Personal Social Stream Technical

Review/Recap – “Accelerate: The Science of Lean Software and DevOps: Building and Scaling High Performing Technology Organizations”

Read Accelerate: The Science of Lean Software and DevOps: Building and Scaling High Performing Technology Organizations
I originally gave this book 3/5 on Goodreads, however, now I’ve revisited the book and made some personal summarisations from my highlights – I feel I need to revisit that.

My original just-finished-it instinct was that the research part of the book was too deep and took me into methodologies that I ultimately don’t care about. In fact, I nearly dropped out during that section. Having battled through, I realise that would have been a mistake – after that, there is a great leadership case study that really hits home some of the messages around the culture the book promotes needing to be developed and learned, rather than being the result of blindly applying archetypes.

Still, the research methodology part is a third of the book – at least structurally. I would advise any potential readers that that is entirely skippable if you’re prepared to trust the source and take the advice it’s pushing at face value. After all, the advice is backed with logic and reasoning – which for me, is more persuasive than “we spotted correlation in these heuristics for high performing organisations”, which isn’t as compelling for me as it may be for others.

The first part of the book is full of absolute gold, end to end. Mostly backing up a lot of the literature around the benefits of things like Continuous Delivery, Lean Product Management and working in agile teams. But it presents it in a nicely joined-up way, built around the hypothesis that high performing teams deliver quickly and build stable systems – offering four simple heuristics that theoretically cement those two features.

It digs a lot into building organisational cultures – citing some of the good research into what makes for a performing team at Google (Westrum generative cultures, learning organisations, etc).

It examines technical practices that contribute to the above – why and how continuous delivery works, the benefits of automation, versioning everything, test strategies, embedded disciplines (devops, devsecops, the job of testers in a highly automated world etc).

There’s large importance put on security, reserving a full section to it – where the conclusion (naturally, given some of the authors’ previous DevOps literature) is that you should build it into your process as early as possible.

I found the culture and leadership sections to be particularly good – backing up a lot of my own personal thoughts and biases. Lots about organisations forming their own paths and not just mimicking their way to culture change – which feels obvious, but probably isn’t given how much Cargo Cultism there is in tech.

I think that all in, there aren’t any new conclusions drawn in this book, but it is great to have them in one place. The book sees its own USP as the exhaustive research, but as I say, I found it exhausting, and not to be its real strength. The density and jumping off points & onward knowledge journeys are what really made it a great read for me.

Links Social Stream Technical

My current favourite thing is screencasts with hilariously over the top soundtracks. Never has writing a netlify function in Vim and deploying a hello world endpoint seemed so cool.

HOW many tabs?

My Problem

Everyone I work with at some point mentions my browser tab mismanagement. As of right now, I have 135 tabs open, but that’s definitely at the lower bounds of where I am typically at.

Screenshot of Firefox close tabs dialog.

For years I suffered under of Chrome’s tab Toblerone of death (“Tablerone”?), which constantly reminded me of my dysfunction with only an icon to identify which tab was which.

Recently, I started using Firefox again and it enabled my problem by presenting my open tabs in a carousel UI. So it wasn’t until my Mac told me I had no memory left that I realised how bad it had got.

The carousel obfuscates the exact number of tabs I have open – leaving the quickest (and only obvious) way to find out in Firefox through hitting CMD+Q – which triggers the above dialogue.

It may sound like this is kind of a weird brag, but it’s not. This problem isn’t one I’m delighted with having. My open tabs present a combination of lackadaisicalness (in that sometimes I just plain forget to quit out of them) and flawed aspiration (“I’ll get to reading that imminently”). By the time I get to declaring tab-bankruptcy and bookmarking them (using a combination of the OneTab extension and Pinboard ), they’ve amassed to a point where I sometimes no longer know why they’re there.

The Tool

Yes, I obviously have a problem. And once I’d admitted it, I started looking into ways to manage it. My chosen approach would be to better surface the above number, so at least it would become more quickly apparent that I’d gone too far.

I don’t want to move away from Firefox, and I don’t want to have to threaten to quit anytime that I want an update on my tab count. Nothing obvious was coming up in my searches until a little later I was pursuing another itch (hyper customisable window management, if that matters) where I stumbled upon some articles about a tool called Mjolnir, which had been forked to a more batteries-included tool called Hammerspoon.

The Hammerspoon splash page does a better job of describing it, but the tl;dr is it adds a scripting layer to MacOS using Lua. This seemed powerful enough for me to craft something that might suit my needs here.

Crafting A Solution

So with my hammer in hand, I needed to craft a way to address my “issue”. I’m a big fan of “Quantified Self“, and Continuous Improvement – both philosophies that involve surfacing and using data about yourself to improve a situation. So I figured – if that number is always in my face, shouldn’t I naturally try to optimise it now and again?

With that question in mind, I started looking at the Hammerspoon API, and initially, I figured this would be easy. Hammerspoon surfaces a method on a Window called tabCount, so this should be incredibly easy to integrate. Unfortunately not – the data it returns in Firefox isn’t representative of how many tabs you have open. At a guess, I expect Firefox isn’t using a native macOS tab implementation, and so this was an immediate dead end.

Then I started looking at methods in Firefox for surfacing this number. I’m sure I could automate a brute force counter somehow, but I don’t know how I might implement that in the background without interrupting the user. Firefox tracks “telemetry” for reporting browser usage. I found that stuff dead interesting, I’d encourage you to give it a read (all of the data is viewable in your instance at about:telemetry).

However, the only relevant number I’ve found is `browser.engagement.max_concurrent_tab_count`, and that only reflects the maximum number of tabs you’ve opened in that session. Which isn’t great when you’re trying to get that number down. Also, it doesn’t seem simple to query that data – you can get all sorts from FF’s SQLite databases, but I never landed on this value.

After lots of faffing, I finally discovered the session restore file found in your Firefox Profile directory at `sessionstore-backups/recovery.jsonlz4`. It’s the file that Firefox uses to restore your session when it runs out of memory because you’re an idiot who had too many tabs open. The file is massive – particularly if you have large amounts of tabs open – and in a slightly arcane format – lz4 – which on reading, you’ll see is a suitable compression format for the problem at hand. I expected that if we can parse that file quickly, it should be suitable for the job we have here – reporting regularly on the current number of tabs open.

Implementing a Firefox Tab Counter in Hammerspoon

First of all, I needed a fast way of reading that file. After some digging and pissing on with weird different libraries, I managed to get a fork of lz4json compiling locally – not only did it work, but it was fast:Screenshot showing that running the lz4jsoncat utility on my session backup took 0.011 secondsThen I needed to parse it. Instinct got me going down the “I’ll write some javascript for this” path, but I quickly realised that for fast feedback, I needed something close to the metal, simple, and again fast. So I dived into what describes itself as “sed for JSON data”, jq, and realised that this type of problem was its bread and butter.

Piping the above catted file into this strange incantation:

| jq 'reduce (.windows[] .tabs | length) as $length (0; . + $length)'

gives you a value that matches up exactly with the dialog box we get when hitting CMD+Q. Woop!

To try and explain the above, `.windows[] .tabs` gives you an array of tabs in each window. `| length` gives you the result of the length method applied to that array. And reduce applies a reduce operation to each of those windows, in our case adding them to get a total amount of tabs open.

Combine all of that into a little shell script, and by applying our ultra scientific `time` method of measuring the speed of it all, we still get sub-tenth of a second times:

timing the shell script gives a result of 0.089 seconds

That strikes me as a fast and feasible way of retrieving a tab count. So next step – surfacing it somewhere obvious. Working through the Hammerspoon getting started guide, you quickly come to a tutorial around simulating Caffeine in the menubar. This was incredibly straightforward and seemed like a nice obvious first step for getting stuff somewhere. So that’s what I did – we run the get open tabs script, and put the result of that in the menubar, using Hammerspoon’s PathWatcher utility to update that number whenever our session restore file changes.

I added a couple of enhancements, but nothing hugely complex – a limit constant that turns our number red when our open tabs are greater than it (at time of writing, I’ve exceeded it by 70 😬). And I had a play with creating an icon, which was incredibly straight forward. It’s based on this “format” (if you can call it that), which is self-documenting enough that I figured it out from looking at a couple of other instances of icons.

The script itself currently lives in my dotfiles, and I’d be happy to help anyone else trying to do something similar if they get stuck trying to replicate any of this. When it all comes together, my menubar looks a little bit like this:

A screenshot of my tab menu in action. Displaying a red

Which is everything I had hoped for. Has it fixed my problem? Goodness me, no. But it has been a fun learning exercise, and it makes me think twice before randomly middle-clicking links.

Links Personal Social Stream Technical

I’ve just signed up for Coursera’s Programming Languages course. Watched the intro, and can’t remember feeling so hyped about an MOOC.

There’s something deeply artistic and elegant in
the way programming languages fit together and I want you to stretch your mind, and
I want you to look at this course as a fresh way to look about software.
A fresh way to think about programming that you’ve never thought about before.
We’re going to make you uncomfortable.
We’re going to do things in a strange way.

Quite a claim.

Meta Social Stream Technical


I have removed analytics from my personal site. The numbers provide no value to me. Nothing they tell me is going to make me change the way I do stuff here, so why feed into the global shitshow that is needless counting and tracking?
Links Technical

Read Measuring Performance behind consent popups (Simon Hearne)

Opting in to cookies can load multiple megabytes of additional content. Here are tips on how to measure web performance both with and without cookie consent.

Thorough and pragmatic approach to measuring real-life page speed. Lots of great tips in here.

Couple of thoughts on React

Pretty sure it’s nothing nobody has noticed already, but I farted out some thoughts about React a while back. I figured I should publish more of this type of rubbish:

I have been watching Kent C. Dodds’ fantastic Advanced React Patterns course on Front End Masters. It occurs to me that a lot of the React patterns and principles which are being uncovered and taught to us all by the likes of Kent relate quite directly to more traditional patterns & principles – in the Gang of Four sense, regarding SOLID and further.

Some examples off the top of my head:

  • Render props feel like a sensible way of implementing dependency injection.
  • Pretty much every major React pattern is a different take on Inversion of Control.
  • Compose-all-the-things is generally a decent approach, good React seems to encourage this.
  • Redux implements Event Sourcing and the concepts of CQRS in an extremely elegant manner.
  • The Single Responsibility Principle underlies the entire concept of web components and React itself.

At first, I was quite snobby about it all: “They’re just coming up with new names for old things”, “all of this has happened before and will happen again”. But when you think about it, this is such a visual way to implement and explain these patterns.

It’s a lot easier to explain event sourcing with “it’s like a distributed redux”. The metaphors feel simpler to solidify when a method hidden in a class called `component.setDependency()` or a hugely overloaded constructor (`new Component(dependency1, dependency2, dependency3);`) becomes ``. I find that coupling, and its costs are so much easier to visualise when you can see the logic spelled out in this way.

I think this has a lot to do with React making interfaces so explicit. If you’re writing sensible components, it is trivial to convert from React’s XML to the correlating UML in your head. (If that doesn’t exist as software yet, I expect it will eventually.)

Projects Technical

Jankteki v0.4.0 released: fixes panel

The more eager eyed Jankteki users may have noticed the latest release. It consists of a small in-game ‘fixes’ panel where you can manually manipulate the game state without having to look up relatively arcane chat commands.

Here’s a video of this feature in action:

It’s the first interaction I’ve done with the game directly and was made pretty simple through Clojurescript / Om’s use of websockets. It was my first experience inspecting websocket frames through the Chrome Dev tools, and it must be said – I found it to be an utter joy. Hopefully this leaves room for other enhancements down the way.

In the meantime, the next slate of work is going to be around creating a fuller user-model – adding notes, annotations, that sort of thing. Maybe even the dreaded “shitlist” feature. More information on all of this / priorities etc can be found on my Trello roadmap

To keep up with this stuff, you might follow me on twitter (but this comes with its own price), you could subscribe to the YouTube, or you could just watch this space.

Again, if you haven’t yet, and this might be useful to you – download it from the Chrome Webstore and raise problems / missing features / pull requests on the Jankteki GitHub page.

Projects Technical

Jankteki: a Chrome Extension

In the past year or so, the collectible card game, Android: Netrunner has just about taken over my life. I won’t go on about why here, but it’s great. You should play it.

One of the more popular ways of playing it is through an online open-source implementation, Legally dubious, it’s a Clojure implementation of the game – involving a huge percentage of cards, providing automations for 1000s of rules and all sorts of interactions / exceptions / custom rulings. Needless to say, it’s a fantastic amount of effort being provided by a dedicated team of developers, led by Minh Tran.

As the game and platform have become more popular, the types of players on there have varied with it. More players is good, it’s more people and decks to play against. It also comes with a trolly underbelly. Ragequits and rude users used to be unheard of there, they’re now part of the landscape.

Enter: Jankteki

Navigating all of this can be frustrating at times. Whether it’s remembering who the bad players are, or looking for people you know to play (or watch play) – these are currently hard to do. And improving the platform at the same time as keeping up to date with card implementations is a slow process. So I built myself a tool – Jankteki, a chrome extension for

I intend for it to be a suite of tools to make using a slightly better experience day-to-day. It’s called Jankteki, because the hacky nature of building features over a ReactJS UI makes for flickers, jank, and breakages all over the place. So use it at your own risk.

The only real feature currently implemented is friends, which you can see wonderfully narrated here:

Why not raise a PR?

I know a bit of Clojure, and I enjoy writing it – but I’m nowhere near being able to contribute to a project the scale of Jinteki. There’s also the fact that my roadmap might not correlate with the Jinteki roadmap, I have features planned to scratch my itch that might be a year or two away from being even discussed in the main repo.

Some of them might not even be the job of the main webapp.

And what is this roadmap?

The full in-progress roadmap can be seen on a Trello board I am working from. But here’s a quick overview (it is completely subject to change):

  • I’m aiming to put in a toolbar for running console commands (deal net damage, remove counters, etc) in order that you don’t have to remember syntax or look it up every time the game state needs manually adjusting.
  • Notes for users. I constantly forget who people are from their usernames, a notes field could track that. It could be used to remember what type of decks they play. Or even to note who “gg”s before they leave (this is disproportionately important to me).
  • The killer feature I’ve talked about, but not implemented yet (I half implemented it and took it out), is some sort of bad-players list. Just a way of flagging a user visually as someone to avoid in the future. It would purely be a personal shit-list, not like a communal feature – both by design, and cos it’s outside the scope of this thing.
  • Game log recording / analysis. This sits in the “I’m unsure this would ever become a feature of the webapp” column. But I love the idea of logging wins and losses vs opponents, factions – methods of losing – and obviously which deck you were playing as at the time. Just as a way of tracking how you’re getting on and what works.


Building this plugin hasn’t been smooth. React’s virtual DOM makes it a nightmare to manipulate constantly changing elements – an element that indicated a friend two seconds ago may contain something else completely.

This plugin is also tightly coupled to class and element names in the codebase, if they change something, we’re always going to be playing catch-up. I’m ok with that. I don’t know how Minh will react to the existence of this (or if he’ll even care :D), but I hope Jankteki will benefit the team in both taking the demand for certain features, and also proving and disproving features before implementing them.

Links for Jankteki

So where to get it?:

And yes – it’s a bit rubbish at the mo, but it proves a concept, so please be open minded with it. There are bugs with pinned friends not showing when navigating to pages, if you have issues, try going directly to or just refreshing the page – I am actively working on that.

Also – things are liable to break, pre version 1 (I don’t know what that will look like), I’m pushing to the web store as often as I’m adding features / fixing bugs. So prepare to be annoyed.

Personal Technical

Quantified Fatigue

I have had a post in draft since the beginning of the year, all about analysing my running streak as it stands. It picks out the miles I’ve done on it, average mile per run, distributions of milages, that sort of thing. Loads of self indulgent wank, that I felt at the time was “interesting”. Reading back on it now, it’s anything but.

In the middle of composing it (according to WordPress, I last touched it 11th February), I stopped measuring pretty much everything. I just got sick. Sick of weighing myself. Timing myself. Caring how far I had ran.

Measuring shit definitely works – I have no doubt of that. Anytime I’ve measured something, I have consistently optimised towards that metric – be it weight, speed, bacon eaten. In the case of weight, the opposite also holds true – I put on about 2kg in the month following my “sick of this shit” tantrum.

I was still exercising plenty – running at least 30 miles per week in that time, in addition to a couple of BJJ sessions. I just happen to have a better appetite.

All of this has made me think – is the value in the “quantified self” movement that people are explicitly going out of their way to measure and observe things that they would like to optimise? If we were able to get these metrics at any point without installing apps, or buying devices, or just plain writing them in spreadsheets, would this value disappear? Beyond being able to say “oh, I’m fat because I do little exercise and eat too many calories” (unlike all of those other fat people), why is a graph such a motivator?

I understand that the whole thing is more faceted than “MAKE THE GRAPH GO BIGGER” – there is accountability (to both the tech and other people), QS allows you to find patterns and correlations in the data that you might not have otherwise noticed, as well as a billion other reasons for its existence. It can’t be a coincidence that the latest Apple and Samsung products have a “health & fitness” spin on them.

I should note here that I’ve not touched this post since March. The above feelings remain true, but I am a flip-flopper. Picking it back up again in mid-June:

Since I started writing this post, I am once again measuring stuff. I want to lose weight (again) because the metaphorical yoyo has retouched the hand. I’m measuring my calorie intake on MyFitnessPal, and that hooks up passively with Endomondo to measure what I’m burning.

But my reservations about QS and “Quantified Fatigue” stand – measuring everything explicitly is too difficult, and I’m concerned about how useful implicit & ubiquitous measurement would be after the fact. There’s no real conclusion here – implicit & ubiquitous doesn’t exist, it remains to see how iOS8 & the S5 will perform in this area, and I’m likely to lose a bunch of weight, get bored and put it all back on again. I just figured it was about time I did some writing here, terrible or otherwise.