
0033   echo "    help     Display the help"
0034 }
0035 
0036 # main()
0037 #
0038 # Handle the main program logic.
0039 #
0040 # @param $@ Arguments from the command line.
0041 function main {
0042   work="$(pwd)/www"
0043   style="http://coffeespace.org.uk/style.css"
0044   header="$work/header.html"
0045   directories[0]="$work"
0046   directories[1]="$work/blogs"
0047   directories[2]="$work/projects"
0048   directories[3]="$work/software"
0049   # Pre-build the header if required
0050   if [[ "$1" == "build" ]]; then
0051     pandoc --section-divs -o "$work/header.html" "$work/header.md_nocompile"
0052   fi
0053   # Iterate over the directories
0054   for dir in "${directories[@]}"; do
0055     cd $dir
0056     case "$1" in
0057       build)
0058         build $style $header
0059         ;;
0060       clean)
0061         clean
0062         ;;
0063       help)
0064         help
0065         exit 1
0066         ;;
0067       *)
0068         echo "Error: Use 'help' for more information"
0069         exit 1
0070         ;;
0071     esac
0072     cd ${directories[0]}
0073   done
0074   echo -e "\n[DONE]"
0075 }
0076 
0077 main $@

Notice the comments - I have no expectation of being able to remember how this works next year ;)

Then it's simply a case of getting the changes pulled from git and build automatically:

0078 #!/bin/bash
0079 
0080 # Infinite loop
0081 while :
0082 do
0083   # Fetch the latest changes
0084   git fetch
0085   # Check whether pull required
0086   if ! git diff --quiet remotes/origin/HEAD; then
0087     # Pull the latest changes
0088     git pull
0089     # Remove the old HTML files
0090     bash run.sh clean
0091     # Rebuild the HTML files
0092     bash run.sh build
0093   else
0094     echo "No changes"
0095   fi
0096   # Sleep for 5 minutes and check again
0097   sleep 300
0098 done

Notice that the site is only rebuilt when there are changes pushed - this means the entire site is down for 5 seconds maximum for every push to the repository.

Conclusion

It's still very "hacky", I know - but I like it. This is run on my 3 euro ScaleWay instance and runs perfectly fine. I once purposely tried to DDoS my own server with no effect, so it should stand up reasonably well to attack.

It's redeeming features are that it's simple, easy to understand, easy to debug and easier to add hacks too. There may be better ways to do this, but from my perspective it does enough. I'm fairly happy with this system.

I'm interested in knowing your thoughts - please post your comments below!