• We’re doing some work to retire “advanced search” finders, which are now unnecessary in the brave new world of supergroup finders, so we can delete a lot of code and then make a bunch of simplifications to finder-frontend. But to avoid confusing people, we’re making the supergroup finders look like the “advanced search” finders if you arrive from a topic page. So visually nothing will change, but internally stuff will.

    I’ve only done a small bit of this, most has been done by others on the team. We’re nearly there now, maybe it’ll get done this week.

  • When we switched to the finder-based site search, we accidentally broke some metadata. It used to be the case that if your search results included content published by a previous government, it would say that next to the result. I fixed that, and you can see it in some of the results when you search for “every child matters”.

  • We had a chat with Amazon about how GOV.UK uses Elasticsearch, and made a small architectural change to our cluster off the back of that (changing from r4.xlarge instances to r5.xlarge). We also made the data sync work, fixed an issue with loading popularity data, and switched the licence-finder to use ES6 in production. We found a problem with the queries being sent to ES6, but once that is solved we should be able to start the A/B test.

    You can query the ES6 cluster right now by using the public search API, but it’s not returning all the results yet because of the query problem.


  • I spotted an odd pattern in GitHub URLs when browsing a directory tree at a particular branch. Here’s a path on GitHub broken down into its components:

              repository                         directory
              vvvvvvvvvvvv                       vvvvvvvvvvvvvvvvvvvvvvv
     ^^^^^^^^                   ^^^^^^^^^^^^^^^^
     user / org                 branch

    Imagine you’re parsing this:

    • User / org names can’t contain slashes, so that’s simple enough.
    • Repository names also can’t contain slashes.
    • The next bit says what sort of thing it is, in this case it’s “tree” meaning we’re browsing a tree.
    • Now it gets tricky: next are the branch and directory names, but those can both contain slashes and there’s nothing between them.

    How do we handle this? And what if it’s not unique? What if there’s a directory /environments/production in the msw/disable-xray branch and also a /disable-xray/environments/production directory in the msw branch?

    Well it turns out that git just doesn’t let you create branches which have this problem:

    $ git init
    $ git checkout -b foo
    Switched to a new branch 'foo'
    $ touch file; git add .; git commit -m "make a new branch"
    [foo (root-commit) 85508af] make a new branch
     1 file changed, 0 insertions(+), 0 deletions(-)
     create mode 100644 file
    $ git checkout -b foo/bar
    fatal: cannot lock ref 'refs/heads/foo/bar': 'refs/heads/foo' exists; cannot create 'refs/heads/foo/bar'

    Pushes are similarly blocked, so you can’t get around it by pushing the foo branch, deleting it, and then doing the same with the foo/bar branch.

    Making branches with slashes in the name actually create multiple levels of directories on the filesystem feels a bit… unsatisfying. But I guess it means if you’ve got a string $branch/$path and a snapshot of the repository, it has an unambiguous meaning and you can work it out without needing to backtrack.

  • I finally fell out of love with static site generators, and wrote some scripts to generate static sites. I wrote a memo: I replaced a static site generator with a script to generate static sites.

  • I started writing a Python script to find a new place to live. It scraped listings from Rightmove, had some metrics to discard incorrectly tagged listings (eg, even when you exclude studios, a lot of the “1 bedroom flat” listings say something like “a lovely studio apartment” in the description), and I was going to add a filter to only keep properties which have Hyperoptic (based on me scraping a list of 30,000 or so postcodes from the coverage map).

    But that all felt like a lot of work, so I asked on the GDS slack how people had found their flats. Naturally, two other devs linked me to their flat-finding scripts…

  • I watched HBO’s Chernobyl, which was really good. After the end of the final episode, there’s a 10-minute segment showing actual historic photos and videos, and giving information about the fates of the real people the characters were based on. It’s definitely worth watching.