blog tweaks

July 6, 2024

Tl;DR:

A few changes happened to this blog in the past few weeks:

RSS feed and web pages no longer embed svg images into <html> and include them via <img src="...">.

This fixes RSS readers like miniflux but might break others. At least now there should be an icon in place of a missing picture instead of just stripped tags.

As a small bonus RSS feed should not be as large to download.
RSS feed now includes source code snippets without syntax highlighting.

I never included css style into rss feed. highlighting-kate uses various tags and decorates them with links heavily. This change fixes source code rendering in liferea.
RSS feed now embeds https:// self-links instead of http:// (except for a few recent entries to avoid breaking reading history).

More words:

I started this blog in 2010. In 2013 I moved it to hakyll static site generator. The initial version was just 88 lines of haskell code.

I did not know much about hakyll back then and I kept it that way for about 10 years: it just worked for me. The only thing I missed were tag-based RSS feeds and article breakdown per tag. It prevented the blog from being added to thematic RSS aggregators like Planet Gentoo. But it was not a big deal. I though I would add it “soon” and never did.

The only “non-trivial” tweaks I did were dot support and gunplot support.

Fast forward to 2024 few weeks ago I boasted to my friend how cool my new gnuplot embeddings are. To what the response was “What pictures?”. Apparently miniflux does not like <svg> tags embedded into <html> and strips them away leaving only bits of <title> tags that almost looks like original graphviz input :)

That meant my cool hack with svg embedding did not quite work for RSS feed. I moved all the embeddings into separate .svg files with this change.

It’s not a big change, but it does violate some hakyll assumptions. Apparently hakyll can output only one destination file for a source file. For example foo.md can only produce foo.html and not foo.html plus indefinite amount of pictures. There is a version support in hakyll, but it assumes that we know number of outputs upfront. It’s not really usable for cases like N unknown outputs from an input. To work it around I’m writing all the auxiliary files without the hakyll dependency tracker knowledge. I do it by defining Writable instance:

data PWI = PWI {
    pandoc :: H.Item String
  , inlines :: [(String, H.Item DBL.ByteString)]
} deriving (GG.Generic)

deriving instance DB.Binary PWI

instance H.Writable PWI where
    write path item = do
        -- emit page itself:
        let PWI pand inls = H.itemBody item
        H.write path pand
        -- emit inlines nearby:
        CM.forM_ inls $ \(fp, contents) -> do
            H.makeDirectories fp
            H.write fp contents

Here inlines is the list of pairs of filenames and their contents to write on disk and pandoc is the primary content one would normally write as H.Item String.

While at it I disabled syntax highlighting in RSS feed as liferea rendered highlighted source as an unreadable mess. And miniflux just stripped out all the links and styles. The change is somewhat long, but it’s gist is a single extra writerHighlightStyle option passed to pandoc render:

pandocRSSWriterOptions :: TPO.WriterOptions
pandocRSSWriterOptions = pandocWriterOptions{
    -- disable highlighting
    TPO.writerHighlightStyle = Nothing
}

The last thing I changed was to switch from http:// links to https:// links by default. In theory it’s a one-character change. In practice that would break unread history for all RSS users. I worked it around by restoring http:// root link for current RSS entries with metadata change.

That way all new posts should contain https:// root links and all site-local links should automatically become https:// links.

Still no tag support. Maybe later.

Have fun!