Goodbye, Ghost

• 4 min read

Once upon a time, this blog was on Blogger.1

When Blogger started to feel a bit long in the tooth, I moved the blog to a self-hosted copy of Ghost.

Alas, over time, I’ve grown annoyed with Ghost. Some of the things I haven’t liked include:

  • changes that have happened to Ghost over time, such as the move away from Markdown and SQLite, and changes to the default theme (that ultimately led me to fork the theme)
  • that fact that if I replace an image on a post, the old image (and any resized copies of it) stay on the server unless I manually delete them (which is a faff)
  • the fact that it still doesn’t support AVIF images in posts
  • some upgrade failures I’ve had that were difficult to recover from
  • various other minor niggles that I won’t bother to mention

I was finally pushed over the edge recently when I had scheduled several posts to be published, and I found, some time later, that none of them had actually been published (and it took some clicking around to make them publish…)

I’ve wanted to move the blog to a static site for some time. And so after the scheduled post snafu, I bit the bullet and put some effort into making that dream happen. And, around a week later, you’re looking at the result.

Here’s the old and new home pages compared:

The old index page, with a list of posts in a grid layout The old index page, with a simple vertical list of posts
While the grid is gone on my blog, Microsoft’s Dev Blogs sadly has switched to cards in a masonry layout

The home page is now a simple (and much more readable) list of posts.

The posts themselves looks very similar to how they previously did:

A post titled ‘__init_subclass__ – a simpler way to implement class registries in Python’ on the old blog A post titled ‘__init_subclass__ – a simpler way to implement class registries in Python’ on the new blog
That underline glitch in the old post only shows up in screenshots taken by Firefox

What else has improved

Here’s some of the things I like about the new set-up:

  • images are stored alongside posts. It’s clear where they are being used, and any unused images have been pruned during the move
  • I no longer need to run any server-side software for the blog. That includes not having to run a database server, and not having to worry about backups in the same way
  • I have full control over the site. That includes data structures (posts are no longer stored in that weird card format), styling, the types of content I can embed in a post and being able to make customisations for a particular post
  • the accessibility of the site has improved
  • automated testing. I can use Playwright, or any other similar tool, to test functionality, test for visual regressions and to perform accessibility checks
  • posts are now primarily back in Markdown (either on its own, or using the MDX format). I can use footnotes again. Syntax highlighting of code blocks is much improved

Some things have been lost:

  • it’s more difficult to have committed private drafts. This has only really been lost because the site is open source
  • scheduled posts requires a bit more effort (using GitHub Actions, for example)
  • the Git repo is a bit on the large side (largely due to images in my post about Japan), and so I may have to do something about that soon (like use Git LFS, or use JPEG XL)

How I migrated data from the old site

The blog posts from Blogger had previously been shoe-horned into Ghost (largely staying in HTML). Ghost itself has gone through various changes in how posts are stored, moving from Markdown to various iterations of its ‘cards’.

So the data was a bit of a mess (even if it mostly rendered okay), and migrating it into Astro was somewhat painful.

I wrote a script that merged a data export from Ghost with the output from Ghost’s Content API, and ran some Unified pipelines on the HTML or Markdown for each post (depending on what was available for the post). That pipeline wrote a new MDX or Markdown file for each post, compatible with Astro and with things like the post title and date stored in front matter.

The pipeline also extracted all URLs of images and other media used in posts, downloaded that media, and saved the downloaded files next to the post they related to. The pipeline then rewrote the image URLs in each post to use Astro’s native image handling (for MDX, this often meant importing the image file, and using the Astro Picture component).

There were various other bits of fiddling required to get the script to output working and correct posts in all cases.

There were still some manual fixes required after all that for things like Ghost image gallery cards, videos and audio.

The blog is on GitHub

As usual, the site is open source for anyone curious.

Footnotes

  1. In fact, that version of the blog still exists ↩

Post a comment
Cahoot: An awful 2FA login process