• I made two changes to organisation pages, which are rendered by collections, an app which I hadn’t touched much previously. Recently the team changed how publications are displayed on org pages, adding the new “Services”, “Guidance and regulation”, “News and communications” (etc) headings, which draw content from “supergroups”. A supergroup is a collection of related document types, where we have some evidence that users understand the theme of the group. Document types, on the other hand, aren’t really something users should need to know about to find what they want.

  • I have a fix for an issue with the new site search where, if you search for something with a quote in it, the little “x” to remove the keyword doesn’t work. I don’t touch frontend stuff much, so this was a bit of a learning experience. I was really surprised to discover that javascript doesn’t have a built-in way to decode HTML entities in a string (without rendering the string to the page as raw HTML first).

  • I have another fix for an issue with synchronising our Elasticsearch data from production to staging. The problem is that:

    • Elasticsearch needs to be able to write to an S3 bucket, even if it’s only ever used to restore snapshots.
    • Sometimes Elasticsearch likes to write a metadata file to the S3 bucket when it does a restore.
    • Elasticsearch (at least, the AWS managed one) can’t be configured to grant ownership of the file it writes to the owner of the bucket.
    • S3 doesn’t have a way to impose a default access policy on new files.

    So sometimes our staging Elasticsearch will restore a snapshot and write a file to the production S3 bucket, which renders the production Elasticsearch unable to take new snapshots because there’s a file in the bucket it can’t read. This isn’t a serious problem, because we use separate (non-shared) buckets for backups, but it does mean our data sync keeps breaking.

    The AWS-suggested solution is to have a lambda which is triggered by an S3 notification, which then fixes the permissions. Which feels like a hacky work-around to me.


  • I added some microdata to memos, the memo listing, and my website. Probably not useful, but it wasn’t too tricky. I found the vocabulary simultaneously really rich and also very limited. For example, you can link a CreativeWork to a Person, like so:

    <span itemprop="author" itemscope itemtype="" style="display: none">
      <meta itemprop="name" content="Michael Walker">
      <meta itemprop="email" context="">
      <link itemprop="url" href="">

    …but you can’t link a Person to a CreativeWork. The closest I could come up with is:

    <li itemscope itemprop="knowsAbout" itemtype="">
      <meta itemprop="author" content="">
      <meta itemprop="datePublished" content="2018">
      <meta itemprop="sourceOrganization" content="University of York">
      <meta itemprop="inSupportOf" content="Ph.D">
      <a itemprop="url" rel="author" class="title" href="/publications/thesis.pdf">
        <span itemprop="headline name">Revealing Behaviours of Concurrent Functional Programs by Systematic Testing</span>
      <span class="description">
        University of York. Ph.D thesis. 2018.
        [<a itemprop="sameAs" rel="alternate" href="/publications/thesis.bib">bib</a>]

    Which feels a bit suboptimal. I more than “know about” my thesis, I wrote it! Another strange thing is that you can link a Person to an Occupation (the hasOccupation relationship), and a Person to an Organization (the worksFor relationship), but you can’t link an Occupation to an Organization or an Organization to an Occupation. So I ended up doing this:

    <section itemscope itemprop="worksFor" itemtype="" class="experience">
      <meta itemprop="roleName" content="Software Engineer">
      <meta itemprop="startDate" content="2018-04">
      <header itemscope itemprop="worksFor" itemtype="">
        <a href="">
          <img itemprop="logo" src="/logos/gds.png" alt="Government Digital Service (GDS)">
            <a itemprop="url" href="">
              <span itemprop="name">Government Digital Service (GDS)</span>
          <h2>Software Engineer</h2>
          <p class="when">Apr 2018&ndash;present</p>
          <li><a href="">alphagov</a> <i class="fa fa-github"></i>
      <p>I've worked on a variety of evidence-driven improvements to
        the <a href="">GOV.UK</a> publishing stack,
        both performance improvements and also resolving long-standing
        technical debt.  Mostly Ruby and Rails 5 / Sinatra, some Python
        2.  Also gave regular support to teams which did not merit a
        full-time developer.  Most of my work for GOV.UK is open

    So, rich in some ways, and strangely impoverished in others. It’s also kind of a pain adding all this markup by hand, so I’m once again wondering about making a machine-readable CV and having scripts to generate the LaTeX and HTML from that.

  • I discovered a neat thing about zsh’s “global aliases” (the killer feature of zsh in my opinion), you can make an alias that executes code. For example, here is a demo of using fzf to select two files which are then cated. I already use global aliases pretty heavily to simplify typing out common pipelines, maybe these $()-style aliases will also become a significant part of my workflow.