Orating honours

2025-09-05T00:00:00+00:00

This is a bit of a niche one but – here in Cambridge</a> we have a number of quaint / historic / weird / cool (take your pick) traditions. One that I think is at the quite cool end is that, when we award honorary degrees, the University Orator</a> writes and delivers an oration introducing the honorand and explaining why they’re getting the degree. In Latin. And, occasionally, also with fragments in Greek and Sindarin.1</a> And when I say “delivers”, I really mean “performs” based on what I saw when in the audience for the honorary degree ceremony this year for the first time.

Happily, as we’re not totally behind the times, English translations are also provided, in the Reporter</a> naturally. And I decided to take a look back through some as I thought some of the ones I saw this year were pretty good. And as it was a slow Sunday afternoon, I went back as far as the online Reporter allowed. And here are the results – a list of links to all of the oration texts that are online, with some honorands called out where I thought they were particularly cool or interesting or moving,2</a> or I’d heard of the recipient.3</a>

Enjoy?

Year</th>	Honorands</th></tr></thead>
2025</a></td>	Stormzy</td></tr>
2024</a></td>	Kip Thorne</td></tr>
2023</a></td>	</td></tr>
2022</a>4</a></sup></td>	Roger Penrose • Simon Schama</td></tr>
2020</td>	no ceremony due to the COVID-19 pandemic so none conferred</td></tr>
2019</a></td>	Doreen Lawrence • Jane Goodall</td></tr>
2018</a></td>	Borys • Michael Edwards</td></tr>
2017</a></td>	Stephanie Shirley • Sophie Wilson • Manuel Castells</td></tr>
2016</a> [pdf</a>]</td>	Keith Peters • Jonny Ive</td></tr>
2015</a> [pdf</a>]</td>	Julia Neuberger • Neil Macgregor</td></tr>
2014</a> [pdf</a>]</td>	Albie Sachs • Yusuf Hamied • Martin Rees • Ian Mckellan</td></tr>
2013</a></td>	Hilary Mantel</td></tr>
2012</a></td>	Peter Higgs • Alfred Brendel</td></tr>
2011</a></td>	</td></tr>
2010</a> [pdf</a>]</td>	Andrew Wiles • Evelyn Glennie</td></tr>
2009</a> [pdf</a>]</td>	The Aga Khan • Melinda Gates • Bill Gates • Shirley Williams • William Broecker [ global warming ] • Amartya Sen</td></tr>
2008</a></td>	John Sentamu</td></tr>
2007</a></td>	Hans Blix • Ann Cotton • Onora O’Neill • Nicholas Stern</td></tr>
2006</a></td>	</td></tr>
2005</a></td>	</td></tr>
2004</a></td>	Robert Horvitz • Quentin Blake • John Dankworth • Cleo Laine</td></tr>
2003</a></td>	Alan Macdiarmid • John Sulston • John Adams</td></tr>
2002</a></td>	</td></tr>
2001</a></td>	Sydney Brenner • Mamphela Ramphele • Margaret Atwood</td></tr>
2000</a></td>	</td></tr>
1999</a></td>	Desmond Tutu • Gordon Moore • Miriam Rothschild</td></tr>
1998</a></td>	Aung San Suu Kyi (in absentia!) • Barbara Castle • Helmut Kohl • Aaron Klug</td></tr>
</td>	</td></tr> </tbody></table> …and there, sadly, the Reporter online record ends.</p> Elvish to those not well versed in Tolkien. I once posted to `rec.arts.sf.tolkien</code>. Just saying. ↩</a></p> </li>` `Seriously. See, for example, those for Doreen Lawrence</a> or Yusuf Hamied</a>. ↩</a></p> </li>` `I am not particularly well-versed in culture, being mostly a geek of sorts. Whatevs. ↩</a></p> </li>` Included celebration of those conferred in absentia in 2021 due to COVID-19. ↩</a></p> </li> </ol> </section> Resizing swap partition on NixOS 2025-08-13T00:00:00+00:00 I recently decided that I had over-egged the pudding and allocated too much space to my swap partition (mounted as /tmp</code>) under NixOS. At the same time, it had got to the point where I was bouncing off the ceiling a little with the other two partitions, /home</code> and /</code> (wherein lives the Nix store).</p> I tried gparted</code> first for a handy GUI but got nowhere due to (I think) Wayland:</p> $</span> nix profile install nixpkgs#gparted.out</span></span> ...</span></span> $</span> gparted</span></span> localuser:root</span> being added to access control list</span></span> Error</span> executing command as another user: No authentication agent found.</span></span> localuser:root</span> being removed from access control list</span></span> $</span> sudo gparted</span></span> Invalid</span> MIT-MAGIC-COOKIE-1 key</span></span> </span> (</span>gpartedbin:360533</span>)</span>:</span> Gtk-WARNING</span> *</span>: 19:19:25.626: cannot open display: :0</span></span></code></pre> So the obvious solution was to reduce the size of the swap partition and increase the other two. However, doing this for my setup – a LUKS-encrypted device partitioned using Logical Volume Manager (LVM) – turned out to be a little less obvious than it might have been, and searching the interwebs turned up several not quite correct answers, I thought I’d make a note here.</p> NB. Don’t pay too much attention to the sizes of partitions below: I cribbed these notes from shell history but re-ran some commands to get the output as I’d already closed that shell so they may not be entirely coherent. The commands should be right though.</p> Find your partition names using lvscan</code>:</p> $</span> sudo lvscan</span></span> ACTIVE</span> '/dev/vg0/home'</span> [<729.61</span> GiB] inherit</span></span> ACTIVE</span> '/dev/vg0/nixos'</span> [94.00</span> GiB] inherit</span></span> ACTIVE</span> '/dev/vg0/nixos-swap'</span> [16.00</span> GiB] inherit</span></span></code></pre> Normally I would then just use lvreduce</code> – but that fails on the swap partition:</p> $</span> sudo lvreduce</span> -r --size</span> 12GiB /dev/vg0/nixos-swap</span></span> File</span> system swap found on vg0/nixos-swap.</span></span> File</span> system size</span> (16.00</span> GiB</span>) is larger than the requested size (</span>12.00</span> GiB</span>).</span></span> File</span> system reduce is required and not supported</span> (swap).</span></span></code></pre> So instead…</p> </li> Delete the swap partition and recreate to the desired size (16GB in my case):</p> sudo</span> lvremove /dev/vg0/nixos-swap</span></span> sudo</span> lvcreate</span> -L</span> 16G</span> -n</span> nixos-swap vg0</span></span> sudo</span> mkswap /dev/vg0/nixos-swap</span></span> sudo</span> swapon</span> -v</span> /dev/vg0/nixos-swap</span></span></code></pre></li> Reallocate the now available space to other partitions:</p> # `/home` is easy: extend logical volume and resize the filesystem at in one go:</span></span> $</span> sudo lvextend</span> -r -L</span> +16G /dev/vg0/home</span></span> # however, can't resize the root filesystem as part of the `lvextend` because it's read-only.</span></span> # instead, find the relevant filesystem:</span></span> $</span> df</span> -Th</span> /</span></span> Filesystem</span> Type Size Used Avail Use% Mounted on</span></span> /dev/disk/by-uuid/8dd0d118-375f-4e8b-84bc-1a05b68889b7</span> ext4 92G 66G 22G 76% /</span></span> # ...and then manually resize it after extending the logical volume:</span></span> $</span> sudo lvextend</span> -L</span> +12G /dev/vg0/nixos</span></span> $</span> sudo resize2fs /dev/disk/by-uuid/8dd0d118-375f-4e8b-84bc-1a05b68889b7</span></span></code></pre></li> </ol> And that’s it. Swap partition is reduced in size, and the space is reallocated to the other two partitions, /home</code> and /</code>.</p> Re-key-remapping 2025-07-26T00:00:00+00:00 As all right-minded people must, I remap the historic, redundant, absurd CAPSLOCK</code> key1</a></sup> to be a more conveniently accessible current, useful, sleek CTRL</code> (control) key on any given keyboard. I also do some other minor mods to assist retention of the little sanity I still have – as I get older and more decrepit, it gets harder for me to handle change. Deal with it.</p> I used</a> to do this</a> on my Macbook</a> using a tool called Karabiner Elements</a>. But now I’m on Linux, that’s not a thing. So I use the excellent keyd</code></a> instead. The UK Thinkpad keyboard being a lot less daft in design than the UK Macbook keyboard, the modifications are also much smaller, which is nice.</p> However the canonical First Modification, CAPSLOCK</code> to CONTROL</code> remains the key one. For added convenience, given I rarely need to “tap” CTRL</code> I also make a tap on the CAPSLOCK</code> key into an ESC</code> (escape). And there the fun starts.</p> It turns out the keyd</code></a> recommended way to achieve this is with the configuration stanza</p> [main]</span></span> capslock = overload(capslock, esc)</span></span></code></pre> …which is mostly fine and has worked well for me for some time. But I realised I was getting increasingly irritated that I seemed to get spurious ESC</code> key presses when using CAPSLOCK</code> as CTRL</code>. sudo keyd monitor</code> and it became clear that the semantics of the overload()</code> function weren’t quite as I thought. From man keyd</code>,</p> overload(<layer>, <action>)</span></span> Activates the given layer while held and executes <action> on tap.</span></span></code></pre> I had assumed this meant that <action></code> did not execute if the assigned key was held, but it doesn’t quite: it seems that a tap is deemed to have occurred if the key is not chorded</em>. So sometimes, being a bit slow, when I hold CAPSLOCK</code> thinking I want to chord and thus have it treated as CTRL</code> but then realise I don’t and I just let go, I still get an ESC</code> generated – even if it was a second or two that I was holding it down for.2</a></sup></p> So I read a bit more of the manpage can came across the timeout</code> function</p> timeout(<action 1>, <timeout>, <action 2>)</span></span> If the key is held in isolation for more than <timeout> ms, activate the</span></span> second action, if the key is held for less than <timeout> ms or another key</span></span> is struck before <timeout> ms expires, execute the first action.</span></span> </span> E.g.</span></span> </span> timeout(a, 500, layer(control))</span></span> </span> Will cause the assigned key to behave as control if it is held for more than</span></span> 500 ms.</span></span></code></pre> …used as capslock = timeout(esc, 250, layer(control))</code> but that turns out also not to be quite right. In the end, it looks like the right answer is overloadt2</code>:</p> overloadt(<layer>, <action>, <timeout>)</span></span> Identical to overload, but only activates the layer if the bound key is held</span></span> for <timeout> milliseconds. This is mainly useful for overloading keys which</span></span> are commonly struck in sequence (e.g letter keys).</span></span> </span> Note that this will add a visual delay when typing, since overlapping keys</span></span> will be queued until the timeout expires or the bound key is released.</span></span> </span> overloadt2(<layer>, <action>, <timeout>)</span></span> </span> Identical to overloadt, but additionally resolves as a hold in the event of</span></span> an intervening key tap.</span></span></code></pre> The end result is the following stanza in my NixOS</a> configuration:</p> keyd</span> =</span> {</span></span> enable</span> =</span> true</span>;</span></span> keyboards</span>.</span>default</span> =</span> {</span></span> ids</span> =</span> [</span> ""</span> ];</span></span> settings</span> =</span> {</span></span> main</span> =</span> {</span></span> # capslock -> (held) ctrl, (tap) ESC</span></span> capslock</span> =</span> "overloadt2(control, esc, 150)"</span>;</span></span> rightalt</span> =</span> "leftalt"</span>;</span></span> };</span></span> shift</span> =</span> {</span></span> grave</span> =</span> "G-4"</span>;</span> # S-` -> €</span></span> };</span></span> };</span></span> };</span></span> }</span>;</span></span></code></pre> The timeout</code> of 150ms feels “about right” at this point – 100ms was borderline not enough, and 200ms was definitely annoyingly long.</p> Not to be confused with the elegant, historic, radical RUN/STOP</code> key of course. C64 FTW! ↩</a></p> </li> Honestly, sometimes I really am that slow. I’m old see? ↩</a></p> </li> </ol> </section> Whither AI? 2025-07-05T00:00:00+00:00 I am hardly the first person to comment1</a></sup> on this – I am given to understand AI has been a topic of some interest to many for a few years now. I’m sure I’ve seen, and possibly even re-tooted</a> things about it in fact. I’m afraid I just don’t keep up.</p> But recent experiences reviewing for a couple of systems/networking venues has led me to feel I need to ask: WHY</strong>? More pointedly, why does the following seem like good motivation for a research paper?</p> There is a complex and important task that currently requires considerable expertise to carry out because it is important to be precise and get it right.</li> The task in question can be described imprecisely using natural language by non-experts.</li> AI (inevitably, some large-language model) can take that natural language description and, after training, produce some output that is stochastically like unto what an expert might produce given the same underlying problem, having brought to bear their expertise.</li> Thus we build an AI that can take the non-expert’s imprecise description and show that sometimes the output it produces is not so wrong as to fail some ad hoc</em> tests of utility that we introduce.</li> </ol> Based on things I’ve recently reviewed “not so wrong” above means “error rate of no more than 25—30% when taking expertly generated natural language prompts as input”. Which is to say, probably not the sorts of input prompt that a non-expert might produce.</p> Network configuration and management is the domain I’ve seen this argument made in most recently. Which seems quite strange to me because I always thought that a 25% error rate in configuring, e.g., your enterprise network security perimeter would be bad. But apparently not if it’s done by an AI.</p> More generally, why do we want to build tools that allow untrained experts to do a job when mistakes are high impact, it requires a trained expert to detect those mistakes, and those tools by design</em> only produce statistically valid output? An error rate of once in a blue moon is categorically worse than a zero error rate if the error involved can leave your entire digital estate open to compromise.</p> If the big issue here is that experts sometimes make typos when editing the configuration files, maybe building some domain-specific languages or better user interfaces or verification techniques or other tooling would be a better way to help them not do that than replacing them with tools that by design</strong> are only ever probably about right.</p> So please stop justifying your AI application research by saying simply that it allows non-experts to carry out expert work! I’m much more likely to be convinced by uses of AI that make experts more productive</em> – though don’t get me started on how to measure productivity because I don’t know except via means which are expensive and time consuming, and it really seems that very few people can be bothered doing that.</p> Ok fine. I admit it. This is a rant. ↩</a></p> </li> </ol> </section> Artificial Intelligence, but at what cost? 2025-05-15T00:00:00+00:00 Our inestimable and most excellent Chaplain, Revd Dr Helen Orchard, likes to have a theme for the Sunday evensong services for the term. Back in Michaelmas 2023 it was … AI. I said I’d help find someone to give a sermon from a technical perspective but then signally failed to do so (sorry!). So in the end I said I’d do it, even though AI is not my thing and I’d never given a sermon before. Or, for that matter, attended evensong. Take the opportunities offered and all that.</p> I realised this week that, although a few people at the time had asked for copies, I’d also done nothing about that (I am nothing if not consistently rubbish). So here’s the text, more or less as given, on 15 October 2023. Note that the golden eagle I mount is a rather fine lectern in our Chapel (pictured). Nothing more salacious than that. Filthy minds.</p> Three editorial notes given that it’s been over a year and a half since I gave this (my! how time flies…):</p> I allude to this but should be clear: the neural network is not the only technological approach to producing AI – several others exist and are both useful and used, machine learning being one that’s particularly productive in recent years. However the most hyped was and still seems to be various forms of neural network so that’s what I focused on.</li> I refer to “static datasets” because the versions of ChatGPT at the time were trained infrequently on a given dataset of the moment. Training updates now seem much more frequent (perhaps weekly), user context is maintained throughout a chat session, and user feedback sought at the end. So while it’s still technically true that the datasets involved are static, it’s much less noticeable.</li> The example of “God save the” worked particularly because this was only about a year after Queen Elizabeth II died, so “queen” was likely still the instinctive response of many.</li> </ol> Finally, just in case it’s not clear – I tend toward the sceptical end regarding AI. Potentially a useful tool in some circumstances but all claims about AGI are nonsense and the singularity won’t happen because of the machines. Human stupidity on the other hand seems without bound. And always follow the money.</p> </a>A photograph of a fine golden-coloured lectern, the head of which is an eagle</a>.</figcaption></figure> As I mount the golden eagle for the first time, I should say that I am not normally given to preaching – though my children might disagree with that statement – but as the theme this term is Artificial Intelligence, Helen asked me to speak to you about that from the perspective of a computer scientist. Unless you catch me in a pub after a couple of pints, I am also not given to philosophising, so I will limit myself to the physical reality of Artificial Intelligence</em>, or AI</em>. Specifically, what is it and what does it cost. I will use AIs that generate text as examples, as these so-called Large Language Models</em> have been the focus of considerable interest in recent months, but the same basic mechanisms and problems apply to AIs used to generate images, music, videos and so on.</p> First, what is it. AI is a catch-all term for a set of technologies that attempt to replicate whatever we call “intelligence”. Computer scientists, cognitive psychologists and mathematicians have worked on these various technologies for decades, but the current vogue is very much for a particular set of mathematical techniques that try to produce brain-like behaviour by modelling inter-connected neurons.</p> Each neuron is stimulated by one or more input signals which it combines to produce an output signal with some probability. The outputs of some neurons are connected to the inputs of some other neurons, creating an enormous network. The effect in our brains might be that an input signal “I want a biscuit” results in an output signal that causes us to move an arm to pick up a biscuit. In a modern “generative AI”, the input might be a sentence or paragraph or two of text, and the resulting output might be an image or a sequence of words.</p> As a simple example of what I mean, if I asked you to give the next few words in the phrase starting “God save the” you might say “king send him victorious”. You have just performed inference using your own language model, generating some likely output text given three words of input. I’ll come back to that example later.</p> </blockquote> I said the inputs were combined to produce the output with some probability, but how exactly? The process for combining inputs involves a set of parameters that are determined by finding the values that give the best fit some a priori data. This is known as training if you’re an AI specialist, or parameter fitting if you’re a statistician.</p> A simple analogy: you may recall that a straight line is defined by two parameters, its slope and any point on the line. If you had a set of two dimensional data points that you thought were straightforwardly related, you might try to discover that relationship by drawing the best straight line you could through them; but which particular line would you think was the best? A reasonable choice might be the one that minimised the total distance from the line to each point. For an AI the maths is a little more complex, but that’s basically what happens: training finds the parameter values that give the best fit to a large set of training data.</p> So that’s a modern AI: a statistical model that, when stimulated by one or more inputs, produces outputs with some probability. The inputs might be words or images or some other thing, and the outputs might be words or images or some other thing. The underlying model might be wrapped up by other models that, for example, try to filter out undesirable outputs or provide for different ways of consuming inputs.</p> It is the sheer scale that makes this work: your brain has perhaps 100 billion neurons each of which might connect to 10,000 other neurons for a total of perhaps one million billion connections, whereas an AI such as a recent version of ChatGPT might have 175 billion parameters but each connected to just hundreds of others. The underlying mathematics has been known for decades; it is the combination of massive training datasets and the enormous computational resources of the cloud that have enabled us to build these AIs.</p> </blockquote> Second, ignoring the hysteria around so-called Artificial General Intelligence and The Singularity, what costs do these AIs incur?</p> To return to the example I used, I said that you might have completed the phrase “God save the” with the words “king send him victorious”. In some sense that is the “correct” completion. But perhaps some of you would have initially thought “queen send her victorious”. And I have at least one friend who would naturally respond “queen and her fascist regime”.</p> Human experience is varied and personal – the training process I described typically uses large static datasets collected by scraping the Internet. While the resulting AI can be configured not always to produce identical outputs given identical inputs, the training process does naturally lead to a kind of homogenisation. Simplistically, if your group is not represented in that training dataset, its experience will not be represented in the AI and thus will not be reproduced in the output. Worse, if the training data contains misrepresentations or attacks on your group, the AI will by default capture and perpetuate them, already observed to be a particular problem for women, Jews, and many minorities.</p> Further, I mentioned that training data is scraped from the Internet – but as the musical Avenue Q famously put it, “the Internet is for porn”. A lot of that text is rather fantastical and describes actions generally unacceptable in polite society, so the companies producing and operating AIs try to create guardrails by building other models that filter offensive outputs generated by their AIs – but how do you train such a model? You need to start with examples of offensive output that are labelled as such so that you can train a model to differentiate between what is offensive and what is inoffensive. But creating that labelled data involves human labour. For example, OpenAI were reported as outsourcing this activity to workers in Kenya paid less than $2 per day to label perhaps 200 paragraphs per day of offensive input text with the type of offensiveness: rape, torture, incest, and so on. Unpleasant and psychologically damaging work.</p> </blockquote> There are also more practical problems posed by the resources used to create and operate AIs. In particular, energy and water.</p> It takes a lot of computation to train and operate a large popular AI – OpenAI reported about three and a half thousand petaflops-per-second-days in 2020 to train their GPT model, where a petaflop represents a million billion computations. That is, about 10 years of a computer running at one petaflop per second. For comparison, your phone might achieve 0.1% of that performance. But as the bumper sticker has it, the cloud is just someone else’s computer – in the case of a training run for a large AI model, several hundred thousand computers in a datacenter. For example, Microsoft’s Iowa datacenter was built out for training models for OpenAI and has 285,000 standard processor cores and 10,000 GPUs (more powerful and power-hungry processors that you might be familiar with using if you’re a gamer).</p> This means CO_{2</sub> from the energy to power the computers plus water to cool them. How much? Well, estimates computed for earlier, smaller, models put the CO_{2</sub> footprint of a single training run at roughly the same as a round-trip flight from New York to San Francisco. Once trained, individual queries are comparatively cheap – but ChatGPT experienced the fastest ever growth of an Internet service. Earlier this year it was estimated as serving hundreds of millions of queries per day resulting in power consumption of perhaps 1 gigawatt-hour each day – the equivalent of 33,000 American households.</p>}} As for water, Microsoft has reported that its global water usage increased 34% from 2021 to 2022; Google’s increased 20% in the same period, but from a higher baseline. The increase is believed to be substantially due to training and operating AI. A group from University of California at Riverside estimate that each “conversation” with ChatGPT uses, directly and indirectly, about a pint of water – and this generally needs to be clean drinking water that will not leave residues that clog systems. The month before GPT-4 training was completed, Microsoft’s Iowa datacenters consumed 11.5 million gallons, about 6% of the district’s drinking water. The amounts vary based on season and location of the datacenter but it seems clear that water consumption is very substantial and could impact local communities and ecosystems. And of course, there is a tension here: cheap and green solar energy improves the carbon footprint but the associated higher temperatures usually also worsens the water footprint as more cooling is required.</p> </blockquote> So there’s a view of AI – an impressive set of mathematical and computational techniques that can recreate some human behaviours to some extent in some circumstances, at significant practical and moral cost. My own view is threefold.</p> First, using the phrase “Artificial Intelligence” to describe these technologies, rather than something less emotive such as Computationally Intensive Statistics, inevitably generates a very strong hype cycle, and we are currently at a point in that cycle where a welcome degree of scepticism is starting to come in and people are more actively questioning what exactly these technologies can and can’t do.</p> Second, we have largely proceeded to date without concern for any of the costs I discussed earlier, and – also welcome – that is changing: the costs are significant and we cannot ignore them.</p> Third, there are interesting legal and economic tussles taking place as to who owns the training data, who owns the weights – that is, the AIs – produced, and by whom and how should AIs be regulated. In particular, it is notable that many companies are claiming that there is a need for regulatory barriers to be introduced – but those are the companies that have already reached a scale where they can overcome those barriers, so such barriers will serve only to keep newcomers out of the marketplace, entrenching the existing power of “big tech” (OpenAI, Google, Microsoft, Amazon, Meta, etc).</p> Finally, as I used the word hysteria earlier to describe hyped fears of Artificial General Intelligence and the Singularity – please</strong> be sceptical of anyone claiming that as a serious existential risk, particularly</strong> if they are associated with aforementioned “big tech”! I view most of that discourse as a “dead cat” strategy, an attempt to distract from the current harms they are causing today by pointing to vague, nebulous, yet potentially infinite future harms. For more about the quite startling beliefs of many of those sounding those alarms, I recommend reading about the TESCREAL set of ideologies – Transhumanism, Extropianism, Singularitarianism, Cosmism, Rationalism, Effective Altruism, Longtermism.</p> Thank-you.</p> </blockquote> References</a></h2> Background</a></h3> “Language Models are Few-Shot Learners”, OpenAI, 2020. https://arxiv.org/abs/2005.14165</a></li> “On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?”, Bender et al, FAcct’21. https://doi.org/10.1145/3442188.3445922</a></li> “The Internet is for porn”, Stephanie D’Abruzzo & Rick Lyon, Avenue Q. https://genius.com/Stephanie-dabruzzo-and-rick-lyon-the-internet-is-for-porn-lyrics</a></li> </ul> Hidden Work</a></h3> “OpenAI Used Kenyan Workers on Less Than $2 Per Hour to Make ChatGPT Less Toxic”, Time.com, 2023, https://time.com/6247678/openai-chatgpt-kenya-workers/</a></li> “Behind the secretive work of the many, many humans helping to train AI”, NPR, 2023. https://www.npr.org/2023/06/26/1184392406/behind-the-secretive-work-of-the-many-many-humans-helping-to-train-ai</a></li> </ul> Energy</a></h3> “Energy and Policy Considerations for Deep Learning in NLP”, Strubell et al, 2019. https://arxiv.org/abs/1906.02243</a></li> “Training a single AI model can emit as much carbon as five cars in their lifetimes”, MIT Technology Review, 2019. https://www.technologyreview.com/2019/06/06/239031/training-a-single-ai-model-can-emit-as-much-carbon-as-five-cars-in-their-lifetimes/</a></li> </ul> Water</a></h3> “Artificial intelligence technology behind ChatGPT was built in Iowa — with a lot of water”, AP News, 2023. https://apnews.com/article/chatgpt-gpt4-iowa-ai-water-consumption-microsoft-f551fde98083d17a7e8d904f8be822c4</a></li> “A.I. tools fueled a 34% spike in Microsoft’s water consumption, and one city with its data centers is concerned about the effect on residential supply”, Fortune, 2023. https://fortune.com/2023/09/09/ai-chatgpt-usage-fuels-spike-in-microsoft-water-consumption/</a></li> “Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models“, Pengfei Li et al, 2023. https://arxiv.org/abs/2304.03271</a></li> </ul> Rusty Nu POSIX 2025-05-07T00:00:00+00:00 A slight delay to this post, but happily I now have no more lecturing to do until 20271</a></sup> :)</p> I switched a year or two ago to using NixOS as my daily driver following about 15 years as an increasingly irritated MacOS user. Shortly before I had become interested in Rust as a systems programming language that seemed to marry several things I like about OCaml with several other desirable things from C and Python.</p> I then more recently observed something that I thought was interesting: there seem to be a lot</strong> of recent replacements of what were once completely standard and changeless POSIX utilities. I’m thinking things like grep</code>, find</code>, ls</code> and the like that I’ve been using uninterrupted, other than the occasional quibble over whether it was the original version or the GNU version, for about 30 years. Indeed, I have already raved (slightly) about just</code></a> as a make</code></a> replacement and its use with OCaml</a> and LaTeX</a>.</p> NixOS’ declarative configuration meant that I could actually see the list growing, all in one place – I suspect on other systems I wouldn’t have noticed in quite the same way because it would’ve been a much more incremental and diffuse process of change without a clear record of the choices made.</p> I thus find in my cli.nix</code></a> config that describes the CLI tools I expect, to have the following collection:</p> nu_posix</span> =</span> [</span></span> bat</span> # better cat</span></span> bottom</span> # btm ~ better top, htop, etc</span></span> broot</span> # interactive directory navigation</span></span> chafa</span> # terminal graphics viewer</span></span> ctpv</span> # terminal file previewer</span></span> cyme</span> # better `lsusb`</span></span> delta</span> # better syntax highlighting diff</span></span> dua</span> # disk usage, interactively</span></span> eza</span> # improved `ls`</span></span> fd</span> # `find` replacement</span></span> fend</span> # better CLI calculator</span></span> hexyl</span> # hex pretty printer</span></span> htop</span> # graphical top</span></span> iotop</span> # io top</span></span> jujutsu</span> # better git</span></span> just</span> # updated gnumake replacement</span></span> procs</span> # better ps</span></span> ripgrep</span> # rg ~ `grep` replacement</span></span> sudo-rs</span> # memory-safe `sudo`</span></span> uutils-coreutils-noprefix</span> # replaces GNU `coreutils`</span></span> viddy</span> # better watch</span></span> ]</span>;</span></span></code></pre> I think that most, if not all, of these are written in Rust: that particular language community seems to have a real enthusiasm for re-implementing long-standing tools but better, and I have to say I really appreciate it! When I say “better” I’m not particularly thinking of esoteric language features or development ideologies either. I mean better in two very particular senses:</p> Usability</strong>. Many of the older tools simply did not have great user interfaces and, when they were ok, they were not built using modern tooling. As a result getting documentation was somewhere between good and great if there was a decent man</code>-page, with a range of potential switches for more short form help or for cases where the man</code>-page was not installed – whether -h</code>, --help</code>, -help</code>, -?</code>, help</code>, or something else. The short-form help would, of course, be formatted in arbitrary ways.</p> The modern Rust-y replacements tend to use clap</code></a> as a reasonably standard command-line parser. As a result, they are remarkably consistent in usage and format, typically producing something that looks a lot like man</code>-page output in response to their -h\|--help</code> switch. In a world where man</code>-pages are often an afterthought or, even worse, replaced by info</code> documentation, I find this invaluable. They are also generally inclined to make greater use of modern terminal environments – eza</code></a> as a replacement for ls</code></a> is a good example of this.</p> </li> Performance</strong>. Old tools were originally built for old computers in old languages (largely C) and, whether this is language ideology or just the practicalities of engineering long-standing widely-used codebases, tended not to be radically updated.</p> Rust re-implementations, on the other hand, are from scratch – and Rust’s memory model appears to make it relatively easy for them to be made multi-threaded. On modern hardware this seems to make them startlingly higher performance than the alternatives. Tools I particularly appreciate for this include fd</code></a> replacing find</code></a> and ripgrep, rg</code>,</a> replacing grep</code></a>.</p> </li> </ol> Perhaps the most immediate example of the benefits of this that I’ve experienced is dua</code></a> via dua i</code>. Traditionally, when trying to clean up an uncomfortably full hard disk I would’ve ended up using some manual iterative application of either du -hS </code> or possibly something like find ... \| xargs du</code>. Or possibly written a Python script to do it for me. And it would’ve taken O</em>(hours) for me to find where the space was being used and to do something about it. And I would’ve found it tedious and deeply irritating.2</a></sup></p> In contrast, dua i</code> gives me a TUI interface to navigate the filesystem from wherever I run it, the ability to cumulatively mark files and directories for trashing or immediate deletion, with subdirectory space summaries – and does so across ~850GB / 3 million files in about 10-15 seconds without using any form of caching, database, or other such thing. As far as I can tell, simply by being efficient and multi-threaded.</p> If this is the future, sign me up. (At least for the bits like this that are good.)</p> …assuming I get back the same courses after my sabbatical that is. ↩</a></p> </li> I’m easily irritated. What can I say. ↩</a></p> </li> </ol> </section> Coping and Capping 2025-04-08T00:00:00+00:00 Well that was fun! Quite high up there in the set of things that I never even considered I might do would’ve been awarding degrees. But by dint of being President and thus standing in for Simon</a>, on Saturday I did exactly that.</p> The University held a Congregation for those being admitted to the Cambridge MA</a> (“Magistri in Artibus” – Master of Arts). Degrees are conferred by the Chancellor, Vice-Chancellor or nominated deputy. Apparently, typically, for this one and the main undergraduate congregation in July, that nominated deputy Vice-Chancellor is usually the Head of House for the College concerned. In this case, as President at Christ’s is effectively deputy Master (~ Head of House), it me.</p> So instead of the usual Batman-style black gown, I got to wear the rather natty Superman-style cope</em>, hence I was coped</em>. I also got to wear one of the fancy hats (“square cap” or mortarboard) for the first time, hence I was also capped</em>.</p> And as many other officers were also appropriately hatted, and out of respect to the role would cap</em> me at many available opportunities, the written advice I received was literally that “officers will cap you – you do not have to cap back” because the cope makes it difficult to do so.</p> So, in brief, it seems I coped with being coped and capped but could not cope with capping while coped. To add to the excitement, I also had to Do Some Latin. Thankfully it’s a dead language so I assume all it could do was turn in its grave while I butchered it.</p> I hope everyone had a good time anyway!</p> Rewarding influence 2025-04-05T00:00:00+00:00 EuroSys 2025</a> was co-located with ASPLOS 2025</a> this year. Other commitments meant I (again) couldn’t stay for the whole conference, attending primarily because two students had papers in the TDIS workshop</a>.</p> But happily I was</em> able to stay for the first day of the conference – “happily” not only because it gave me a chance to catch up with some old friends I hadn’t seen in a decade or more, but also because the Mirage unikernels paper</a> which appeared at ASPLOS 2013</a> won one of two ASPLOS 2025 Influential Papers awards</a> :)</p> This is obviously very flattering – typically doing research is necessarily its own reward because the work can seem fruitless much of the time. Even when a paper gets written and submitted it will most likely be rejected – I think EuroSys this year reported something like a 12% acceptance rate, so rejection is a priori</em> the most likely outcome. Finally, if the paper does finally get accepted, it will most likely sink without trace – perhaps a brief flurry of interest for a few months or so, the paper gets cited a few times, and then it fades away. This seems inevitable in a reasonably fast moving field that is also growing at pace – EuroSys had ~160 attendees in 2006, growing to ~330 in the 10 years to 2016, but hitting ~1100 this year; while submissions grew from ~200 in 2019 to 696 this year.</p> So to win an award recognising that others feel a paper actually had some influence is rare, and makes me very happy :) At the same time, it reinforces a couple of lessons that I really should’ve internalised by now.</p> The first is that papers inevitably get better for thoughtful considered feedback from experts a step or several away from the work – so drafts should be produced in plenty of time and distributed to anyone who’s willing to take the time for feedback. In the case of this paper the previous failed submission to OSDI 2012</a> had, let’s say, reviews of mixed quality. But one stood out, from Jon Howell (who signs his reviews so I know it was him), who gave us a firm “reject” which (in retrospect) was actually fairly well-deserved but in an incredibly constructive way. To paraphrase him slightly, the work was interesting but the paper was crap – and here’s how to rewrite it so it makes sense</em>. We basically did what he said, ASPLOS accepted it, and the rest is now history. (Over a decade ago, good grief.)</p> The second is that I simply cannot predict whether any research I’m doing is actually going to turn out to have any value. The only other equivalent award I’ve had was an INFOCOM 2024 Test of Time award</a> for our 2012 paper on a system called Thinkair</em></a>, about mobile-code offload from devices to the cloud. That paper had received considerably more than just one rejection prior to acceptance, and if I recall my final contribution correctly, I recommended not submitting it to INFOCOM as I didn’t think we’d done enough to address previous review comments.</p> Shows what I know. But then, how boring would life be without a little ignorance… :)</p> Not one but two accepted papers and a Ph.D. 2025-03-30T00:00:00+00:00 As I find myself once more on a train to parts unknown (to me at least), a brief update :)</p> The parts unknown in question is Rotterdam, NL (so really quite well-known to quite a lot of people, just not me) for EURO/SYS 2025</a> (being held jointly with ASPLOS 2025</a>, although I can’t stay for the whole thing unfortunately) and specifically the 3rd International Workshop of Testing Distributed Internet of Things Systems (TDIS)</a>.</p> Why? Happily the programme committee decided to accept two papers from my (ex-)students – which is nice :) The two in question are</p> Reckon-ing Kubernetes at the Edge using Emulated Clusters</a></strong> with Alessandro Sassi (University of Cambridge / Politecnico di Milano) and Christopher Jensen (University of Cambridge / Microsoft Research). This describes Alessandro’s M.Sc. research project undertaken as a visitor with my group. He built on Chris’ earlier work on Reckon, an emulator setup for examining consensus system behaviour</a>. Alessandor extended this to use ContainerNet enabling it to emulate Kubernetes clusters on a single node, and used this to examine Kubernetes performance in some edge network scenarios. Source available on GitHub at https://github.com/AleSassi/reckon-k8s</a>.</p> </li> LoRaLive: Efficient LoRaWAN Traffic Generation</a></strong> with Vadim Safronov (University of Oxford / University of Cambridge). This reports a component of Vadim’s Ph.D. work where he built a system to enable dense deployment LoRaWAN trace-playback using a minimal number of nodes while respecting legal constraints on duty cycles. Source available on GitHub at https://github.com/LoRaLive/LoRaLive</a>.</p> </li> </ol> Both nice tools that we hope might be of community interest!</p> The Ph.D. in question is Chris Jensen’s – happily he passed his viva on Thursday just gone, titled “Separating conflict-recovery from failure-recovery in distributed consensus”, examined by Tim Harris</a> and Aleksey Charapko</a>. Other recent passes include Al-Amjad Tawfiq Isstaif, titled “Contention-resilient overcommitment for serverless deployments” and Andrew Jeffery, titled “Modelling orchestration”. The race1</a></sup> is now on for the first to tech report</a>…</p> It’s not really a race. That would be weird. ↩</a></p> </li> </ol> </section> `nmcli` connection to internal WiFi 2025-02-14T00:00:00+00:00 Using my fancy (?) new(-ish) Linux laptop running NixOS</a>, I finally had cause to connect to our internal Wi-Fi network. This was not entirely trivial due to the various configuration options required. So here goes, for the record, what I did as an aide memoir for me and in case it’s useful for anyone else…</p> First, create the connection – the Wi-Fi network in question is named Internal-CL</code>:</p> $</span> sudo nmcli connection add type wifi con-name Internal-CL ssid Internal-CL</span></span> Connection</span> 'Internal-CL'</span> (8f1ddcc9-4b1f-4e5d-9992-522714685eb4) successfully added.</span></span></code></pre> Then, configure it:</p> $</span> sudo nmcli connection edit Internal-CL</span></span> </span> ===</span>\|</span> nmcli</span> interactive connection editor</span> \|</span>===</span></span> </span> Editing</span> existing '802-11-wireless' connection: 'Internal-CL'</span></span> </span> Type</span> 'help' or '?' for available commands.</span></span> Type</span> 'print' to show all the connection properties.</span></span> Type</span> 'describe [<setting>.<prop>]' for detailed property description.</span></span> </span> You</span> may edit the following settings: connection, 802-11-wireless</span> (wifi), 802-11-wireless-security (</span>wifi-sec</span>), 802-1x, ethtool, match, ipv4, ipv6, hostname, link, tc, proxy</span></span> nmcli</span>></span> set 802-1x.eap peap</span></span> nmcli</span>></span> set 802-1x.phase2-auth mschapv2</span></span> nmcli</span>></span> set 802-1x.identity YOUR-IDENTITY</span></span> nmcli</span>></span> set 802-1x.password YOUR-PASSWORD</span></span> nmcli</span>></span> set wifi-sec.key-mgmt wpa-eap</span></span> nmcli</span>></span> save</span></span> Connection</span> 'Internal-CL'</span> (8f1ddcc9-4b1f-4e5d-9992-522714685eb4) successfully updated.</span></span> nmcli</span>></span> activate</span></span> Monitoring</span> connection activation</span> (press</span> any key to continue</span>)</span></span> Connection</span> successfully activated</span> (D-Bus</span> active path: /org/freedesktop/NetworkManager/ActiveConnection/12</span>)</span></span> </span> nmcli</span>></span> quit</span></span></code></pre> Obviously you will need to provide your own values for YOUR-IDENTITY</code> and YOUR-PASSWORD</code> :)</p> `just` OCaml 2025-02-07T00:00:00+00:00 In similar vein to a recent post</a>, I have also started using just</code></a> when I periodically need to rebuild my OCaml</a> tool1</a></sup> ocal</code></a>. So I ended up replacing the old Makefile</code></a> with a shiny new Justfile</code></a>.</p> As it also proved useful in another (more esoteric) tool I wrote for parsing out exam results for my students so I can paste into email easily</a>, I thought I’d put it here for the record. So here it is…</p> Usual preamble of course:</p> _</span>default</span>:</span></span> @</span>just --list</span></span></code></pre> Then set some common variables:</p> PWD</span> :=</span> env</span>(</span>"PWD"</span>)</span></span> DOCDIR</span> :=</span> "_build/default/_doc/_html"</span></span> BUILDDIR</span> :=</span> "_build/install/default/bin"</span></span></code></pre> Then set the target — the tool name, in this case ocal</code> (so named as this is an OCaml re-implementation of a tool approximating the trad Unix cal</code></a> tool):</p> TARGET</span> :=</span> "ocal"</span></span></code></pre> Now for the actually useful stuff: some targets. Mostly these just call out to dune</code> but in a way I find more intuitive.</p> # build targets</span></span> build</span>:</span></span> dune build @all</span></span> </span> # cleanup</span></span> clean</span>:</span></span> dune </span>clean</span></span> </span> # uninstall targets</span></span> uninstall</span>:</span></span> dune uninstall</span></span> </span> # run any tests</span></span> test</span>:</span></span> dune runtest</span></span> </span> # format sources</span></span> format</span>:</span></span> dune fmt</span></span></code></pre> Some compound calls next.</p> First, before building we might need to install dependencies, so do so in the time-honoured fashion:</p> # install dependencies</span></span> depends</span>:</span></span> opam install --yes dune-release odoc</span></span> opam install --yes . --deps-only</span></span></code></pre> Next, to install I first build ready to install, then symlink the resulting binary into the right place in my home directory:</p> # install targets</span></span> install</span>:</span> build</span></span> dune build @install</span></span> ln -sf </span>{{PWD}}</span>/</span>{{BUILDDIR}}</span>/</span>{{TARGET}}</span> ~</span>/</span>.local</span>/</span>bin</span>/</span></span></code></pre> To lint all the things, invoke dune</code> twice:</p> # lint everything</span></span> lint</span>:</span></span> dune build @lint</span></span> dune-release lint</span></span></code></pre> Similarly, to build the docs, build all</em> the docs:</p> # build docs</span></span> doc</span>:</span></span> dune build @doc</span></span> dune build @doc-private</span></span></code></pre> Try to open the docs on Linux and if that fails, on MacOS:</p> # open the docs for reading</span></span> read</span>: doc</span></span> handlr open </span>{{DOCDIR}}</span>/</span>index.html </span>\|\|</span> open </span>{{DOCDIR}}</span></span></code></pre> Finally, tag and create a release; not actually done this in ages so no idea if dune-release</code> invocations are still a thing, let alone correct!</p> # tag and create a release</span></span> release</span>:</span></span> dune-release tag</span></span> dune-release -vv</span></span></code></pre> Largely due to NixOS</a> upgrades moving tools into different locations. ↩</a></p> </li> </ol> </section> Reverse `find` 2025-02-05T00:00:00+00:00 In the last few days I discovered I needed to search back up the filesystem from $CWD</code> to find the first occurence of a file (specifically, a Justfile</code> but that’s by-the-by). Got bored of doing it by hand so wrote a bash</code></a> shell function; here ’tis:</p> rf</span> () {</span></span> local</span> D</span></span> while !</span> eza</span> -l</span> "</span>${</span>D</span>:=</span>.</span>}</span>/</span>$</span>1"</span>;</span> do</span> # first, check `$CWD`</span></span> [</span> "$(</span>realpath</span> "</span>$D</span>/</span>$</span>1")"</span> ==</span> "/</span>$</span>1"</span> ] &&</span> break</span> # stop if we hit `/` already</span></span> D</span>=</span>$D</span>/..</span> # else, iterate one layer up</span></span> done</span></span> }</span></span></code></pre> Invoke as (e.g.,) rf Justfile</code>. Alternatively, as a one-liner:</p> F</span>=</span>Justfile</span>;</span> while !</span> eza</span> -l</span> ${</span>D</span>:=</span>.</span>}</span>/</span>$F</span>;</span> do</span> [</span> "$(</span>realpath $D</span>/</span>$F</span>)"</span> ==</span> "/</span>$F</span>"</span> ] &&</span> break</span>;</span> D</span>=</span>$D</span>/..</span>;</span> done</span>;</span> unset</span> D</span></span></code></pre> PhD viva, UK style 2025-01-31T00:00:00+00:00 Having recently, happily, had several PhD students completing in short order, I’ve been approaching external PhD examiners. Occasionally I find myself asking someone who’s not done any / many in the UK previously. As our system – as all such systems! – is a bit different to those in other parts of the world, I’ve written a few notes on a couple of occasions about what to expect. So I figured I might as well publish them.</p> What follows is my impression / understanding based on experience here in the Department of Computer Science & Technology</a>, Cambridge University</a> ca. 2025. Your Mileage May Vary anywhen and anywhere else, including anywhere else in the UK – check local regulations to be sure.</p> In terms of process, the system here is that candidates submit their complete dissertation and then undergo a “viva voce” (oral examination). It would be usual for the viva to take place within 2—3 months of submission. It’s better (IMO, much better) if it can happen in person but we may still be able to arrange to do it online in extremis.</p> There are two examiners:</p> The internal examiner</em> (typically connected to the Department) who ensures the process is followed properly but may not be a deep expert in the specific topic, and</li> The external examiner</em> (from outside the University) who is there as the subject matter expert.</li> </ol> The viva consists of the two examiners asking the candidate questions about their dissertation until they’re satisfied; typically this takes at least 2h and can go longer, though more than 3.5—4h is unusual in my experience.</p> The examiners are each expected to read the dissertation in detail before the viva and each write a short (typically 1—2pp) independent report giving their opinion, outlining any concerns they will have and the resulting topics/questions they will be exploring in the viva, and indicating what their a priori judgement is in terms of recommendation (roughly: pass/pass with corrections/revise & resubmit/no Ph.D. but you can have a Masters/fail).</p> The examiners will then typically meet ~30min or so before the viva to discuss their independent reports and decide on the approach to take in the viva. After the viva they write a joint report (usually shorter than their independent reports; perhaps 0.5pp) outlining what happened in the viva, as well as making a final recommendation and providing (if appropriate) a list of corrections that must be satisfied for the candidate to pass.</p> Finally, the University pays a (risibly small) honorarium to the external examiner for doing the viva plus reasonable expenses.</p> Fontsizing in LaTeX 2025-01-24T00:00:00+00:00 I recently had colleagues hit an issue that I have hit myself in the past, and so I finally decided to figure out a fix.</p> Specifically, when building EPSRC research proposals in LaTeX, getting a complaint that the font size is non compliant – it should be 11pt Arial, but the standard LaTeX options generate something slightly smaller, with Adobe Acrobat and Microsoft tools both reporting a size of 10.45pt or so.</p> One proposed solution was the to add the following in the preamble:</p> \usepackage</span>{</span>anyfontsize</span>}</span></span> \AtBeginDocument</span>{</span>\fontsize</span>{11bp}{13.35bp}</span>\selectfont</span>} </span></span></code></pre> …but that did not work unfortunately.</p> After some poking about and staring at output and searching the interwebs, it appears that this was triggered, at least for me, by the fontspec</code></a> package that was being used to sort out fonts and unicode and so on in conjunction with XeLaTeX</a> as a backend driver.</p> A key piece of debug logic was to add the following text in a document:</p> The quick fox --- </span>\the\fontdimen</span>6</span>\font\relax</span></span></code></pre> …which ensured there was some text and then inserted the font dimesions according to LaTeX. It did indeed produce the output The quick fox — 9.54147pt</code> when it should’ve been 10pt</code>.</p> The font runes I was using were</p> \usepackage</span>[T1]{</span>fontenc</span>}</span></span> \usepackage</span>{</span>lmodern</span>}</span></span> \usepackage</span>{</span>amssymb,amsmath</span>}</span></span> \usepackage</span>{</span>eurosym</span>}</span></span> \usepackage</span>{</span>upquote</span>}</span></span> \usepackage</span>{</span>microtype</span>}</span></span> \usepackage</span>{</span>fontspec</span>}</span></span> \usepackage</span>{</span>xltxtra,xunicode</span>}</span></span> \defaultfontfeatures</span>{Mapping=tex-text,Scale=MatchUppercase}</span></span> \renewcommand</span>{</span>\familydefault</span>}{</span>\rmdefault</span>}</span></span> \setmainfont</span>{Arial}</span></span> \setmonofont</span>{Hack Nerd Font}</span></span></code></pre> …and it seemed to be the Scale=MatchUppercase</code> clause that caused the problem. Further investigation suggested that most of that was actually copypasta legacy code that was no longer required; replacing with</p> \usepackage</span>{</span>fontspec</span>}</span></span> \setmainfont</span>{Arial}</span></span> \setmonofont</span>{Hack Nerd Font}</span></span></code></pre> …gave the output The quick fox — 10.0pt</code> in Arial as expected. Which was nice.</p> I also realised in the course of doing this that xelatex</code> is now deprecated as a backend, so I have started using the still actively developed luatex</code></a> backend driver instead by passing -lualatex</code> to latexmk</code></a> and that has worked fine so far</p> `just` LaTeX it! 2025-01-18T00:00:00+00:00 I have recently become a fan of just</code></a> as a replacement for the venerable make</code></a>. I find that nowadays I rarely need the built-in dependency rules that make</code></a> provides. Perhaps more radically, I also rarely need to write my own as the prevelance of format-specific build tools such as cargo</code></a>, uv</code></a>, latexmk</code></a> and the like mean I don’t need to write my own either.</p> Recently, while writing references and then helping out get submissions for Rossfest</a> consistently formatted and building cleanly, I found myself extending my various just</code></a> targets for LaTeX</a>. So I thought I’d document them here.</p> I always begin my Justfile</code> with the apparently idiomatic “just show me the targets and associated help text” target:</p> _</span>default</span>:</span></span> @</span>just --list</span></span></code></pre> This seems considerably easier – and more powerful – than the equivalent hack I used to use in a Makefile</code>!</p> .DEFAULT</span>: help</span></span> </span> .PHONY</span>: help</span></span> help</span>:</span></span> @</span>echo "Targets are:"</span></span> @</span>grep -E '^(^## \|[^.#][ a-zA-Z0-9-]+:.#)' Makefile </span>\</span></span> \| sed -E 's/:[[:print:]]+#/:/;s/^([^#])/-- \1/'</span></span></code></pre> Next setup up some useful variables: the command we’ll use (latex</code>) plus sources (texs</code>) and targets (pdfs</code>), and droppings that might be produced but not cleaned up by latexmk -[cC]</code>:</p> latex</span> :=</span> "latexmk -pdf"</span></span> </span> texs</span> :=</span> `</span>echo</span> [0-9][0-9]-</span></span>.tex</span>`</span></span> pdfs</span> :=</span> replace</span>(texs, </span>'.tex'</span>, </span>'.pdf'</span>)</span></span> droppings</span> :=</span> "$f.nav $f.snm $f.bbl"</span></span></code></pre> Now to actually building and cleaning things; first, individual targets:</p> # build a PDF</span></span> pdf</span> tgt</span>:</span></span> {{latex}} {{</span>file_stem</span>(tgt)}}</span>.tex</span></span> </span> # clean generated files</span></span> clean</span> tgt:</span></span> {{latex}}</span> -C </span>{{</span>file_stem</span>(tgt)}}</span>.tex</span></span> for f in </span>{{</span>file_stem</span>(tgt)}}</span>; do rm -f </span>{{droppings}}</span> ; done</span></span></code></pre> (Yes, ok, so it seems a bit silly to have to wrap a for</code> loop around simply to propagate a variable from the Justfile</code> into the shell. But no matter.)</p> Next, all available targets:</p> # build all PDFs</span></span> pdfs</span>:</span></span> for f in </span>{{texs}}</span>; do just pdf $f ; done</span></span> </span> # clean all PDFs</span></span> clean</span>-pdfs:</span></span> for f in </span>{{pdfs}}</span>; do just </span>clean</span> $f ; done</span></span></code></pre> Finally, watch</em> a target, rebuilding on save – it may be helpful therefore to avoid automatically saving the source while in a state in which it will not successfully build!</p> # watch a file, rebuilding when saved</span></span> watch</span> tgt</span>:</span></span> while inotifywait -e close_write </span>{{tgt}}</span> ; do just pdf </span>{{tgt}}</span> ; done</span></span></code></pre> Introducing `bibtox` 2025-01-03T00:00:00+00:00 After some time using various tools and scripts to format and sort my files of BibTeX/BibLaTeX entries, I finally gave up back in March and wrote one myself – bibtox</code></a>. This replaced some very nasty combination of server-side Python</a> and in-page CoffeeScript</a>, plus a third-party tool bib2bib</code></a> I would run on an ad hoc basis.</p> Per the README</a>, this processes either a stream of entries on stdin</code> or a set of files arranged into sections specified by a simple configuration file. It outputs sorted or unsorted, either as canonicalised entries or marked up HTML.</p> And it means that, hopefully, finally, my list of publications on this site is consistently formatted and sorted. Completeness is, of course, a process rather than a state so achieving that is left for other times and places.</p> Issues</a> or pull requests</a> welcome!</p> Retreat from TUM 2024-12-18T00:00:00+00:00 Ok, ok, I exaggerate – it’s not really that far. But any time I have to set the alarm for 02.30, it feels like it’s a long long way away!</p> TU Munchen</a> host an annual retreat</a>, and thanks to Prof. Joerg Ott</a> I was invited this year for the first time. It’s held in TUM’s Retreat Centre at Raitenhaslach</a>, originally a Cistercian monastery which is a lovely location except that it’s a bus ride from the hotel used and a bus and three trains from the nearest airport – which is, mildly confusingly, Salzburg</a> not Munich</a>. The latter point made me assume that there would be good directions via public transport from Salzburg, but that turned out not to be the case. And as Google Maps is, at best, patchy in terms of public transport coverage in this part of the world – it doesn’t know about all the buses at least – I thought it might be useful to record the process of getting there from Cambridge.</p> </a> First, to fly to Salzburg or Munich? In the end I picked Salzburg as it was closer to the retreat centre itself and meant I didn’t need to fly via Heathrow or Gatwick. As I had to be in Cambridge the night before for family reasons, picking a closer airport for my very early flight seemed sensible. Unfortunately this means flying Wizzair UK from London Luton Airport</a>, which after just twice flying from there remains my least favourite UK airport. Given the flight departed at 05.55, a £77.50 taxi from Cambridge was the only option.</p> Per Panther’s recommmendation I allowed 70 minutes for the journey, and on my own recommendation I aimed to arrive ~1.5h before the flight resulting in a taxi booking for 03.15. Happily when I get in the taxi the driver said it would be around 45-50 minutes, which was nice. Less happily when I checked in the day before – using a laptop as neither Safari nor Firefox on iOS was able to display the boarding pass – I got an automatic email from Wizzair telling me that I needed to arrive at the airport by 02.15 as the airport was upgrading their “central search facilities”. After trying to call the airport half a dozen times but getting trapped in the IVR menus at the inevitable “read the FAQ on the website” end state, I gave up trying to check if that really was necessary and decided to risk my original timings.</p> I did however spot one useful thing: “prebooked security check”. This appears to be the ability to book, at 15 minute granularity, a security check during peak hours (03.00-04.30). Given my estimated arrival time of 04.15, that seemed perfect. As it was also completely free</strong> requiring nothing more than my flight number and an email address (not even the one associated with my ticket), I booked it resulting in an email with a QR code to show to security. This worked perfectly when I arrived at the airport: show the person on security the QR code on the phone screen, and they simply jump you past the security queue to the front. Did I mention this was completely free? Seems a strange system to me but hey, I’ll take it!</p> In the end, the public transport worked though there was a short walk at the end that was longer than I expected – rebuilding a road meant I could get through but the bus couldn’t! And then the retreat happened – but you can read about that elsewhere</a> :)</p> Workshopping, across the universe 2024-10-02T00:00:00+00:00 One of the pleasures of being an academic is to travel to nice places to meet interesting people, and interesting places to meet nice people. In one of my first such trips for a few years I recently went to Sweden to participate in the 17th Cloud Control Workshop</a>.</p> </a> I’d previously attended the 15th Cloud Control Workshop</a> shortly before the pandemic happened, causing the series to pause briefly. This was the reboot, and as before, it was a great deal of fun: good company, good food, beautiful location. If you get the chance to go, take it if you can!</p> The workshop is a really nice mix of keynote presentations – not too many and not too long at 20 minutes (mostly) – and discussion sessions proposed beforehand or on site by participants. I gave one of the keynotes, talking about the challenges posed and opportunities offered by edge computing. Can share slides on request, or when I decide on a good way to publish them!</p> I also enjoyed many good discussions and conversations, topped off with a great BBQ, but two topics stand out. First, a great discussion session organised by Prof. Ana Klimović</a> about serverless computing which immediately triggered some thoughts about possible followup publications from EDGELESS</a>, and collaboration possibilities.</p> Second, several conversations on a topic I know little about, low-earth orbit (LEO) satellite networking, and what could be done with it.</p> For what it’s worth, my thought was whether a LEO constellation plus a little edge compute could provide a difficult-to-disrupt out-of-band monitoring network for critical infrastructure like datacenters and power grids: although data bandwidth is limited and a bit complicated due to the constraints on ground stations, they have pretty decent cameras, so why not use those to capture and locally process images of the roofs of large rectangular buildings like datacenters and the like which could continually display various status data.</p> Such status information would only need to be locally generated, so it would take fairly substantial physical disruption of the facility (also presumably noticeable by camera) to prevent that working. Alternatively, the constellation (perhaps shared across multiple facilities in different jurisdictions) would need to be substantially disrupted to prevent it being able to monitor targets.</p> It was a fun discussion anyway, combining a network technology I knew little about with possibly interesting applications of edge computing to resiliency.</p> Postcript: in case you’re curious, the title is a passing reference to Star Trekkin’</a>, a fine popular music single from my youth. Which has a much longer back-story on Wikipedia than I had anticipated.</p> Part II Projects 2024-09-20T00:00:00+00:00 Undergraduate final-year (“Part II”) project supervision goes in fits and starts. After a couple of years of having almost no interest, this year I’ve had several enquiries and it seems I might end supervising 3–4 projects. So herewith a record of the things I’ve found myself repeating!</p> Project structure</a></h2> The key thing for the structure of the project is to make sure that there is a core piece that is (essentially) guaranteed to be deliverable. This is the piece that you know you can do, and once done and written up, you know you can get an adequate (if not great) mark. Ensuring this takes the risk out of the project.</p> On top of this core piece, it’s then usually sensible to build “a few” (2? 3? 4?) extensions which will make the project spicy if done well. You may wish to phrase these extensions as “for example, extensions might include…” or words to that effect, to give some wiggle room in the final dissertation. Getting these done successfully is what should put you in line for a very good mark rather than a simply adequate mark.</p> Project framing</a></h2> It is also very helpful, particularly if you are aiming for a high mark, to try to frame the project in terms of a research question you will answer</strong> rather than an artefact you will build</strong>. Often in systems the appropriate way to answer the research questions we pose will be to build an artefact – but by framing it in terms of the question you seek to answer, it makes it easier to write the dissertation as a piece of research rather than a small matter of programming</a>. Empirically, this seems to have an outsize effect on the chances of the examiners thinking the project difficult and marking accordingly.</p> It can also be useful to try to be explicit about where your project requires you to go beyond the CST taught material, particularly from Part IA/IB.</p> Project proposal</a></h2> To my mind there are two critical pieces of the proposal around which everything else sits.</p> First, the evaluation plan</strong>: if you can write this well, then you understand what you’re going to build, and what it means to have done it well (or badly). Writing the evaluation plan therefore usually means you have, for the most part, got the rest of the project figured out.</p> Second, the workplan</strong>: this divides time until submission into two week (never larger, sometimes smaller) chunks, each of which has attached a calendar date</em> (so there’s no confusion over weeks in term or suchlike) and a milestone</em>/deliverable</em> (so that we can immediately tell whether you’ve completed, or at least are making progress against, that chunk of work in our weekly meeting). Don’t forget to take account of any relevant module assessment deadlines in your plan!</p> Note that it’s a plan not a contract! You don’t lose marks because you deviate from the plan– but if you can’t tell whether you’re ahead or behind, you might well find yourself in a sticky position at the end of Lent term or start of Easter term when you find you’ve got module assessments to complete, revision to start, two weeks to go until dissertation submission and still four weeks of work to do on your project…</p> Supervision process</a></h2> I normally supervise projects by scheduling weekly half-hour meetings with each student. Longer meetings can be arranged on an ad hoc basis as required. The key purpose of the meeting is to check progress against the workplan, and to make sure that any difficulties and roadblocks are aired and dealt with (whether in the meeting or by scheduling a longer discussion).</p> For an example of a reasonable target timeline, consider trying to get implementation completed by the end of Christmas vacation, evaluation completed by the division of Lent Term, and the dissertation completed by the end of Lent Term. That then gives you flexibility as to whether you do more project work, extensions etc., or focus on exam revision, or whatever.</p> Hopefully that’s helpful. At the very least, I can now point potential project students at it, so it’s helpful for me :) Some of the above may also be relevant writing research proposals (Part III / MPhil projects, even Ph.D.s) but that’s a topic for another day.</p> Post-COVID19 TPC experience 2024-09-09T00:00:00+00:00 I do not participate in a huge number of TPCs (Technical Programme Committees) as a general rule– partly time constraints but mostly no-one knows who I am so I don’t often get asked… (!)</p> I have done a few though, some big (e.g., USENIX NSDI</a>, ACM IMC</a>), some small (UK Systems</a>, PAM</a>), and perhaps because I only do a couple every few years, while doing ACM CoNEXT</a> and ACM/IEEE SEC</a> this week, I found myself particularly noticing some changes in practice since that last TPCs I recall (notably NSDI</a> and IMC</a>). So here are three observations.</p> 1. Online, first and only</a></h2> The biggest obvious change is that TPC meetings are now online rather than in-person. This has one big disadvantage for me: I really enjoyed travelling to the meeting to meet colleagues and (usually) participate in some TPC-oriented workshop with presentations of recent an in-progress work. In many ways, I found this sort of activity more interesting than the conference itself (sorry!). It also has the unfortunate effect that, at least for international TPCs, timezones make scheduling tricky – one benefit of travelling was that at least the meeting took place in localtime for (almost) everyone.</p> However, it also has clear benefits: the CO2 footprint of the event is dramatically reduced which can only be a good thing. The financial cost reduction probably also opens up the experience to attendees who would never previously have been able to make it. I’ve rarely seen really poor meeting behaviour on the TPCs I’ve been involved in, but I also find that in an online meeting, chairing tends to be more easily more rigorous and the sometimes dominating effect of a single confident (perhaps I might say over-confident, or even just loud) individual is significantly reduced. Which is good.</p> 2. Offline dominance</a></h2> I now see dramatically</strong> more use being made of commenting and discussions in HotCRP</a>, which remains the only conference management and paper reviewing platform I will willingly use (thanks Eddie</a>!). I don’t know whether this is a post-pandemic effect, or just the fact that the old old guard has largely shuffled out of active TPC duties and we now have a new old guard (ie., people of my era) and younger who are perhaps happier to communicate and express views without needing to be in the same room.</p> I think this is another dramatic improvement in process, and puts us in a similar place to how, for example, we handled marking final year undergraduate projects at Nottingham University</a>. It means that discussion is recorded, and usually more coherently and explicitly argued due to the need to write it down.</p> 3. Reduction in extremism</a></h2> The one arguably slightly negative comment I would have is – and this is only my anecdotal impression, and not something I can pretend to have data on – that I think I see a tendency for reviewers to perhaps be a little less clear in their scoring. Once the across the board strong rejects (a fair number) and strong accepts (a much smaller number) were taken out, I saw an awful lot of weak reject (but happy to consider weak accept after discussion) and weak accept (but happy to consider weak reject after discussion). This made it quite hard for me, as another reviewer, sometimes to get a clear signal as to what the other reviews were recommending.</p> This is a challenge I sometimes see in admissions interviewing: less experienced interviewers are sometimes reluctant to give a clear signal, exhibiting instead a tendency to score near or perhaps just above the middle of the scale. (I remember doing this myself.) This feels more comfortable – putting one’s head above the parapet by taking a clear stance often feels more socially awkward than giving a “fine, ok, pretty good, sure” response, not least because it needs stronger supporting argument – but in the end I think it misses the point of being an interviewer / a TPC member which is exactly to accept or reject papers based on having been recognised for your expertise.</p> So in the end, two good points, two potential (but minor) negatives, and one completely unjustifiable and purely selfish negative. Which on the whole is a good thing. Though having come back to this after a break for a couple of years, I find I am even more sceptical of the whole process than I was. More later perhaps, once I get my thoughts in order.</p> Stop, Start 2024-08-30T00:00:00+00:00 Well, here we go again, again. Having stalled out back in 2013 to restart in 2015, I stalled out again in 2017, so am finally restarting seven years later. Again. I do have a dozen or more half-written posts from that period that may leak out slowly, in backdated form</a>.</p> The stack for the blog has been updated again, naturally – the tools got better after all. So now this is using Zola</a> with, initially, the abridge theme</a>. I think it’s shinier, and hopefully also involves (almost) no JavaScript, is finally accessible, and not too heavyweight. In removing the excessive JavaScript, I also updated my publication list</a> generation process</a>: no more CoffeeScript and JavaScript, just some plain ol’ Python</a> using the slick new uv</a> tooling. The Justfile</a> incantation is something like:</p> papers</span> := "../templates/shortcodes/publications.html"</span></span> bibinputs</span> := "../papers/bibinputs.json"</span></span> </span> # build papers data for site</span></span> @papers</span>:</span></span> cd bib2html && uv sync && uv run -- ./src/bib2html {{bibinputs}} >\| {{papers}}</span></span></code></pre> I’ve also added a register of interests</a> so I stop having to scrabble around remembering what I’m officially doing when I’m occasionally asked.</p> The stack underneath has also changed. I finally kicked the Mac OS habit after over 10 years – decided I’d had enough of Apple hiding things and generally getting in my way – at the same time I switched from Android to iPhone. Never do things the easy way, that’s my motto. So now I’m back on a PC … but it turns out that 2024 (well, 2023 in fact) is the year of Linux on the Laptop for me. Not something I thought would actually happen when we joked about it in the late 90s. Truly, times change. And not just any old wannabe-Windows distribution but the resolutely idiosyncratic and copiously yet poorly documented NixOS</a> using Flakes and Home-Manager of course, with Wayland</a> and Sway</a>. I’ve even finally made the leap to Codium</a>, the de-Microsofted version of VSCode, from Emacs</a> (mostly – org-mode</a> has not yet been replaced). I blame Andrew</a> and Chris</a>.</p> For the terminally curious, this meant a fairly heavy reworking of my 30 years old dotfiles</a> – still work in progress but adequately organised for now I think. See the NixOS specific bits</a> if curious.</p> And, because I’m going to try to include at least one vaguely technical bit (or at least, one bit which isn’t just me bemoaning that I haven’t written anything in years), I confess to being particularly pleased with the following fragment to enable my Sway startup script</a> to start several apps on particular workspaces, by switching to the workspace, starting the app, and then waiting for a suitable window to appear. The latter step entails</p> subscribing swaymsg</code> to window</code> messages with jq</a> to extract the messages of interest indicating a new window in the background,</li> executing the command to start the app,</li> using grep</code>/pkill</code> in tandem to kill the subscription when the new window is detected, and finally</li> waiting for the subscription to have been killed.</li> </ul> It seems fairly (indeed, surprisingly) reliable.</p> wait_for</span> () {</span></span> {</span> swaymsg</span> -r -m -t</span> subscribe '["window"]'</span> \|</span></span> jq</span> -c --unbuffered</span> '. \| select(.change == "new")'</span> \|</span></span> {</span> grep</span> -m1</span> .</span> ></span>/dev/null</span> ;</span> pkill</span> swaymsg</span> ;} &</span></span> }</span> 2></span>/dev/null</span></span> pid</span>=</span>$!</span></span> swaymsg</span> --</span> "exec </span>$</span>"</span> &&</span> sleep</span> 0.5</span></span> wait $pid</span> 2></span>/dev/null</span></span> }</span></span></code></pre> Use via something like:</p> wayland</span>.</span>windowManager</span>.</span>sway</span>.</span>config</span>.</span>startup</span> =</span></span> let</span></span> msg</span> =</span> cmds:</span> "swaymsg '</span>${</span>builtins</span>.</span>concatStringsSep</span> ", "</span> cmds</span>}</span>'"</span>;</span></span> workspace</span> =</span> ws: msg [</span> "workspace --no-auto-back-and-forth </span>${</span>ws</span>}</span>"</span> ]</span>;</span></span> after</span> =</span> delay: cmds:</span> "sleep </span>${</span>toString</span> delay</span>}</span> && </span>${</span>msg cmds</span>}</span>"</span>;</span></span> startup</span> =</span> pkgs</span>.</span>writeShellScriptBin</span> "startup.sh" ''</span></span> wait_for () {</span></span> { swaymsg -r -m -t subscribe '["window"]' \|</span></span> jq -c --unbuffered '. \| select(.change == "new")' \|</span></span> { grep -m1 . >/dev/null ; pkill swaymsg ;} &</span></span> } 2>/dev/null</span></span> pid=$!</span></span> swaymsg -- "exec $" && sleep 0.5</span></span> wait $pid 2>/dev/null</span></span> }</span></span> </span> ${</span>workspace</span> "</span>${</span>mediaws</span>}</span>"</span>}</span></span> wait_for "rhythmbox"</span></span> ...</span></span> ''</span>;</span></span> in</span></span> [ {</span> command</span> =</span> "</span>${</span>startup</span>}</span>/bin/startup.sh"</span>; } ]</span>;</span></span></code></pre> Discovering `restic` 2024-08-26T00:00:00+00:00 I recently had cause to try to recover some files from my restic</code></a> backups. These go back for over a year now, and I could not remember at which point I’d mistakenly nuked the directory I now wanted to recover. restic find</code> purports to be able to do this by searching through snapshots but I found that it’s quite slow, and can only search within a time range which is not that helpful when you don’t know the time range you need.</p> So I did it by hand, which turned out to be rather faster.</p> RESTIC_PASSWORD_FILE</span>=</span>/your/backup/password/file</span> RESTIC_REPOSITORY</span>=</span>/your/backup/repository/</span> \</span></span> # list snapshots, filtering by DATE regex, grabbing just the snapshot hash</span></span> sudo</span> -E</span> restic snapshots</span> -c \</span></span> \|</span> rg</span> DATE</span> \</span></span> \|</span> cut</span> -b1-8 \</span></span> \| while</span> read</span> ss</span>;</span> do</span> </span></span> echo</span> "=== </span>$ss</span>"</span></span> sudo</span> -E</span> restic ls</span> $ss</span> \|</span> rg</span> /DIRECTORY/</span> ;</span> done</span> </span></span> done</span></span></code></pre> The key thing with the above approach is that it’s also quite amenable to bisection, which makes it a lot faster.</p> Debugging Electron apps 2024-08-14T00:00:00+00:00 Another short one, this time the magic flags to get debug output and access to webdev tools under Electron apps such as Slack and whatnot on Wayland:</p> APPLICATION</span> --enable-logging --devtools \</span></span> --platform=wayland --enable-features=UseOzonePlatform --password-store=</span>"gnome_libsecret"</span></span></code></pre> Reinstalling `maestral` 2024-06-28T00:00:00+00:00 A short one this; just a crib of commands to redo maestral</code> (unofficial Dropbox</a> client) configuration.</p> # remove existing configuration</span></span> rm</span> ~/.config/maestral/maestral.ini</span></span> # appears to fail with `/usr/bin/env 'bash': No such file or directory`</span></span> maestral</span> autostart</span> &&</span> systemctl</span> --user</span> daemon-reload</span> </span></span> # ...so instead, just remmeber to do the following on restart</span></span> maestral</span> stop</span> &&</span> maestral</span> start</span> </span></span> # if desired, add autocompletion for `maestral` in bash</span></span> [[ $(</span>which</span> maestral</span> 2></span>/dev/null</span>) ]] &&</span> source</span> <(</span>maestral</span> completion bash)</span></span></code></pre> NixOS channels 2023-12-13T00:00:00+00:00 I don’t pretend to understand NixOS</a> configuration fully yet, what with Flakes and channels and so forth. But I did find the following usful to setup channels consistently so that I could have a single config that used both.</p> sudo</span> nix-channel</span> --list</span> # to list known channels</span></span> sudo</span> nix-channel</span> --remove</span> nixos</span> # to remove a channel</span></span> sudo</span> nix-channel</span> --add</span> https://nixos.org/channels/nixos-unstable nixos</span> # to go bleeding edge</span></span></code></pre> …ultimately resulting in:</p> $</span> nix-channel</span> --list</span> # channels are stored per-user, and $(whoami) != root</span></span> $</span> sudo nix-channel</span> --list</span></span> nixos</span> https://nixos.org/channels/nixos-unstable</span></span> nixpkgs</span> https://nixos.org/channels/nixos-unstable</span></span></code></pre> Upgrading to the latest release is then something like:</p> pushd </span>~</span>/rc-files/nixos/</span></span> sudo</span> nix-channel</span> --update</span></span> nix</span> flake update</span></span> sudo</span> nixos-rebuild switch</span> --upgrade-all</span></span> popd</span></span></code></pre> …and finally, garbage collecting old versions once you’re satisfied the new one works:</p> sudo</span> nix-collect-garbage</span> -d</span></span> nix-collect-garbage</span> -d</span></span> sudo</span> nix-store</span> --gc</span></span> nix-store</span> --gc</span> </span></span></code></pre> And yes, some of the incantations above might be a little cargo-cultish and not strictly necessary. But at various points they’ve seemed necessary to me, and now they’re in my shell history, they’re what I got.</p> OneDrive and NixOS 2023-08-22T00:00:00+00:00 Starting by reading instructions at:</p> https://github.com/abraunegg/onedrive/blob/master/docs/sharepoint-libraries.md</a></li> https://github.com/abraunegg/onedrive/blob/master/docs/business-shared-items.md</a></li> https://github.com/NixOS/nixpkgs/pull/77734#issuecomment-575874225</a></li> </ul> FWIW I also use that package on a NixOS system (via a Nix package that can presumably be installed on other systems if you add nix as a package manager), https://github.com/NixOS/nixpkgs/blob/master/nixos/modules/services/networking/onedrive.nix</a>.</p> I have successfully used it with the University and my startup’s tenancies (both personal OneDrive and enterprise Sharepoint sites) simultaneously – this was fairly simple in the end but I recall it took a while to do the token dance to get the necessary tokens, and then to get all the runes right. In short it was something like:</p> Follow the instructions Robert pointed to sort out access to the various accounts, refresh tokens, and the like.</p> </li> Create file ~/.config/onedrive-launcher</code> comprising each subdirectory of ~/.config</code> that is a onedrive configuration directory corresponding to each account – I named these after the accounts in question so mine contained</p> </li> </ol> onedrive-rmm1002@cam.ac.uk</span></span> onedrive-mort@ikva.ai</span></span> sharepoint-mort@ikva.ai-iKVALimited</span></span></code></pre> (I now prefix the last two lines with # </code> to comment them out as I don’t need those synced any more.)</p> The systemd service onedrive-launcher.service</code> then uses the file to kick off a systemd onedrive@...</code> service for each entry.</p> Edit the ~/.config/ACCOUNT/config</code> files appropriately; the only changes I made were to</li> </ol> # for my University account</span></span> sync_dir = "~/OneDrive/rmm1002@cam.ac.uk"</span></span> </span> # for my startup personal OneDrive</span></span> sync_dir = "~/OneDrive/mort@ikva.ai"</span></span> sync_business_shared_folders = "true"</span></span> </span></span> # for startup Sharepoint sites</span></span> sync_dir = "~/OneDrive/mort@ikva.ai-iKVA_Limited"</span></span> drive_id = "..." # rune found per instructions Robert pointed to I think</span></span></code></pre> WSL2 / Windows 11 2023-03-13T00:00:00+00:00 I naively thought I could just use WSL2 on Windows on my new laptop. But it turned out this was the year of Linux on the Laptop for me. For posterity here’s the crib sheet though.</p> wsl</span> --set-default-version</span> 2</span></span> sudo</span> apt update</span> &&</span> sudo</span> apt upgrade</span> -yy</span></span> </span> sudo</span> apt install locales</span></span> sudo</span> locale-gen en_GB.UTF-8</span></span> </span> sudo</span> apt install</span> -yy</span> emacs-gtk direnv gedit</span></span> git</span> clone ./..rc-files</span></span> ./scripts/install.sh</span></span></code></pre> Some Windows native packages using Chocolatey</a></p> choco</span> install signal skype wire slack zoom</span></span> choco</span> install git</span></span> choco</span> install rustup.install rust-analyzer python</span></span> choco</span> install docker</span></span> choco</span> install powertyos dropbox googledrive wiztree</span></span></code></pre> However, choco install texlive</code> didn’t work so well, so I fell back to WSL2: sudo apt install latexmk texlive-latex-base texlive-xetex ttf-mscorefonts-installer</code>.</p> Use MS Powertoys to remap keyboard for CAPSLOCK</code>, €</code>.</p> Timesync is a bit broken, cf https://stackoverflow.com/a/72318510</a>.</p> Unfortunately it all went pear-shaped when I tried to rsync</code> files across from MacOS into Windows/WSL2.</p> I mapped the network drive via Network and Sharing Center > Settings > Network & internet > Advanced network settings > Advanced sharing settings > Public networks > Network discovery = ON</p> …and then</p> </span> for</span> d</span> in</span> admin christs docs rc-files research src teaching me</span> ;</span> do</span></span> echo</span> "=== </span>$d</span>"</span></span> rsync</span> -uavzsP --log-file=</span>$d</span>.$(</span>date</span> -Iseconds) -e</span> ssh mort@IPADDRESS:/Users/mort/</span>$d</span>/ ./</span>$d</span></span> done</span></span></code></pre> …but found that moving files to the host mashed things a bit (rw</code> bits carried, but x</code> not; hidden files not; no content translation; owner carried) while moving host files to wsl was sort-of ok (owner carried, rw bits carried) except that group</code> and other</code> access bits are all set to whatever the user</code> access bits were.</p> See https://stackoverflow.com/questions/41513597/how-do-i-access-the-wsl-linux-file-system-from-windows</a> for more, perhaps.</p> Getting Docker installed was also rather painful:</p> # remove old distribution dockers</span></span> sudo</span> apt remove docker.io containerd runc</span> &&</span> sudo</span> apt autoremove</span></span> </span> # install dependencies to use an alternative package repo</span></span> sudo</span> apt-get update</span> &&</span> sudo</span> apt-get install ca-certificates curl gnupg lsb-release</span></span> </span> # install the new package repo</span></span> sudo</span> mkdir</span> -m</span> 0755</span> -p</span> /etc/apt/keyrings</span></span> curl</span> -fsSL</span> https://download.docker.com/linux/ubuntu/gpg</span> \|</span> sudo</span> gpg</span> --dearmor -o</span> /etc/apt/keyrings/docker.gpg</span></span> echo</span> "deb [arch=$(</span>dpkg</span> --print-architecture</span>) signed-by=/etc/apt/keyrings/docker.gpg] https://download.docker.com/linux/ubuntu $(</span>lsb_release</span> -cs</span>) stable"</span> \|</span> sudo</span> tee /etc/apt/sources.list.d/docker.list</span> ></span> /dev/null</span></span> </span> # install up-to-date Docker</span></span> sudo</span> apt-get update</span> &&</span> sudo</span> apt-get install</span> \</span></span> docker-ce</span> \</span></span> docker-ce-cli</span> \</span></span> containerd.io</span> \</span></span> docker-buildx-plugin</span> \</span></span> docker-compose-plugin</span></span></code></pre> Finally, some further references that may or may not be useful:</p> https://stephenreescarter.net/how-to-shrink-a-wsl2-virtual-disk/</a></li> https://www.linkedin.com/pulse/acceso-wsl2-desde-windows-con-samba-manuel-nicol%C3%A1s-ortu%C3%B1o/</a></li> https://www.howtogeek.com/193013/how-to-create-an-encrypted-container-file-with-bitlocker-on-windows/</a></li> </ul> OCaml operators 2023-02-26T00:00:00+00:00 An OCaml operator cheatsheet</a> for OCaml</a> operators that I have found useful.</p> Software folklore 2023-02-26T00:00:00+00:00 Truly cursed bugs that have become software folklore</a>.</p> Hosting HotCRP 2021-10-17T00:00:00+00:00 I once had cause to setup HotCRP</a> for local hosting. Specifically on a local Lab-hosted VM image. Some of what follows is specific to the CUCL VM hosting service, but I think most of it is HotCRP generic and so may be of use. Anyway, here’s the crib sheet, starting from https://mbtech.github.io/Setting-up-hotcrp/</a>…</p> # setup some variables</span></span> YOUR-DOMAIN</span>=</span>"hotcrp-test.cl.cam.ac.uk"</span></span> YOUR-WORKSHOP</span>=</span>"sysws18"</span></span> YOUR-PASSWORD</span>=</span>"mybestpassword"</span></span> YOUR-EMAIL</span>=</span>"postmaster@example.com"</span></span></code></pre>Lab specifics</a></h2> Assume we start from a default Ubuntu template VM, and then…</p> Configure the VM</li> </ol> cl-asuser</span> passwd</span> # set UNIX password for sudo</span></span> </span> # create some space</span></span> for</span> d</span> in</span> /usr/src/</span> ;</span> do</span></span> export</span> K</span>=</span>$(</span>uname</span> -r</span> \|</span> sed</span> 's/-generic$//'</span>)</span></span> echo</span> -n</span> $K $d</span> ...</span></span> case</span> $d</span> in</span></span> "/usr/src/linux-headers-</span>$K</span>"</span> \|</span> "/usr/src/linux-headers-</span>${</span>K</span>}</span>-generic"</span> )</span></span> echo</span> keep</span></span> ;;</span></span> )</span></span> echo</span> remove</span></span> sudo</span> rm</span> -rf</span> $d</span></span> ;;</span></span> esac</span></span> done</span></span> </span> # THIS IS UNSAFE! BE CAREFUL! IT CALLS `sudo rm -rf`!</span></span> for</span> d</span> in</span> /lib/modules/</span> ;</span> do</span></span> echo $d</span> ...</span></span> case</span> $d</span> in</span></span> "/lib/modules/$(</span>uname</span> -r</span>)"</span> )</span></span> echo</span> keep</span></span> ;;</span></span> )</span></span> echo</span> remove</span></span> sudo</span> rm</span> -rf</span> $d</span></span> ;;</span></span> esac</span></span> done</span></span> </span> # if necessary, resize the partition. this shouldn't be necessary with the new</span></span> # VM image! if you need more than ~1GB space for papers, setup xvdb1</span></span> sudo</span> fdisk /dev/xvda</span> <<</span>EOF</span></span> p</span></span> d</span></span> n</span></span> </span> </span> </span> </span> w</span></span> EOF</span></span> sudo</span> partprobe</span></span> </span> # resize the default filesystem to use the entire partition</span></span> sudo</span> resize2fs /dev/xvda1</span> # blank SIZE means use whole partition</span></span></code></pre> Install packages</li> </ol> # sort out packages</span></span> export</span> TZ</span>=</span>Europe</span>/</span>London</span></span> sudo</span> apt update</span> &&</span> sudo</span> apt install</span> --no-install-recommends -qq -yy \</span></span> apache2</span> \</span></span> ca-certificates</span> \</span></span> git</span> \</span></span> libapache2-mod-php</span> \</span></span> mailutils</span> \</span></span> mysql-server</span></span> php-curl</span> \</span></span> php-json</span> \</span></span> php-mysql</span> \</span></span> poppler-utils</span> \</span></span> postfix</span> \</span></span> zip</span></span></code></pre> Configure postfix</code></li> </ol> # configure postfix: accept defaults if offered, setup postfix to use ppsw</span></span> sudo</span> sed</span> -i</span> 's/relayhost =/relayhost = ppsw.cam.ac.uk/' /etc/postfix/main.cf</span></span> sudo</span> /etc/init.d/postfix reload</span></span> sudo</span> systemctl restart postfix.service</span></span> # test mail sending</span></span> echo</span> "Test mail from postfix"</span> \|</span> mail</span> -s</span> "Test Postfix"</span> ${</span>YOUR-EMAIL</span>}</span></span></code></pre> For more email help, see https://help.uis.cam.ac.uk/email-telephony-and-collaboration/email/specialist-email-advice/sending-email</a> using YOUR-DOMAIN</code> as mail domain, and ppsw.cam.ac.uk</code> as relay host.</p> Install HotCRP</li> </ol> Get latest release:</p> git</span> clone https://github.com/kohler/hotcrp.git</span></span> cd</span> hotcrp</span></span> git</span> checkout tags/v2.101</span> -b</span> v2.101</span></span></code></pre> Setup root</code> account for MySQL</li> </ol> sudo</span> /etc/init.d/mysql stop</span> # stop the running service</span></span> </span> # configure and run mysql in the console</span></span> sudo</span> mkdir</span> -p</span> /var/run/mysqld</span></span> sudo</span> chown mysql:mysql /var/run/mysqld</span></span> sudo</span> mysqld_safe</span> --skip-grant-tables</span> &</span> sleep</span> 5</span></span> </span> # smash a new `root` password in place </span></span> sudo</span> mysql</span></span> ALTER</span> USER 'root'@'localhost' IDENTIFIED WITH mysql_native_password BY '${YOUR-PASSWORD}'</span>; </span></span> FLUSH</span> PRIVILEGES</span>;</span></span> exit</span>;</span></span> </span> # restart mysql properly as a service</span></span> mysqladmin</span> -uroot -p</span>${</span>YOUR-PASSWORD</span>}</span> -h127.0.0.1 --protocol=tcp</span> shutdown</span></span> sudo</span> /etc/init.d/mysql start</span></span></code></pre> …alternatively</p> mysql</span> -uroot</span><<</span>_EOF</span></span> USE mysql;</span></span> </span> UPDATE mysql.user SET authentication_string = PASSWORD('</span>${</span>YOUR-PASSWORD</span>}</span>')</span></span> WHERE User = 'root' AND Host = 'localhost';</span></span> FLUSH PRIVILEGES;</span></span> </span> QUIT</span></span> _EOF</span></span></code></pre> Secure your MySQL installation</li> </ol> sudo</span> systemctl stop mysql</span></span> sudo</span> mkdir</span> -p</span> /var/run/mysqld</span></span> sudo</span> chown mysql:mysql /var/run/mysqld</span></span> sudo</span> mysqld_safe</span> --skip-grant-tables --skip-networking</span> &</span></span> sudo</span> mysql_secure_installation</span> -p</span>${</span>YOUR-PASSWORD</span>}<<</span>EOF</span></span> n</span></span> n</span></span> y</span></span> y</span></span> y</span></span> y</span></span> EOF</span></span></code></pre> Setup the HotCRP MySQL tables and config</li> </ol> lib/createdb.sh</span> --user=root --password=</span>${</span>YOUR-PASSWORD</span>} <<</span>EOF</span></span> ok</span></span> YOUR-WORKSHOP</span></span> EOF</span></span> </span> # edit conf/options.php</span></span> # - contactName</span></span> # - contactEmail</span></span> # - sendEmail</span></span> # - emailFrom</span></span> # - emailSender</span></span> # - timezone</span></span> # - upload_max_filesize [ if you care ]</span></span></code></pre> Turn on the HotCRP site in your Apache configuration</li> </ol> # apache2: turn on hotcrp site</span></span> sudo</span> sh</span> -c</span> 'cat >>/etc/apache2/conf-available/hotcrp.conf <<_EOF</span></span> <Directory "$(pwd -P)"></span></span> Options Indexes Includes FollowSymLinks</span></span> AllowOverride all</span></span> Require all granted</span></span> </Directory></span></span> Alias /YOUR-WORKSHOP $(pwd -P)</span></span> _EOF</span></span> '</span></span> </span> sudo</span> a2enconf</span> <<</span>EOF</span></span> hotcrp</span></span> EOF</span></span> </span> sudo</span> chgrp www-data conf/options.php</span></span> sudo</span> service apache2 reload</span></span> sudo</span> apache2ctl graceful</span></span></code></pre> …and you should now be able to access your hotcrp site at http://${YOUR-DOMAIN}/${YOUR-WORKSHOP}</a></p> Use Let’s Encrypt</a> to create and configure certificates for HTTPS support</li> </ol> sudo</span> apt install</span> -yy</span> software-properties-common</span></span> sudo</span> add-apt-repository ppa:certbot/certbot</span></span> sudo</span> apt update</span></span> sudo</span> apt install</span> -yy</span> certbot-auto</span></span> wget</span> https://dl.eff.org/certbot-auto</span></span> chmod</span> a+x ./certbot-auto</span></span> sudo</span> ./certbot-auto</span> -n --os-packages-only</span></span> </span> sudo</span> ./certbot-auto</span> -a</span> webroot</span> -i</span> apache</span> -w</span> $(</span>pwd</span> -P</span>)</span> \</span></span> --agree-tos --redirect --uir --hsts --staple-ocsp \</span></span> -d</span> YOUR-DOMAIN</span> --email</span> YOUR-EMAIL</span></span> </span> sudo</span> ./certbot-auto</span> --install-only</span></span></code></pre> Set permissions on the certificate directories</li> </ol> sudo</span> chgrp www-data /etc/letsencrypt/live</span></span> sudo</span> chmod g+rx /etc/letsencrypt/live</span></span> sudo</span> chgrp www-data /etc/letsencrypt/archive/</span></span> sudo</span> chmod g+rx /etc/letsencrypt/archive/</span></span></code></pre> End state is the Apache config looks something like the following, with unindented lines being those I added:</p> $</span> cat /etc/apache2/sites-available/hotcrp.conf</span></span> <</span>IfModule mod_ssl.c</span>></span></span> </span> SSLStaplingCache</span> shmcb:/var/run/apache2/stapling_cache</span>(</span>128000</span>)</span></span> </span> <</span>VirtualHost</span> _default_:44</span>3></span></span> ServerAdmin</span> webmaster@localhost</span></span> DocumentRoot</span> /home/hotcrp/hotcrp</span></span> ErrorLog</span> ${</span>APACHE_LOG_DIR</span>}</span>/error.log</span></span> CustomLog</span> ${</span>APACHE_LOG_DIR</span>}</span>/access.log combined</span></span> SSLEngine</span> on</span></span> </span> SSLCACertificateFile</span> /etc/letsencrypt/live/hotcrp.sysws.org.uk/fullchain.pem</span></span> SSLUseStapling</span> on</span></span> </span> <</span>FilesMatch</span> "\.(cgi\|shtml\|phtml\|php)$"</span>></span></span> SSLOptions</span> +StdEnvVars</span></span> <</span>/FilesMatch</span>></span></span> <</span>Directory</span> /usr/lib/cgi-bi</span>n</span>></span></span> SSLOptions</span> +StdEnvVars</span></span> <</span>/Directory</span>></span></span> </span> ServerName</span> hotcrp.sysws.org.uk</span></span> SSLCertificateFile</span> /etc/letsencrypt/live/hotcrp.sysws.org.uk-0001/fullchain.pem</span></span> SSLCertificateKeyFile</span> /etc/letsencrypt/live/hotcrp.sysws.org.uk-0001/privkey.pem</span></span> Include</span> /etc/letsencrypt/options-ssl-apache.conf</span></span> Header</span> always set Strict-Transport-Security "max-age=31536000"</span></span> Header</span> always set Content-Security-Policy upgrade-insecure-requests</span></span> </span> <</span>/VirtualHost</span>></span></span> <</span>/IfModule</span>></span></span> </span> 11.</span> Add DNS entry for the name assigned</span> (in</span> my case, `</span>hotcrp.DOMAIN</span>`</span>).</span></span></code></pre> Tarballing some files 2021-08-25T00:00:00+00:00 I recently discovered, to some irritation, that the --include PATTERN</code> option to tar</code> seems only to apply to directories – and if the PATTERN</code> doesn’t match, it won’t traverse subdirectories. But I wanted to include .php</code> for some reason. So instead pipe the output of find</code>, or better these days, fd</code></a>:</p> fd</span> -e</span> php</span> -0</span> \|</span> tar</span> -cvjf</span> TARBALL.bz2</span> --null --files-from</span> -</span></span></code></pre> Data visualisation links 2020-09-07T00:00:00+00:00 Some possibly useful data visualisation links:</p> https://medium.economist.com/why-you-sometimes-need-to-break-the-rules-in-data-viz-4d8ece284919</a></li> https://colororacle.org/</a></li> </ul> Renovating an old media PC 2020-08-28T00:00:00+00:00 Some notes from my first attempt to renovate an old media PC that had a SYSLINUX install without any package management, and a crufty BIOS. Probably outdated now, but I may go back to it one day…</p> First, some background links:</p> https://en.wikipedia.org/wiki/Cylinder-head-sector</a></li> https://en.wikipedia.org/wiki/FAT_boot_sector</a></li> https://en.wikipedia.org/wiki/Logical_Block_Addressing#CHS_conversion</a></li> https://en.wikipedia.org/wiki/Master_Boot_Record</a></li> https://en.wikipedia.org/wiki/Volume_boot_record</a></li> https://wiki.archlinux.org/index.php/Syslinux</a></li> https://wiki.syslinux.org/wiki/index.php?title=Common_Problems#Failed_to_load_ldlinux</a></li> https://wiki.syslinux.org/wiki/index.php?title=Hardware_Compatibility#USB_related_problems</a></li> https://wiki.syslinux.org/wiki/index.php?title=Hdt_(Hardware_Detection_Tool)</a></li> </ul> I explored two ways forward: SYSLINUX</a> and GRUB</a>.</p> SYSLINUX</a></h2> I found that getting SYSLINUX working required moving the partition to 0/1/1 – using sectors per track of 63 or 32, and heads per cylinder or 16 or 64 with appropriate cylinder values simply didn’t help.</p> Diagnosed by observing that console displayed only CRLF but no banner – SYSLINUX code ends up with the banner to be displayed just falling into the second sector on the disk, so it can’t be read unless the geometry is correct. Don’t ask why old fashioned whirling metal disk geometry needs to be set for a USB stick, you’ll be sad.</p> Formatting the USB stick</a></h3> Some runes, use at your own risk.</p> sudo</span> dd if=/dev/zero of=/dev/sdd status=progress bs=1M count=</span>256</span></span> sudo</span> fdisk /dev/sdd</span> <<</span>EOF</span></span> o</span></span> x</span></span> h</span></span> 64</span></span> s</span></span> 32</span></span> r</span></span> n</span></span> p</span></span> 1</span></span> </span> </span> t</span></span> 6</span></span> a</span></span> 1</span></span> w</span></span> </span> EOF</span></span> sudo</span> mkfs.fat /dev/sdd1</span></span></code></pre> One exciting gotcha: the fdisk</code> utility in the util-linux</code> package didn’t work</strong> – but the one in busybox</code> did!</p> Putting MBR in place</a></h3> sudo</span> dd bs=</span>440</span> count=</span>1</span> conv=notrunc if=/usr/share/syslinux/mbr.bin of=/dev/sdd</span></span></code></pre>Obtaining and installing memtest86</code></a></a></h3> cd</span></span> wget</span> http://memtest.org/download/4.10/memtest86+-4.10.zip</span></span> unzip</span> memtest86+-4.10.zip</span></span> sudo</span> cp ~/memtest86+-4.10.bin /mnt/boot/</span></span></code></pre>Putting locally built SYSLINUX in place</a></h3> sudo</span> mount /dev/sdd1 /mnt</span></span> sudo</span> mkdir</span> -p</span> /mnt/boot/syslinux</span></span> sudo</span> syslinux</span> --directory</span> boot/syslinux</span> --install</span> /dev/sdd1</span></span> sudo</span> cp /usr/share/syslinux/</span></span>.c32 /mnt/boot/syslinux/</span></span> cd</span> ~/syslinux</span></span> make</span> bios</span></span> sudo</span> cp ~/syslinux/bios/com32/hdt/hdt.c32 /mnt/boot/syslinux/</span></span> sudo</span> cp /usr/share/hwdata/pci.ids /mnt/boot/syslinux</span></span></code></pre>sudo</span> sh</span> -c</span> "cat > /mnt/boot/syslinux/syslinux.cfg"</span> <<</span>EOF</span></span> # UI menu.c32</span></span> PROMPT 1</span></span> DEFAULT hdt</span></span> </span> LABEL some_label</span></span> LINUX memdisk</span></span> INITRD ../alpine-standard-3.12.0-x86_64.iso</span></span> APPEND iso-scan/filename=../alpine-standard-3.12.0-x86_64.iso</span></span> </span> LABEL memtest</span></span> LINUX ../memtest86+-4.10.bin</span></span> </span> LABEL hdt</span></span> COM32 hdt.c32</span></span> </span> LABEL reboot</span></span> COM32 reboot.c32</span></span> </span> LABEL poweroff</span></span> COM32 poweroff.c32</span></span> EOF</span></span></code></pre> Unfortunately, getting hdt</code> working required rebuilding as the Alpine package version doesn’t appear to statically link against libupload.a from SYSLINUX tree so doesn’t work. Fixing required make bios</code> in the SYSLINUX tree after installing dependencies including:</p> sudo</span> apk</span> -U</span> add nasm xzlinux-headers util-linux-dev</span></span></code></pre>GRUB</a></h2> Similar behaviour: GRUB2 displayed the GRUB </code> message and nothing else, The Ubuntu wiki</a> says this is the “can’t find MBR or euqivalent” information. In fact, it’s the same issue: subsequent progress requires reading the second sector, but I had a CHS/LBA mismatch meant it wasn’t reading from the right sector and so hanging.</p> To wipe the stick</a></h3> sudo</span> dd if=/dev/zero of=/dev/sdd status=progress bs=4M</span></span></code></pre>To partition the stick</a></h3> In this case, to be bootable with a single ext4</code> partition</p> sudo</span> parted /dev/sdd</span></span> mklabel</span> msdos</span></span> unit</span> s</span></span> mkpart</span> primary ext2 2048s 100%</span></span> set</span> 1</span> boot on</span></span> set</span> 1</span> lba off</span></span></code></pre> …or alternatively, possibly</p> sudo</span> fdisk /dev/sdd</span> <<</span>EOF</span></span> o</span></span> n</span></span> p</span></span> 1</span></span> </span> </span> a</span></span> 1</span></span> </span> EOF</span></span></code></pre>To format the partition, and install grub</code> and the master boot record</a></h3> sudo</span> mkfs.ext4 /dev/sdd1</span></span> sudo</span> mount /dev/sdd1 /mnt</span></span> sudo</span> grub-install</span> --recheck --boot-directory=/mnt/boot</span> /dev/sdd</span></span></code></pre> At this point, booting off the stick will bring htpc to GRUB </code> error stage, indicating GRUB has loaded but doesn’t know anything about how to continue.</p> Install memtest</a></h3> sudo</span> cp memtest86+.bin /mnt</span></span> sudo</span> cat</span> ></span>/mnt/boot/grub/grub.cfg</span> <<</span>EOF</span></span> set timeout=10</span></span> set default=0</span></span> </span> menuentry "Memtest 86+" {</span></span> linux16 /memtest86+.bin</span></span> }</span></span> EOF</span></span></code></pre>Install Alpine ISO for booting</a></h3> Add the following stanza to GRUB config, above:</p> insmod</span> loopback</span></span> </span> menuentry</span> "alpine" {</span></span> set</span> isofile=/boot/alpine-standard-3.12.0-x86_64.iso</span></span> loopback</span> loop</span> $isofile</span></span> linux</span> (loop)/boot/vmlinuz-lts iso-scan/</span>filename</span>=</span>$isofile modules</span>=</span>loop,squashfs,sd-mod,usb-storage</span> modloop</span>=</span>(</span>loop</span>)</span>/boot/modloop-lts</span></span> initrd</span> (loop)/boot/initramfs-lts</span></span> }</span></span></code></pre>Miscellaneous cribs</a></h2> I did of course forget that the</p> Bluetooth keyboard requires a dongle to be plugged in which is stored inside the battery compartment. Doh. Making a note so I don’t forget again.</p> Adding internal hard-disk</a></h3> Required partitioning and formatting:</p> find where disk is mounted: sudo lshw -C disk -short</code> assume new disk is /dev/sdX</code>, flash disk is /dev/sdF</code></li> partition disk: sudo parted /dev/sdX mkpart primary ext4 1 -1</code></li> format disk: sudo mkfs.ext4 -m0 /dev/sdX1</code></li> label disk: sudo e2label /dev/sdX1 Harddisk</code></li> remount flash disk rw: mount -o remount,rw /dev/sdF</code></li> edit /boot/extlinux.conf</code> so APPEND line reads: APPEND boot=LABEL=System disk=LABEL=Harddisk quiet</code></li> </ul> I screwed up the first time by not correctly labelling the disk so had to make an Ubuntu rescue USB stick. Couldn’t get this to work using MacOS, though didn’t try putting GRUB on via MacOS.</p> download ISO: http://ubuntu-rescue-remix.org/</a></li> boot ubuntu-rescue-remix-12-04.iso</code> via virtualbox</li> mount USB stick on /dev/sdX</code> at /mnt</code>: mount /dev/sdX /mnt</code></li> format the stick: mkfs.vfat -n multiboot /dev/sdX1</code></li> cd /mnt && mkdir boot iso</code></li> grub-install --force --no-floppy --boot-directory=/mnt/boot /dev/sdX</code></li> create ISO from mounted cd: dd if=/dev/cdrom of=/mnt/iso/ubuntu-rescue-remix-12-04.iso</code></li> create /boot/grub/grub.cfg</code> with</li> </ul> menuentry</span> 'Ubuntu Rescue Remix ISO ' {</span></span> set</span> isofile="/iso/ubuntu-rescue-remix-12-04.iso"</span></span> loopback</span> loop</span> (hd0,N)</span>$isofile</span></span> linux</span> (loop)/casper/vmlinuz</span> boot</span>=</span>casper</span> iso-scan/filename</span>=</span>$isofile</span> noprompt noeject</span></span> initrd</span> (loop)/casper/initrd.gz</span></span> }</span></span></code></pre> where N</code> is partition number, typically 1.</p> Finally, for backup purposes, addons are stored in /storage/.xbmc/addons/packages</code>, and the following Alpine packages were useful to install for some of the above, diagnostics, etc:</p> sudo</span> apk add busybox-static apk-tools-static</span></span> sudo</span> vi /etc/apk/repositories</span></span> sudo</span> apk.static update</span></span> sudo</span> apk.static upgrade</span> --no-self-upgrade --available</span></span> sudo</span> apk add lshw lshw-doc</span></span> sudo</span> lshw</span> -C</span> storage</span> -short -numeric</span></span> sudo</span> apk add lsblk</span></span> sudo</span> lsblk</span></span></code></pre> Discord discord 2020-08-01T00:00:00+00:00 So for some reason I wanted to do this – use Discord on an iPad without installing the app. This proved surprisingly tricky as Safari on the iPad really</em> wanted to make you use the app and certainly wouldn’t display the desktop site.</p> However, Firefox Focus can though it still forced you into the app from the link in the invite email.</p> However, you can go to the website to signup and then manually invite yourself via a real desktop browser. And this then means that Firefox Focus on the iPad believes in the new account and just lets you in.</p> Seems something of a palaver but hey, these twisted webs we weave.</p> Setting up a Macbook 2020-01-01T00:00:00+00:00 This is the set of things that I roughly did to set up my old new Macbook: configurations, packages, etc. Not guaranteed complete, but hopefully captures much of it.</p> Keyboard</a></h2> Set the touchbar to expanded</li> Turn off various emacs conflicts (e.g., S-M-6</code>)</li> Use Karabiner Elements</a> to make CAPSLOCK</code> become ESC</code> on a single press</li> </ul> Homebrew packages</a></h2> Install Homebrew</a> and then</li> </ul> # install packages</span></span> brew</span> install</span> \</span></span> aspcud</span> \</span></span> aspell</span> \</span></span> bash-completion</span> \</span></span> coreutils</span> \</span></span> direnv</span> \</span></span> emacs</span> \</span></span> evernote</span> \</span></span> ffmpeg</span> \</span></span> font-hack-nerd-font</span> \</span></span> gawk</span> \</span></span> gcc</span> \</span></span> get_iplayer</span> \</span></span> ghostscript</span> \</span></span> git</span> \</span></span> git-lfs</span> \</span></span> gnupg2</span> \</span></span> gnuplot</span> \</span></span> gpg-agent</span> \</span></span> graphviz</span> \</span></span> imagemagick</span> \</span></span> jq</span> \</span></span> lua</span> \</span></span> mu</span> \</span></span> ncftp</span> \</span></span> nmap</span> \</span></span> ocaml</span> \</span></span> offlineimap</span> \</span></span> omnigraffle</span> \</span></span> opam</span> \</span></span> python</span> \</span></span> python@2</span> \</span></span> qemu</span> \</span></span> rcs</span> \</span></span> readline</span> \</span></span> rsync</span> \</span></span> socat</span> \</span></span> sshfs</span> \</span></span> telnet</span> \</span></span> tmux</span> \</span></span> unrar</span> \</span></span> wget</span></span></code></pre># install casks</span></span> brew</span> tap homebrew/cask-versions</span></span> brew</span> cask install</span> \</span></span> adium</span> \</span></span> disk-inventory-x</span> \</span></span> docker</span> \</span></span> dropbox</span> \</span></span> emacs</span> \</span></span> evernote</span> \</span></span> font-hack-nerd-font</span> \</span></span> get-iplayer-automator</span> \</span></span> google-backup-and-sync</span> \</span></span> google-drive-file-stream</span> \</span></span> gpodder</span> \</span></span> inkscape</span> \</span></span> iterm2</span> \</span></span> karabiner-elements</span> \</span></span> keepingyouawake</span> \</span></span> keybase</span> \</span></span> mactex</span> \</span></span> mp3tag</span> \</span></span> musicbrainz-picard</span> \</span></span> omnigraffle6</span> \</span></span> onedrive</span> \</span></span> osxfuse</span> \</span></span> sharepod</span> \</span></span> signal</span> \</span></span> slack</span> \</span></span> vlc</span></span></code></pre>git</span> lfs install</span> --system</span> # enable `git-lfs` system-wide</span></span> mu</span> index</span> --rebuild</span> # rebuild `mu` index</span></span> opam</span> init</span> # initialise `opam`</span></span></code></pre>Podcasts</a></h2> Extract RSS feeds from iTunes:</li> </ul> grep</span> Location Library.xml</span> \|</span> sed</span> 's/[[:space:]]//g'</span> \|</span> sort</span> \|</span> \</span></span> grep</span> -iv</span> "\.$mp[34]\\|m4[av]$"</span> \|</span> grep</span> http</span> \|</span> cut</span> -b</span> 28-</span> \|</span> \</span></span> cut</span> -f</span> 1</span> -d</span>"<"</span> ></span> PODCAST.URLS</span></span></code></pre>OCal</a></a></h2> opam</span> install</span> -y</span> ocal</span></span></code></pre> Falsehoods 2019-01-11T00:00:00+00:00 Being a list of some cool “Falsehoods programmers believe about …” sites, now redundant thanks to https://github.com/kdeldycke/awesome-falsehood</a>:</p> …time</a></li> …time, more</a></li> …names</a></li> …addresses</a></li> …geography</a></li> </ul> Wipeout 2018-12-17T00:00:00+00:00 I recently decided to go through some old PCs and hard disks (yes, actual spinning bits of metal) and recycle or simply junk them. I figured I should wipe them properly first, and given that they had been installed with a range of OSs, setup a bootable USB stick so that I could boot and wipe in one easy motion.</p> Download Alpine Linux</a> 3.8.1 ISO</a> – I had only x86_64</code> machines, YMMV obviously</li> Write this to a USB stick using dd</code> (on Linux), or Balena Etcher</a> on Windows</li> Insert the stick and boot the machine after making any necessary BIOS changes</li> If the machine has been unbooted for too long or (as one of mine) has a flat CMOS clock battery so cannot retain time across reboots, set the time manually: date --set="20181217"</code></li> automatically: setup-alpine</code> to start configuring things, then CTRL-C</code> after network setup, and execute setup-ntp</code></li> </ul> </li> Then setup-alpine; apk add coreutils</code></li> Finally, shred --verbose /dev/sdXN</code> where X</code> is the device id and N</code> the partition number (eg., /dev/sda2</code>), or shred --verbose -n1 /dev/sdXN</code> if you’re using a modern disk (apparently) and only want one pass of random data</li> </ul> Installing Alpine</a></h2> After installing Alpine as above:</p> adduser mort</code></li> create ~mort/.ssh/authorized_keys</code> containing you preferred public key, and set permissions (chmod 600 ~mort/.ssh/authorized_keys</code>)</li> apk add sudo</code> and then visudo</code> to allow members of group sudo</code> to sudo</code></li> add mort</code> to group sudo</code></li> logout and then back in</li> </ul> You can then configure storage as you see fit; it seems I once did probably using sudo</code>:</p> apk</span> add lvm2 git bash xfsprogs</span></span> pvcreate</span> /dev/sd[bc]</span> # create persistent volumes group</span></span> vgextend</span> vg0 /dev/sdb</span> # assign storage devices to volume group</span></span> vgextend</span> vg0 /dev/sdc</span> #</span></span> lvcreate</span> --name</span> lv_home</span> --size</span> 60G vg0</span> # create logical volume `lv_home`</span></span> mkfs.xfs</span> /dev/vg0/lv_home</span> # format `lv_home` using XFS</span></span> lvcreate</span> --name</span> lv_backup</span> --size</span> 60G vg0</span> # create logical volume `lv_backup`</span></span> mkfs.xfs</span> /dev/vg0/lv_backup</span> # format `lv_backup` using XFS</span></span> </span> # create `/etc/fstab` to match the above configuration</span></span> cat</span> >></span>/etc/fstab</span> <<</span>EOF</span></span> /dev/vg0/lv_home /home xfs defaults 0 0</span></span> /dev/vg0/lv_backup /backup xfs defaults 0 0</span></span> EOF</span></span> mount</span> -a -v</span> # mount everything, verbosely</span></span></code></pre> Back to the Future 2017-12-14T00:00:00+00:00 Ed: the first in a likely series, this is one of the posts I drafted but never published during my seven years in the blogging wilderness</p> </blockquote> I have been having persistent issues on MacOS with Time Machine backups interacting badly with several configuration elements and other services. This is exacerbated by my use of maildirs to backup emails, with my Gmail account leading to a directory with some millions of files. (Yeah, ok, maybe not wise but hey, we are where we are.)</p> So in the course of trying to get backups to work reliably, I found the following commands useful to at least some degree:</p> mdutil -sav</code> to get info about spotlight indexing status</li> sudo mdutil -Ea -i [off\|on]</code> to turn off\|on spotlight indexing where possible</li> sudo fs_usage -w mdworker mds_stores backupd</code> obtain the current filesystem usage for those services</li> sudo fs_usage -w -e iTerm2</code> to obtain current filesystem usage for services excluding</em> those listed</li> sudo sysctl debug.lowpri_throttle_enabled=[0\|1]</code> stop\|start throttling to go faster\|be nice!</li> to obtain the various different hostnames that MacOS seems to want to give; note that renaming ComputerName</code> (also possible via SysPrefs>Sharing>>name</code>) caused the Time Machine directory to be renamed when the backup started</li> </ul> for</span> n</span> in</span> ComputerName LocalHostName HostName</span> ;</span> do</span> </span></span> scutil</span> --get</span> $n</span></span> done</span></span></code></pre> tmutil listbackups</code> to list backups</li> tmutil listlocalsnapshotdates /</code> to list local snapshots</li> tmutil destinationinfo</code> to list the Time Machine destination volumes</li> to stream Time Machine log entries live</li> </ul> log</span> stream</span> --style</span> syslog</span> --debug --info \</span></span> --predicate</span> 'senderImagePath contains[cd] "TimeMachine"'</span></span></code></pre> to show Time Machine log entries from YYYY-MM-DD</li> </ul> log</span> show</span> --style</span> syslog</span> --debug --info --start</span> YYYY-MM-DD</span> \</span></span> --predicate</span> 'senderImagePath contains[cd] "TimeMachine"'</span>\</span></span></code></pre> Quelle dommage 2017-11-05T00:00:00+00:00 Ed: this tool is perhaps less relevant now that both Mirage</a> and OPAM</a> have moved on. But perhaps it’ll be resurrected one day so here it is.</p> </blockquote> Largely because I wanted to make a feeble attempt at a French pun, dommage</code></a> is a tool for Docker</a> containerising Mirage unikernels. From the README</a>:</p> Dommage, Dockerised Mirage</a></h2> dommage</code> is a shell script that wraps the Mirage</a> CLI to make use of Docker containers meaning that:</p> you can cache the OPAM build artefacts in the container image, speeding up local builds;</li> you can re-use the build container image in Travis builds by publishing it, speeding those up considerably; and</li> you can easily test build -t xen</code> targets on OSX.</li> </ul> I’ve tried to minimise interference with the normal operation of Mirage</a> CLI so simply replacing mirage</code> with dommage</code> is supposed to work. To publish the resulting container image, dommage publish <image></code>.</p> Issues, comments, suggestions and bug fixes all welcome!</p> Operation</a></h3> To start, dommage</code> provides a few management commands to manipulate the build container:</p> dommage init BASE-IMAGE</code> creates a new container, based off BASE-IMAGE</code> from the Docker Hub</a></li> dommage publish IMAGE</code> commits the current container and pushes it to Docker Hub</a> as IMAGE</code></li> dommage destroy</code> stops and removes the current build container</li> dommage run ...</code> executes a command inside the current build container</li> </ul> In addition, it wraps the main Mirage</a> CLI commands:</p> dommage configure ...</code> runs mirage configure ... && make depends</code> inside the build contianer</li> dommage build ...</code> runs mirage build ...</code> inside the build container</li> dommage clean ...</code> runs mirage clean ...</code> inside the build container</li> </ul> Spring Loading Karabiner in the Autumn 2017-10-30T00:00:00+00:00 I recently took the plunge and upgraded my OS X. Not to vN of Sierra</em> as I’d hoped, but to v0 High Sierra</em>– the perils of waiting too long…</p> Unfortunately, this toasted1</a></sup> my carefully curated keyboard remappings as Karabiner</a> used a kernel extension, for which everything changed. All was not lost however, as the rewrite to support Sierra/High Sierra was well underway. Or so I thought until I realised that the configuration file had changed from XML to JSON. And so my configuration journey began. (But it all ends well, so that’s good.)</p> Controlling the config</a></h2> The first thing was to get the new configuration matters under control. I did this per the documentation, symlinking the config subdirectory from my rc-files</code> repo:</p> cd</span> ~/.config/</span></span> mv</span> karabiner/ ~/rc-files/</span></span> ln</span> -s</span> ~/rc-files/karabiner</span></span></code></pre>Internal Apple keyboard</a></h2> In the interests of keeping all configuration in one place (but see below), I decided to do this via a set of complex modifications</a>. In summary this meant:</p> swap (caps_lock)</code> and (control)</code>:</li> </ul> {</span></span> "description"</span>:</span> "mort: caps_lock -> ctrl"</span>,</span></span> "manipulators"</span>: [</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "caps_lock"</span>,</span></span> "modifiers"</span>: {</span>"optional"</span>: [</span>"any"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "left_control"</span></span> }</span></span> ]</span></span> }</span></span> ]</span></span> },</span></span></code></pre> swap "</code> (glyph S-'</code>) with @</code> (glyph S-2</code>):</li> </ul> {</span></span> "description"</span>:</span> "mort: S-' (</span>\"</span>) <-> S-2 (@)"</span>,</span></span> "manipulators"</span>: [</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "quote"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "2"</span>,</span></span> "modifiers"</span>: [</span>"shift"</span>]}</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "2"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "quote"</span>,</span></span> "modifiers"</span>: [</span>"shift"</span>]</span></span> }</span></span> ]</span></span> }</span></span> ]</span></span> },</span></span></code></pre> map (backslash)</code> (glyph \</code>) to #</code>, and S-\</code> (glyph \|</code>) to ~</code>:</li> </ul> {</span></span> "description"</span>:</span> "mort: </span>\\</span> -> #; S-</span>\\</span> (\|) -> ~"</span>,</span></span> "manipulators"</span>: [</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "backslash"</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "3"</span>,</span></span> "modifiers"</span>: [</span>"option"</span>]</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "backslash"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span>,</span></span> "modifiers"</span>: [</span>"shift"</span>]</span></span> }</span></span> ]</span></span> },</span></span> ...</span></span> }</span></span></code></pre> map (non_us_backslash)</code> (glyph §</code>) to `</code> and S-(non_us_backslash)</code> (glyph ±</code>) to €</code>, and then patch things up so that the usual window switching works (using (command)-`</code>):</li> </ul> {</span></span> "description"</span>:</span> "mort: § -> `; ± (S-§) -> €"</span>,</span></span> "manipulators"</span>: [</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "2"</span>,</span></span> "modifiers"</span>: [</span>"option"</span>]</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"command"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span>,</span></span> "modifiers"</span>: [</span>"command"</span>]</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"command"</span>,</span> "shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span>,</span></span> "modifiers"</span>: [</span>"command"</span>,</span> "shift"</span>]</span></span> }</span></span> ]</span></span> }</span></span> ]</span></span> },</span></span></code></pre> finally, map `</code> to \</code> and S-`</code> to \|</code></li> </ul> {</span></span> "description"</span>:</span> "mort: ` -> </span>\\</span>; S-` (~) -> \|"</span>,</span></span> "manipulators"</span>: [</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "backslash"</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "backslash"</span>,</span></span> "modifiers"</span>: [</span>"shift"</span>]</span></span> }</span></span> ]</span></span> }</span></span> ]</span></span> },</span></span></code></pre>iTerm2</a></h2> Unfortunately for me, iTerm2 then gets a bit confused as it wants to leave (command)</code> alone, only allowing mapping of (option)</code> to (meta)</code> (or, in fact, (esc+)</code>). In the past I swapped (left_command)</code> and (left_option)</code> to make the usual shell (bash</code>) CLI editing combinations (roughly, emacs</code>) work. That wasn’t ideal though as I then had to fix up the window cycling commands ((command)-` </code> and so on). Fortunately, the fix this time seems easier: just configure the two tricky mappings (involving generating a keypress modified with (option)</code>) to be interpreted by iTerm2 to just send the appropriate text through. Again, I did this in the UI (Preferences > Profiles > Keys) but the resulting configuration change is also straightforward:</p> <</span>key</span>></span>Keyboard Map</span></</span>key</span>></span></span> <</span>dict</span>></span></span> ...</span></span> <</span>key</span>></span>0x32-0x80000</span></</span>key</span>></span></span> <</span>dict</span>></span></span> <</span>key</span>></span>Action</span></</span>key</span>></span></span> <</span>integer</span>></span>12</span></</span>integer</span>></span></span> <</span>key</span>></span>Text</span></</span>key</span>></span></span> <</span>string</span>></span>€</span></</span>string</span>></span></span> </</span>dict</span>></span></span> ...</span></span> <</span>key</span>></span>0x33-0x80000</span></</span>key</span>></span></span> <</span>dict</span>></span></span> <</span>key</span>></span>Action</span></</span>key</span>></span></span> <</span>integer</span>></span>12</span></</span>integer</span>></span></span> <</span>key</span>></span>Text</span></</span>key</span>></span></span> <</span>string</span>></span>#</span></</span>string</span>></span></span> </</span>dict</span>></span></span> ...</span></span> </</span>dict</span>></span></span></code></pre>Microsoft Digital Media keyboard</a></h2> Examining the key codes using the Karabiner Event-Viewer, it seemed that the first thing to do was to swap (grave_accent_and_tilde)</code> (glyph `</code>) and (non_us_backslash)</code> (slightly confusingly, glyph \</code> on my keyboard). I started out trying to do this as a complex modification so that all the remappings were in one file</a>, but couldn’t: I couldn’t figure out how to control the application order of mappings in that file. However, simple modifications are applied before complex modifications, and this is</em> a simple modification as it’s a direct swap, so I just used the UI and did it there. For the sake of completeness, the resulting modification to karabiner.json</code></a> is:</p> {</span></span> ...</span></span> "profiles"</span>: [</span></span> {</span></span> ...</span></span> "devices"</span>: [</span></span> {</span></span> "disable_built_in_keyboard_if_exists"</span>:</span> false</span>,</span></span> "fn_function_keys"</span>: [],</span></span> "identifiers"</span>: {</span></span> "is_keyboard"</span>:</span> true</span>,</span></span> "is_pointing_device"</span>:</span> false</span>,</span></span> "product_id"</span>:</span> 180</span>,</span></span> "vendor_id"</span>:</span> 1118</span></span> },</span></span> "ignore"</span>:</span> false</span>,</span></span> "simple_modifications"</span>: [</span></span> {</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span></span> },</span></span> "to"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span></span> }</span></span> },</span></span> {</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_backslash"</span></span> },</span></span> "to"</span>: {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span></span> }</span></span> }</span></span> ]</span></span> }</span></span> ],</span></span> ...</span></span> }</span></span> ]</span></span> }</span></span></code></pre> The next step was to patch up the complex modifications. Once I realised that the event viewer was claiming that the key with glyph #</code> was emitting (backslash)</code> while it was, in fact, emitting (non_us_pound)</code>, this was fairly straightforward:</p> swap (command)</code> (glyph Alt</code>) and (option)</code> (glyph Start</code>):</li> </ul> {</span></span> "conditions"</span>: [</span></span> {</span></span> "type"</span>:</span> "device_if"</span>,</span></span> "identifiers"</span>: [{</span>"vendor_id"</span>:</span> 1118</span>,</span> "product_id"</span>:</span> 180</span>}]</span></span> }</span></span> ],</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "left_option"</span>,</span></span> "modifiers"</span>: {</span>"optional"</span>: [</span>"any"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "left_command"</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "conditions"</span>: [</span></span> {</span></span> "type"</span>:</span> "device_if"</span>,</span></span> "identifiers"</span>: [{</span>"vendor_id"</span>:</span> 1118</span>,</span> "product_id"</span>:</span> 180</span>}]</span></span> }</span></span> ],</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "left_command"</span>,</span></span> "modifiers"</span>: {</span>"optional"</span>: [</span>"any"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "left_option"</span></span> }</span></span> ]</span></span> }</span></span> ]</span></span> },</span></span></code></pre> add coverage of (non_us_pound)</code> to the rule that remaps \</code> to #</code>:</li> </ul> {</span></span> "conditions"</span>: [</span></span> {</span></span> "type"</span>:</span> "device_if"</span>,</span></span> "identifiers"</span>: [{</span>"vendor_id"</span>:</span> 1118</span>,</span> "product_id"</span>:</span> 180</span>}]</span></span> }</span></span> ],</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_pound"</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "3"</span>,</span></span> "modifiers"</span>: [</span>"option"</span>]</span></span> }</span></span> ]</span></span> },</span></span> {</span></span> "conditions"</span>: [</span></span> {</span></span> "type"</span>:</span> "device_if"</span>,</span></span> "identifiers"</span>: [{</span>"vendor_id"</span>:</span> 1118</span>,</span> "product_id"</span>:</span> 180</span>}]</span></span> }</span></span> ],</span></span> "type"</span>:</span> "basic"</span>,</span></span> "from"</span>: {</span></span> "key_code"</span>:</span> "non_us_pound"</span>,</span></span> "modifiers"</span>: {</span>"mandatory"</span>: [</span>"shift"</span>]}</span></span> },</span></span> "to"</span>: [</span></span> {</span></span> "key_code"</span>:</span> "grave_accent_and_tilde"</span>,</span></span> "modifiers"</span>: [</span>"shift"</span>]</span></span> }</span></span> ]</span></span> }</span></span></code></pre> …and that’s it. My keyboard is, once again, my castle.</p> To be honest, I suspect even the Sierra</em> upgrade would’ve done this. ↩</a></p> </li> </ol> </section> Platforms, Packaging, Progress— Addendum 2017-09-01T00:00:00+00:00 This is a short addendum to my post of a couple of days ago</a> caused by my carelessness in writing the OPAM file</a>. Careful readers will observe the lack of any dependency on notty</a>. Read on for what happened next…</p> The result of this carelessness was that everything worked just fine locally, but my PR to the OPAM package repository</a> failed. Cue much wailing and gnashing of teeth.</p> However, thanks to a moment’s assistance from Daniel Bünzli</a>, this was easy to fix:</p> $</span> git checkout</span> 0.2.0</span> # checkout the relevant release version tag</span></span> $</span> topkg opam pkg</span> # create the release metadata</span></span> $</span> e _build/ocal.0.2.0/opam</span> # invoke editor so I can add the missing dep</span></span> $</span> topkg opam submit</span> # submit the updated OPAM metadata, updating the PR</span></span> Submitting</span> _build/ocal.0.2.0</span></span> [ocal-0.2.0.tbz] http://github.com/mor1/ocal/releases/download/0.2.0/ocal-0.2.0.tbz downloaded</span></span> Updating</span> existing pull-request</span> #10176</span></span> Pull-requested:</span> https://github.com/ocaml/opam-repository/pull/10176</span></span></code></pre> For me, the main thing to note here is that the OPAM metadata in the repo at the commit ref tagged for release doesn’t match that which OPAM uses to install the release. But as Sebastien Mondet</a> pointed out to me, this is neither relevant nor (in the long term) likely, as (e.g.) version constraints on dependencies may need to be added to old versions of dependent packages to keep them working. (Though I did add and commit the dependency to master</code>, naturally.)</p> So, all-in-all, an easy fix to a common problem. Which is the way it should be…</p> Platforms, Packaging, Progress 2017-08-28T00:00:00+00:00 I recently decided to refresh and update my ocal</a> package,1</a></sup> primarily to port it to use the excellent notty</a> before adding support for indicating week-of-year. At the same time, I took the opportunity to update the build infrastructure now that the OCaml world has some shiny new packaging and build tools to go with OPAM</a>, namely topkg</code></a> and jbuilder</code></a>. So, starting from Dave Scott’s</a> wiki entry</a> about how to package Mirage</a> libraries, here’s what I had to do…</p> Remove Oasis remnants</a></h2> git</span> rm _oasis setup.ml Makefile</span></span> _tags myocamlbuild.ml .merlin</span></span> mv</span> ocal.opam/opam o</span> &&</span> git</span> rm</span> -rf</span> ocal.opam</span> &&</span> mv</span> o ocal.opam</span> &&</span> git</span> add ocal.opam</span></span> cat</span> >\|</span> .gitignore</span> <<</span>_EOF</span></span> _build</span></span> .merlin</span></span> .install</span></span> _EOF</span></span></code></pre> Although we’re removing the ocal.opam/descr</code> file, we’re not going to lose the content: we’re going to let topkg opam pkg</code> use its default --readme</code> option to extract the relevant info from the first marked up section of the README.md</code></a>:</p> # ocal — An improved Unix </span>`cal`</span> utility</span></span> </span> %%VERSION%%</span></span> </span> A replacement for the standard Unix </span>`cal`</span> utility. Partly because I could,</span></span> partly because I'd become too irritated with its command line interface.</span></span></code></pre> We also remove but don’t lose the functionality of the .merlin</code> and OPAM ocal.install</code> files, as jbuilder</a> will generate them for us.</p> Create src/jbuild</code> file</a></h2> cat</span> >\|</span> src/jbuild</span> <<</span>_EOF</span></span> (jbuild_version 1)</span></span> </span> (executable</span></span> ((public_name ocal)</span></span> (package ocal)</span></span> (name main)</span></span> </span> (libraries</span></span> (</span></span> astring</span></span> calendar</span></span> cmdliner</span></span> notty</span></span> notty.unix</span></span> ))</span></span> (flags (:standard -w "A-44-48-52" -safe-string))</span></span> ))</span></span> _EOF</span></span></code></pre> This corresponds to the 0.2.0</a> release of ocal</a>. Note that the name</code> parameter refers to the module that contains the entrypoint for the executable, and that we turn on all warnings (A</code>) except for three that we wish to ignore:</p> 44</code>: Open statement shadows an already defined identifier.</li> 48</code>: Implicit elimination of optional arguments.</li> 52</code>: (see 8.5.1) Fragile constant pattern.</li> </ul> After I did some tidying up of the code to deal with the newly imposed warnings, make</code> and make install</code> satisfactorily (and quickly!) used jbuilder</a> to build and install the executable as ~/.opam/system/bin/ocal</code> (thanks to the public_name</code> stanza in the src/jbuild</code> file, above). make uninstall</code> then caused jbuilder</a> to remove it, before I opam</code> pinned it and then reinstall through opam</code> to check that workflow worked as well:</p> opam</span> remove ocal</span></span> opam</span> pin add</span> -yn --dev-repo</span> ocal .</span></span> opam</span> install ocal</span></span></code></pre>Create the topkg</code> skeletons</a></h2> Having refreshed the basic build infrastructure, next it’s time to update the packaging workflow. For a simple library we could use the automatic jbuilder</a>/topkg</a> plugin per the wiki entry</a>:</p> mkdir</span> pkg</span></span> cat</span> >\|</span> pkg/pkg.ml</span> <<</span>_EOF</span></span> #!/usr/bin/env ocaml</span></span> #use "topfind"</span></span> #require "topkg-jbuilder.auto"</span></span> _EOF</span></span></code></pre> However, this isn’t a library so we don’t have documentation to build so we don’t bother with the odoc</code> skeleton. As a result we also need to customise pkg/pkg.ml</code></a> so as to stop topkg publish</code> failing when it can’t build docs:</p> #!/usr/bin/env ocaml</span></span> #</span>use </span>"topfind"</span></span> #require </span>"topkg-jbuilder"</span></span> </span> open</span> Topkg</span></span> </span> let</span> publish</span> =</span></span> Pkg</span>.</span>publish</span> ~</span>artefacts</span>:</span>[</span>`</span>Distrib] ()</span></span> </span> let</span> ()</span> =</span></span> Topkg_jbuilder</span>.</span>describe</span> ~</span>publish</span> ()</span></span></code></pre>Prepare a release</a></h2> Finally, we follow the standard topkg</a> workflow to prepare a release. First, add an entry to CHANGES.md</code></a> with the correct formatting and commit the result, and then:</p> distrib</span>:</span></span> [ -x </span>$$</span>(opam config var root)/plugins/opam-publish/repos/ocal ] \|\| </span>\</span></span> opam-publish repo add ocal mor1/ocal</span></span> topkg tag</span></span> topkg distrib</span></span></code></pre> …which creates tokens for accessing the GitHub repo for this project (if they don’t already exist), creates a release tag based on entries in CHANGES.md</code></a>, and then creates the release tarballs (without the edits to pkg/pkg.ml</code></a> this would also build the docs, but we have none).</p> Publish a release</a></h2> Finally, we publish the release to GitHub and issue a pull request to the OPAM repository</a> to add the new release into OPAM after linting and tests have passed.</p> publish</span>:</span></span> topkg publish</span></span> topkg opam pkg</span></span> topkg opam submit</span></span></code></pre> Given that this repo has only a single package, we could in fact simply issue</p> topkg tag && topkg bistro</span></span></code></pre> Also, as an alternative to customising the pkg/pkg.ml</code></a> as indicated above, we could simply remember to indicate the appropriate customisation on the command line:</p> topkg publish distrib</span></span></code></pre> …but topkg bistro</code> wouldn’t then work.</p> Conclusion</a></h2> So that’s it: a simple executable distribution taken from old-school Oasis</a> and OCamlBuild</a> infrastructure to shiny new modern jbuilder</a> and topkg</a>. The new scheme seems to me to be an improvement: faster build times, simpler (to my eyes) metadata, autogeneration of more of the repeated metadata (.merlin</code> etc), and a reasonably simple Makefile</code></a> that I actually think I understand. Definitely progress :)</p> A somewhat over-featured replacement for the standard UNIX cal</code> utility, because I got irritated by its American-centricity and my initial Python replacement</a> was just too slow… ↩</a></p> </li> </ol> </section> Redwood Trail, Big Basin Redwood Park, USA 2017-08-24T00:00:00+00:00 A short stop off to see some big trees: a small redwood grove in California.… </p> May Lake, Yosemite National Park, USA 2017-08-22T00:00:00+00:00 A short walk to May Lake in Yosemite National Park, USA. The day after the eclipse. Genuinely beautiful. Some of the photos I’m most happy with in this lot.… </p> ARMing LinuxKit 2017-06-02T00:00:00+00:00 As some may know, following the Unikernel Systems</a> acquisition, I currently do contract work for Docker Inc.</a> in addition to my day job here at the Cambridge University Computer Laboratory</a>. Recently this has centred on LinuxKit</a>, “A toolkit for building secure, portable and lean operating systems for containers</em>” and, specifically, enabling ARM64 support. I’m pleased to say that a basic proof-of-concept is now complete, and we’re working towards getting support merged upstream.</p> The proof-of-concept was developed using the great ARM64 support provided by packet.net</code></a>, on one of their type 2A</code> boxes.</p> If you fancy trying it out, then hopefully the following instructions will be of use – or just bug me on the packet.net</code> Slack</a>!</p> Building</a></h2> Start by getting an ARM64 box setup. If you have one to hand, great! If not, you could head over to packet.net</a> and create type 2A Ubuntu box to use as a build environment.</p> Then clone the source, either git clone</code> my dev branch</a>, or see https://github.com/linuxkit/linuxkit/pull/1654</a> for the open PR which may be a bit more stable.</p> The essence of it then is to build the containers based off aarch64/alpine</code>, along with an ARM64 version of the moby</code> CLI</a> if needed. Specifying the container images you just built in your moby.yml</code> file will then cause moby</code> to assemble things that should boot on ARM64.</p> The output should be a gzipped kernel, currently slightly misleadingly named bzImage</code> as well as a suitable initrd</code>.</p> Booting</a></h2> Setup another ARM64 box on which to boot the results. You could setup a type 2A packet.net</a> box once more, but this time set it to custom OS</em> and iPXE boot</em>. For the iPXE boot URL, give a URL pointing to a suitable boot file. I use:</p> #!ipxe</span></span> set</span> base-url URL-TO-DIRECTORY-HOLDING-IMAGES</span></span> set</span> kernel-params ip=dhcp nomodeset ro serial console=ttyAMA0,115200 earlycon earlyprintk=serial,keep initrd=arm64-initrd.img</span></span> initrd</span> ${</span>base-url</span>}</span>/arm64-initrd.img</span></span> imgstat</span></span> boot</span> ${</span>base-url</span>}</span>/arm64-bzImage</span> ${</span>kernel-params</span>}</span></span></code></pre> Note that, currently at least, the packet.net</a> iPXE boot only occurs on the first boot as it is assumed that the iPXE boot will install a working image to the local disk. Thus, if it doesn’t work first time, get an SOS console and break in by hitting ^B</code> at the appropriate moment, before issuing chain URL</code> where URL</code> points to your iPXE boot file.</p> Conclusion</a></h2> This just does the barest minimum for now – I did say it was a proof-of-concept… :) Work is currently ongoing to upstream this rather than developing this PoC further, but if anyone has a particular interest or would like to provide patches to, e.g., support network devices on packet.net</a>, please get in touch</a>, file an issue or send a pull request!</p> Docker Docker 2017-05-06T00:00:00+00:00 Bootstrapping Docker for ARM64 (aka AARCH64)</a></h1> Basic process is:</p> bootstrap ARM64 go</code> toolchain on x86, and install</li> build ARM64 go1.7.5</code> toolchain needed for docker</code> build</li> bootstrap ARM64 docker</code> v1.10.3</li> use bootstrapped docker</code> to provide containerised build environment for building later versions</li> </ul> Instructions below are for CentOS 7 for Reasons(tm). Package details and so on will vary on other distros.</p> Build go</code> bootstrap toolchain</a></h2> On x86 host:</p> build basic go1.4 sufficient to bootstrap</li> </ul> cd</span></span> curl</span> -O</span> https://storage.googleapis.com/golang/go1.4-bootstrap-20161024.tar.gz</span></span> tar</span> xzvf go1.4-bootstrap-20161024.tar.gz</span></span> mv</span> go go1.4</span></span> cd</span> go1.4/src</span></span> ./make.bash</span></span></code></pre> cross-compile go1.7 (latest)</li> </ul> mkdir</span> -p</span> ~/go/src</span></span> cd</span> ~/go/src</span></span> git</span> clone https://go.googlesource.com/go</span></span> cd</span> go</span></span> git</span> checkout go1.7.5</span></span> cd</span> src</span></span> GOOS</span>=</span>linux</span> GOARCH</span>=</span>arm64</span> ./bootstrap.bash</span></span></code></pre> transfer cross-compiled toolchain to ARM64 host</li> </ul> scp</span> ~/go/src/go-linux-arm64-bootstrap.tbz HOST:~</span></span></code></pre>Build ARM64 go1.7.5</code></a></h2> produce bootstrap toolchains</li> </ul> cd</span></span> tar</span> xvf go-linux-arm64-bootstrap.tbz</span></span></code></pre> use the bootstrap toolchain to build a modern go</code> install</li> </ul> mkdir</span> -p</span> ~/go/src</span></span> cd</span> ~/go/src</span></span> git</span> clone https://go.googlesource.com/go</span></span> cd</span> go</span></span> git</span> checkout go1.7.5</span></span> cd</span> src</span></span> GOROOT_BOOTSTRAP</span>=</span>~/go-linux-arm64-bootstrap</span> ./make.bash</span></span> mv</span> ~/go/src/go/bin ~/go/bin</span></span></code></pre>Bootstrap ARM64 Docker</a></h2> get a recent git</code></li> </ul> sudo</span> yum remove git</span></span> wget</span> https://github.com/git/git/archive/v2.12.2.tar.gz</span></span> tar</span> xvf v2.12.2.tar.gz</span></span> cd</span> git-</span></span></span> which</span> git</span></span> sudo</span> yum install perl-devel hg curl-devel</span></span> make</span> configure</span></span> ./configure</span> --prefix=/usr/local</span></span> make</span> -j8</span></span> sudo</span> make install</span></span></code></pre> install dev dependencies</li> </ul> sudo</span> yum install btrfs-progs-devel device-mapper-devel</span></span></code></pre> clone source</li> </ul> cd</span> ~/go</span></span> mkdir</span> -p</span> src/github.com/docker</span></span> cd</span> src/github.com/docker</span></span> git</span> clone git@github.com:docker/docker</span></span> cd</span> docker</span></span></code></pre> build components</li> </ul> git</span> co v1.10.3</span></span> (</span> cd</span> vendor</span> &&</span> for</span> n</span> in</span> src/</span> ;</span> do</span> ln</span> -s</span> $n</span> ;</span> done</span> )</span></span> ./hack/make.sh</span> dynbinary</span></span> rm</span> -rf</span> vendor</span> &&</span> git</span> checkout .</span> # tidy up symlinking</span></span></code></pre> Note that a current bug in Ubuntu packaging metadata means a small edit needs to be made to ./Dockerfile.aarch64</code>: change the apt-get update &&</code> to apt-get update ;</code> so that the build doesn’t stop at the first hurdle, updating packages.</p> run daemon</li> </ul> # sudo rm -rf /var/lib/docker /etc/docker/config.json # DANGEROUS!</span></span> sudo</span> ./bundles/1.10.3/dynbinary/docker daemon</span> -D --group=wheel</span></span></code></pre> run client to check</li> </ul> $</span> PATH=</span>$(</span>pwd</span> -P</span>)</span>/bundles/1.10.3/dynbinary/:</span>$PATH</span></span> $</span> docker version</span></span> Client:</span></span> Version:</span> 1.10.3</span></span> API</span> version:</span> 1.22</span></span> Go</span> version: go1.7.5</span></span> Git</span> commit: 20f81dde9</span></span> Built:</span> Tue Apr</span> 4</span> 00:27:13</span> 2017</span></span> OS/Arch:</span> linux/arm64</span></span> </span> Server:</span></span> Version:</span> 1.10.3</span></span> API</span> version:</span> 1.22</span></span> Go</span> version: go1.7.5</span></span> Git</span> commit: 20f81dde9</span></span> Built:</span> Tue Apr</span> 4</span> 00:27:13</span> 2017</span></span> OS/Arch:</span> linux/arm64</span></span></code></pre>Build Docker</a></h2> reissue make</code> using containerised build</li> </ul> export</span> DOCKER_BUILDTAGS</span>=</span>'apparmor selinux seccomp'</span></span> git</span> co v17.05.0-ce</span> # or v1.12.3 or master or whatever</span></span> make</span> # transient failure of first build; restart succeeded</span></span> make</span> deb</span></span> mkdir</span> contrib/builder/rpm/aarch64</span></span> make</span> rpm</span></span></code></pre>Notes</a></h2> Resulting 17.05-dev</code> binaries fail due to missing libapparmor.so</code></li> Static binary build no longer supported?</li> DOCKER_BUILDTAGS</code> environment variable no longer supported?</li> 1.12.3</code> and 1.13.1</code> also built</li> Build of RPM or DEB packages fails due, I think, to DIND not working</li> </ul> Lab GitLab 2017-04-19T00:00:00+00:00 Recently had cause to do this as part of the SRG’s</a> and OCaml Labs</a> infrastructure. Thought it might be useful to make some notes, so here they are! Assuming your local sys-admin</code> has kindly created you a suitable VM running Ubuntu with login credentials, etc, read on…</p> Note that several commands that follow must be run as root</code>, via use of sudo</code> below. Given that, think twice before just cutting and pasting them in, obviously… And I am not held responsible for anything either way!</p> Install Docker</a></h2> On a new Ubuntu stretch/sid (testing) VM:</p> $</span> lsb_release</span> -drc</span></span> Description:</span> Ubuntu</span> 16.04.1</span> LTS</span></span> Release:</span> 16.04</span></span> Codename:</span> xenial</span></span></code></pre> Next, install up-to-date Docker</a>:</p> sudo</span> apt-get install apt-transport-https ca-certificates</span></span> apt-key</span> adv</span> --keyserver</span> hkp://p80.pool.sks-keyservers.net:80</span> \</span></span> --recv-keys</span> 58118E89F3A912897C070ADBF76221572C52609D</span></span> sudo</span> echo "deb https://apt.dockerproject.org/repo debian-stretch main"</span> \</span></span> ></span> /etc/apt/sources.list.d/docker.list</span></span> sudo</span> apt-get update</span></span> sudo</span> apt-get install</span> -y</span> docker-engine</span></span></code></pre> Tweak the systemd</code> Docker</a> configuration by adding a fragment to point all Docker</a> to the /data</code> partition, lest the root partition /</code> fill:</p> cat</span> ></span> /etc/systemd/system/docker.service.d/data-disk.conf</span> <<</span>EOF</span></span> [Service]</span></span> ExecStart=</span></span> ExecStart=/usr/bin/dockerd -H fd:// -g /data/docker</span></span> EOF</span></span></code></pre> Then start the Docker</a> daemon and run hello-world</code> just to check all is well:</p> sudo</span> systemctl daemon-reload</span></span> sudo</span> service docker start</span></span></code></pre> Finally, test the install by running hello-world</code>:</p> $</span> docker run hello-world</span></span> </span> Hello</span> from Docker!</span></span> This</span> message shows that your installation appears to be working correctly.</span></span> </span> To</span> generate this message, Docker took the following steps:</span></span> 1.</span> The Docker client contacted the Docker daemon.</span></span> 2.</span> The Docker daemon pulled the "hello-world" image from the Docker Hub.</span></span> 3.</span> The Docker daemon created a new container from that image which runs the</span></span> executable</span> that produces the output you are currently reading.</span></span> 4.</span> The Docker daemon streamed that output to the Docker client, which sent it</span></span> to</span> your terminal.</span></span> </span> To</span> try something more ambitious, you can run an Ubuntu container with:</span></span> $</span> docker run</span> -it</span> ubuntu bash</span></span> </span> Share</span> images, automate workflows, and more with a free Docker Hub account:</span></span> https://hub.docker.com</span></span> </span> For</span> more examples and ideas, visit:</span></span> https://docs.docker.com/engine/userguide/</span></span></code></pre> If appropriate, you may also wish to add yourself to the docker</code> user group:</p> sudo</span> usermod</span> -aG</span> docker</span> $(</span>whoami</span>)</span></span></code></pre>GitLab</a></h2> Assuming you have rights to run docker</code>, install and run Gitlab-CE:</p> S</span>=</span>128.232.xxx.yyy</span></span> H</span>=</span>gitlab.srg.cl.cam.ac.uk</span></span> docker</span> run</span> --detach \</span></span> --hostname</span> $H</span> \</span></span> --publish</span> $S</span>:443:443</span> \</span></span> --publish</span> $S</span>:80:80</span> \</span></span> --publish</span> $S</span>:2222:22</span> \</span></span> --name</span> gitlab</span> \</span></span> --restart</span> always</span> \</span></span> --volume</span> /data/gitlab/config:/etc/gitlab</span> \</span></span> --volume</span> /data/gitlab/logs:/var/log/gitlab</span> \</span></span> --volume</span> /data/gitlab/data:/var/opt/gitlab</span> \</span></span> --volume</span> /data/gitlab/backups:/var/opt/gitlab/backups</span> \</span></span> --volume</span> /data/gitlab/sync:/var/opt/gitlab/sync</span> \</span></span> --env</span> HOST_UID=</span>$$</span>(</span>id</span> -u</span>)</span> --env</span> HOST_GID=</span>$$</span>(</span>id</span> -g</span>)</span> \</span></span> mor1/gitlab-ce-cron:latest</span></span></code></pre> …or use the make start</code> target in the Makefile</a> in the related GitHub repo</a>.</p> TLS Certificates</a></h2> Self-certified certificates:</p> openssl</span> req</span> -nodes -newkey</span> rsa:2048</span> -keyout</span> gitlab.srg.cl.cam.ac.uk.key</span> -out</span> gitlab.srg.cl.cam.ac.uk.csr</span></span> cd</span> ssl</span></span> chmod</span> 600</span> </span></span> openssl</span> x509</span> -req -days</span> 1460</span> -in</span> gitlab.srg.cl.cam.ac.uk.csr</span> -signkey</span> gitlab.srg.cl.cam.ac.uk.key</span> -out</span> gitlab.srg.cl.cam.ac.uk.crt</span></span></code></pre>Run Backups</a></h2> backup</code> script to create backup tarballs and extract</li> sync</code> script to rsync extracted tarballs to filer</li> </ul> Recovering Password</a></h2> To change the root</code> password you need to use the Ruby-on-Rails console to access the relevant object, modify it, and save it back:</p> gitlab</span>-</span>rails console production</span></span> </span> irb</span>(main):</span>001</span>:</span>0</span>></span> user </span>= User</span>.</span>where</span>(</span>id:</span> 1</span>).</span>first</span></span> =></span> #<User id: 1, email: "admin@example.com", created_at: "2016-11-16 22:57:21", updated_at: "2016-12-05 23:42:50", name: "Administrator", admin: true, projects_limit: 10, skype: "", linkedin: "", twitter: "", authentication_token: "secrettoken", theme_id: 2, bio: nil, username: "root", can_create_group: true, can_create_team: false, state: "active", color_scheme_id: 1, password_expires_at: nil, created_by_id: nil, last_credential_check_at: nil, avatar: nil, hide_no_ssh_key: false, website_url: "", notification_email: "admin@example.com", hide_no_password: false, password_automatically_set: false, location: nil, encrypted_otp_secret: nil, encrypted_otp_secret_iv: nil, encrypted_otp_secret_salt: nil, otp_required_for_login: false, otp_backup_codes: nil, public_email: "", dashboard: 0, project_view: 0, consumed_timestep: nil, layout: 0, hide_project_limit: false, otp_grace_period_started_at: nil, ldap_email: false, external: false, organization: nil></span></span> irb</span>(main):</span>002</span>:</span>0</span>></span> user.</span>password</span> =</span> 'secretpassword'</span></span> =></span> "secretpassword"</span></span> irb</span>(main):</span>003</span>:</span>0</span>></span> user.</span>password_confirmation</span> =</span> 'secretpassword'</span></span> =></span> "secretpassword"</span></span> irb</span>(main):</span>004</span>:</span>0</span>></span> user.</span>save!</span></span> Enqueued</span> ActionMailer</span>::</span>DeliveryJob</span> (</span>Job</span> ID:</span> 5f74573d</span>-</span>dfa2</span>-</span>4778</span>-</span>b365</span>-</span>cbebd88e454e) to </span>Sidekiq</span>(mailers) with </span>arguments:</span> "DeviseMailer"</span>,</span> "password_change"</span>,</span> "deliver_now"</span>,</span> gid:</span>//</span>gitlab</span>/</span>User</span>/</span>1</span></span> =></span> true</span></span> irb</span>(main):</span>005</span>:</span>0</span>></span></span> </span> gitlab</span>-</span>ctl reconfigure</span></span></code></pre>Hook up to GitHub</a></h2> Per https://docs.gitlab.com/ce/integration/omniauth.html#initial-omniauth-configuration</a> and https://docs.gitlab.com/ce/integration/github.html</a>:</p> Edit via sudo docker exec -it gitlab /bin/bash</code>:</p> root@gitlab:/#</span> vi /etc/gitlab/gitlab.rb</span></span></code></pre>gitlab_rails[</span>'omniauth_enabled'</span>]</span> =</span> true</span></span> gitlab_rails[</span>'omniauth_allow_single_sign_on'</span>]</span> =</span> [</span>'saml'</span>,</span> 'github'</span>]</span></span> gitlab_rails[</span>'omniauth_block_auto_created_users'</span>]</span> =</span> true</span></span></code></pre>Email</a></h2> Use SMTP via <ppsw.cam.ac.uk>, for which the from address must have a valid MX</code> record and</strong> not be under <cam.ac.uk> per http://help.uis.cam.ac.uk/email-telephony-and-collaboration/email/technical/sending</a>.</p> Configuration can be tested via the console:</p> Notify</span>.</span>test_email</span>(</span>'your@email.address, '</span>Hello World</span>', '</span>This</span> is a </span>test</span> message</span>').deliver_now</span></span></code></pre> Google screening 2017-01-29T00:00:00+00:00 Some time ago, for reasons best known to themselves, a Google recruiter decided to “reach out” on the basis of my GitHub profile</a> to see if I were interested in a role as a Site-Reliability Engineer or possibly a Software Engineer. This entailed a short (~30min) telephone interview to answer some questions. I made a note of those I recalled, in case anyone’s interested.</p> The hawk-eyed and keen-minded among you may discern a certain amount of ambiguity in answers to some of the questions – e.g., is the opposite of malloc()</code>, free()</code> or a garbage collector? are we assuming an Ethernet MAC address? – which the recruiter did not seem to be happy to deal with. But so long as my answer included a reasonable approximation to (presumably) the string they had written down, all was well.</p> What is the Big-O complexity of quicksort?</li> What is the search complexity for a red-black tree, a binary tree, a linked list, a hashtable, and a B-tree?</li> What’s the opposite of malloc()</code>?</li> What are the semantics of an ACL?</li> Which of the following fields are not</em> part of the passwd</code> file? shell, comment, initial umask, login name, ui, home directory, gid, preferred language</li> </ul> </li> What does the fstat</code> syscall do?</li> What’s the default signal for kill</code>?</li> What’s in an inode?</li> How do you make a socket accept inbound connections?</li> How many bytes in a MAC address?</li> What are the packets involved in a TCP connection setup?</li> How many hosts are in a /23 subnet?</li> What’s the DNS resource record type for an IPv6 address?</li> Estimate the value of 2^{24</sup>.</li> </ul> In the end, I passed even though I could only remember the name, not the number, of the default signal for kill</code>. It then got mildly amusing: the next stage is apparently to “jump on a call” (sigh) with a recruiter and an engineer to work through some coding problems. I explained that I generally refuse to engage in whiteboard coding during interviews (it’s not a useful measure of anything useful, and I don’t see why I should). They said oh but of course I could do it on a call so it wouldn’t actually be a whiteboard. I said, yes I could but no I wouldn’t and I thought they were rather missing my point. They said really, it was very unusual for someone to refuse. I said, to be honest it makes little sense anyway given they contacted me because of all the code I’d written under my GitHub account</em>. They said oh well.</p> And then some time later – 6 months I think – a different recruiter “reached out” to ask why the process had stalled and did I want to jump on a call.</p> I said No. They haven’t called back since. Oh well…</p>} Customising Homebrew service 2017-01-04T00:00:00+00:00 As I could never remember the command to start the offlineimap</code> service using my customised configuration, here it is:</p> brew</span> services start offlineimap ~/rc-files/homebrew.mxcl.offlineimap.plist</span></span></code></pre> Building Up Your ARMs 2016-11-16T00:00:00+00:00 Due to the impending finish of the EU FP7 funded User Centric Networking</a>1</a></sup> I recently had cause to revisit the excellent work that Thomas Leonard</a> did for the project in getting Xen/ARM running on the Cubieboard2</a> and Cubietruck</a> (aka Cubieboard3</a>).</p> The resulting repo, mirage/xen-arm-builder</a>, had languished for several months and the past SD card images had some problems and had been allowed to drop off the ’Net as a result. However, sterling work by Ian Campbell</a> at a recent Mirage hackathon</a> had started to resurrect this work based on the Alpine Linux</a> distribution. This seemed a promising place to start, so I did :)</p> Building an Image</a></h2> The end result was an enormous pull request</a> that splatted a Brave New World on top of Thomas’</a> work. The README</code></a> is hopefully reasonably self-explanatory but in summary,</p> Clone the repo:</p> git</span> clone https://github.com/mor1/arm-image-builder.git</span></span> cd</span> arm-image-builder</span></span></code></pre></li> Use the make</code> targets:</p> make</span> all</span> # runs `make prepare build image`</span></span> # make prepare # clones repos, pulls tarballs</span></span> # make build # use Docker to build the `linux/` and `u-boot/` trees</span></span> # make image # finally, create the on-disk `sdcard.img`</span></span></code></pre></li> </ol> This clones the necessary repos (Linux, u-boot), builds them, and then puts together the image file sdcard.img</code> in the current directory. If on OSX, make sdcard</code> will then attempt to write that to a blank, mounted SD card. This does a rather hacky auto-discovery of where the SD card might be mounted; if in doubt, and in any case, always safer to simply</p> MNT</span>=</span>the-correct-mount-point</span> make</span> sdcard</span></span></code></pre> …or simply use your favourite tools to write the sdcard.img</code> file to your SD card.</p> Using the Image</a></h2> The end result should be an SD card that you can use to boot your device into Alpine Linux v3.4</a>. At present, completing installation requires then:</p> resetting the environment</a>,</li> completing Alpine setup</a> via the setup-alpine</code> script,</li> (if desired) installing Xen via the /media/mmcblk0p1/alpine-dom0-install.sh</code> script created as part of building the SD card image,</li> (if desired) finally, building Alpine</a> and/or Debian</a> domU</code>s via the /media/mmcblk0p1/alpine-domU-install.sh</code> and /media/mmcblk0p1/debian-domU-install.sh</code> scripts, also created as part of building the image.</li> </ul> Hopefully the net result is you end up with a Cubieboard2/3 running Xen with an Alpine Linux dom0</code> and some domU</code> images available.</p> As ever, comments, patches, pull requests welcome</a>!</p> Grant No. 611001 for those who care. ↩</a></p> </li> </ol> </section> Debugging with jQuery 2016-11-16T00:00:00+00:00 I had cause to do this recently, so here’re the runes from http://stackoverflow.com/questions/7474354/include-jquery-in-the-javascript-console</a> for the record:</p> var</span> script</span> =</span> document</span>.</span>createElement</span>(</span>'script'</span>);</span></span> script</span>.</span>src</span> =</span> "https://ajax.googleapis.com/ajax/libs/jquery/1.6.3/jquery.min.js"</span>;</span></span> document</span>.</span>getElementsByTagName</span>(</span>'head'</span>)[</span>0</span>].</span>appendChild</span>(</span>script</span>);</span></span></code></pre> Zen and the Art of Research Management 2016-11-11T00:00:00+00:00 I think this is a bit of a classic, the first written form of which I came across in Prof. Roger Needham</a>’s 50+5 Festschrift</a> celebrating his time at the Cambridge University Computer Lab</a> and Microsoft Research Cambridge</a>. I don’t know who originated it, but the copy there is certainly due to John Naughton</a> and Bob Taylor</a>. I suppose one might quibble point 12, in that I seem to recall Roger did a lot of pacing about, but a good chair is certainly a worthwhile thing to provide.</p> Anyway, I find myself wanting to point at it from time-to-time, so here it is!</p> By John Naughton</a> (Open University, Milton Keynes, England</em>), and Robert W. Taylor</a> (Woodside, California, USA</em>).</p> HIRE ONLY THE VERY BEST PEOPLE, EVEN IF THEY ARE CUSSED. Perhaps especially if they are cussed. Your guiding principle should be to employ people who are smarter than you. One superb researcher is worth dozens of merely good ones.</p> </li> ONCE YOU’VE GOT THEM, TRUST THEM. Do not attempt to micro-manage talented people. (Remember rule #1.) Set broad goals and leave them to it. Concentrate your own efforts on strategy and nurturing the environment.</p> </li> PROTECT YOUR RESEARCHERS FROM EXTERNAL INTERFERENCE, whether from company personnel officers, senior executives or security personnel. Remember that your job is to create a supportive and protective space within which they can work.</p> </li> MUCH OF WHAT YOU DO WILL FALL INTO THE CATEGORY OF ABSORBING THE UNCERTAINTY OF YOUR RESEARCHERS.</p> </li> REMEMBER THAT YOU ARE A CONDUCTOR, NOT A SOLOIST. (Rule #1 again.) The Lab is your performance.</p> </li> DO NOT PAY TOO MUCH ATTENTION TO ‘RELEVANCE,’ ‘DELIVERABLES’ and other concepts beloved of Senior Management.</p> </li> REMEMBER THAT CREATIVE PEOPLE ARE LIKE HEARTS – they go where they are appreciated. They can be inspired or led, but not managed.</p> </li> KEEP THE ORGANISATION CHART SHALLOW. Never let the Lab grow beyond the point where you cannot fit everyone comfortably in the same room.</p> </li> MAKE YOUR RESEARCHERS DEBATE WITH ONE ANOTHER REGULARLY. Let them tear one another’s ideas to pieces. Ensure frank communication among them. Observe the strengths and weaknesses which emerge in the process.</p> </li> BE NICE TO GRADUATE STUDENTS. One day they may keep you, even if only as a mascot. (Moreover, they are a lot of fun!)</p> </li> INSTALL A WORLD-CLASS COFFEE MACHINE and provide plenty of free soft drinks.</p> </li> BUY AERON CHAIRS. Remember that most computer science research is done sitting down.</p> </li> INSTITUTE A ‘TOY BUDGET’, enabling anyone in the Lab to buy anything costing less than a specified amount on their own authority. And provide a darkened recovery room for accountants shocked by the discovery of this budget.</p> </li> PAY ATTENTION TO WHAT GOES ON IN UNIVERSITIES. Every significant breakthrough in computing in the last four decades has involved both the university and corporate sectors at some point in its evolution.</p> </li> REMEMBER TO INITIATE AND SPONSOR CELEBRATIONS when merited.</p> </li> WHEN IN DOUBT, ASK YOURSELF: “WHAT WOULD ROGER NEEDHAM DO IN SIMILAR CIRCUMSTANCES?”</p> </li> </ol> Evolving the Unikernel Revolution 2016-11-10T00:00:00+00:00 I’ve had the pleasure of giving a couple of talks at some fun venues recently, extolling both the virtues of unikernels</a> and talking a bit about where we currently see them as usefully being deployed.</p> Specifically, Operability.io 2016</a> a couple of weeks ago was enlightening about some of the problems faced in operating production systems. Some great audience questions and follow-ups after the talk, including some who were even wondering when we’ll see unikernels as ready for the desktop! Of course, with the release of the Docker for Mac</a> and [Docker for Windows][dfw] products, it’s arguable that we’ve beaten Linux to that accolade, as both products make extensive use of MirageOS</a> unikernel libraries. Having said that, I was pleased to be told that the message about unikernels having a range of deployment scenarios, and particularly partial deployments into micro-service environments made sense to many who came to speak to me afterwards.</p> This was followed by a slightly expanded version of that talk earlier today at the Devox Belgium</a> conference. Devoxx</a> is primarily a Java community so I was interested to see how the talk would go down given that MirageOS</a> is staunchly OCaml-centric, and the unikernels</a> movement in general is language specific and (at least until now) somewhat weighted toward functional programming, our good friends at IncludeOS</a> notwithstanding. In the end it seemed to go pretty well, based on what little I could see through the bright lights– maybe one day I’ll get used to that when being videoed! Certainly some good questions again, on the specific utility of unikernels to IoT, the relationship between unikernels and Docker, and more besides.</p> Anyway, I hope anyone who came to either talk enjoyed it and found it interesting. Happy to respond to comments or questions via email or on Twitter</a>!</p> Rat Racing Jetlag 2016-11-04T00:00:00+00:00 As I’ve found myself repeating the same information several times recently, and have to dig out the links in question every time, I figured it’d be useful to write this down once so I can point at it.</p> Jetlag is a first-world problem but can be an annoying one– in recent years I’ve found I have a particular problem getting up in the morning when flying east. So, one day, bored in an airport in the US and with only the entire Internet to hand, I thought I’d look around for an explanation at least.</p> I ended up coming across reports of some physiological research – in rats of course – that suggested a way to avoid jetlag. I’ve since tried it over a dozen times, and found it to work in all cases. At this point I no longer care if it’s placebo affect or a genuine explanation or something else :)</p> The TL;DR is: fast for >16 hours before your time of arrival, then eat as appropriate. Definitely >16 hours though – the occasions I’ve not quite managed it (at least once I miscounted and managed only about 14 hours) it didn’t seem to work as effectively.</p> As I interpret the information in the article</a>, found via BBC</a> and Harvard</a> reports, through my “I can barely do computer science and certainly not biology” brain, mammals have two body clocks, one driven by daylight and one by metabolism. The daylight can’t be shifted quickly, but the metabolic one can be made to float by fasting. The metabolic also being a lower level clock means that, when you resync it by consuming calories, it syncs back to whatever light conditions are current.</p> As I say, this isn’t my area of expertise – but it seems to work anyway.</p> Also, because a surprisingly large (or perhaps not) number of people also ask– as far as I know and have experienced, fasting means no</strong> calories, including alcohol</strong>, even it it’s free… :)</p> Grubbing Around 2016-09-13T00:00:00+00:00 Nothing earth-shattering here: I recently had the “pleasure” of setting up an ARM64 server. After considerable support, several firmware upgrades, corruption of the main HDD, reinstallation of CentOS7 (recommended, somewhat to my surprise), all that remained was to get an up-to-date Linux built and installed with 32 bit binary support. This took a bit of make config</code> fiddling, but got there after a few tries.</p> And then I had to relearn how grub</code>/grub2</code> works in this brave new (to me) UEFI CentOS7 world. Herewith some brief commands I found useful while doing so…</p> sudo</span> grep "^menu entry" /boot/efi/EFI/centos/grub.cfg</span> \</span></span> \|</span> tr</span> -s</span> " "</span> \|</span> cut</span> -f</span> 2</span> -d</span> "'"</span> \|</span> cat</span> -n</span></span></code></pre> Edit /etc/default/grub</code> to set GRUB_DEFAULT=N</code> for desired value of N</code></p> Temporarily set the default for the next reboot:</p> sudo</span> grub2-reboot</span> 1</span> # based on output of above</span></span></code></pre> Regenerate the grub2 configuration:</p> sudo</span> grub2-mkconfig</span> -o</span> /boot/efi/EFI/centos/grub.cfg</span></span></code></pre> Walking The Ridgeway 2016-08-28T00:00:00+00:00 After roughly 35 years of my dad and I threatening to do so, following a rather formative experience I had in the Scouts, my brother finally sorted us out with a plan to walk The Ridgeway</a>, popularly known as “Britain’s oldest road”. It was a blast. Local Knowledge 2016-03-11T00:00:00+00:00 A relatively long hiatus this time – blame the Cambridge housing market and the need to simultaneously act both incredibly quickly and mind-numbingly slowly. Also obtaining a College Fellowship</a> and experiencing a Company Acquisition</a> were considerable (interesting, welcome) distractions… :)</p> As an interim measure, and because I’ve been asked relatively frequently over the last few months, I thought I’d collect some local knowledge^Wopinion, garnered over many long hard years of hanging about in both Cambridge and Nottingham. What follows should be viewed as indicating nothing more than my lack of imagination.</p> Both Cambridge and Nottingham are blessed with many fine pubs, as well as several truly terrible ones. Less said about the latter the better. But among those that I like, and will occasionally drag various people to (including the SRG</a>) are, arranged roughly in geographical order from the Computer Lab to the Station:</p> The Plough, Coton</a>. Gastro-pub-ish. Frequently changed management over recent years, but currently seems pretty consistent and (finally!) is not limited to Greene King beers.</li> The Castle Inn</a>. Major long-standing Adnams “proper pub”. Very consistent. Castle Burger and chips please.</li> The Pickerel</a>. Also largely unchanged in a couple of decades. Good for a pint before entering the Greene King desert.</li> The Radegund</a>. Tiniest pub I know of. Recently quite radically refurbished. But, while not what it was, seems pretty decent nonetheless.</li> The Free Press</a>. Has improved significantly while I was in Nottingham. Very good food and beer. Stocks the only actually pleasant Greene King beer (XX Mild</a>), as well as an interesting range of others. Quite small and often packed if it’s too cold to sit outside.</li> The Cambridge Blue</a>. Surely one of the best pubs for many many miles. Excellent and huge range of beer (and, apparently, cider; won’t touch the stuff myself). Good food. Large and with a large garden too – rarely a problem finding somewhere to sit during the week.</li> The Kingston Arms</a>. Another excellent pub in the station area. Possibly better food than the Blue, potentially slightly less wide ranging set of beers (and, certainly, ciders). Also a nice garden, though in all respects somewhat smaller than the Blue so prone to being absolutely packed. Doesn’t serve lager – a certain German colleague makes do with JHB though.</li> The Live and Let Live</a>. Yet one more excellent and, though small, often less busy pub near the station. No food in the evenings.</li> The Devonshire Arms</a>. Largest range of Milton Brewery beers I know of. Food = pizzas I believe (though not, last time I tried, as good as those at The Carpenters Arms</a>).</li> </ul> I don’t get out to the villages as often as I’d like now, but The Green Man, Grantchester</a>, The Blue Lion, Hardwick</a> and especially The Red Lion, Histon</a> are (or were last time I went) all excellent too. The latter is still possibly my favourite pub anywhere in fact.</p> I only spent a few years in Nottingham, and it’s a Proper City unlike Cambridge, so I can’t claim to have tested it thoroughly. However, places I did visit fairly regularly in Nottingham, Beeston, and Wollaton that I certainly enjoyed include:</p> The Roundhouse</a>. A little pricey for Nottingham, but good quality beer and food, nice atmosphere.</li> The Lincolnshire Poacher</a>. Very excellent pub. Good whisky range too.</li> Fellows, Morton & Clayton</a>. Consistently good, proper pub food, about 4 minutes walk from the station platforms.</li> The Vat & Fiddle</a>. Only finally tried more recently. Castle Rock brewery tap. Good food and beer, also about 4 minutes walk from your station platform.</li> The Hand & Heart</a>. In a cave! It’s a pub in a cave! With excellent beer and good food! And not owned by Greene King! (Unlike Ye Olde Trip to Jerusalem</a> which has bigger, better caves but markedly less good beer or food.)</li> The Organ Grinder</a>. Another fine pub at Canning Circus. Blue Monkey brewery this time – rather good. Food limited to (nice!) pork pies and such, but the beer is good.</li> The Plough Inn</a>. Closest (decent) pub to Jubilee Campus. Nottingham City brewery tap. Excellent beer, no food in the evenings, and prices that make me think I’m 15 years younger…</li> The Victoria Hotel, Beeston</a>. Consistently excellent pub. Excellent beer, very good food.</li> The Crown Inn, Beeston</a>. Also consistently excellent beer, though no food in the evenings.</li> The Wollaton Pub and Kitchen, Wollaton</a>. Gastro-pub-ish. Generally good beer, often excellent but occasionally patchy food.</li> </ul> So there you go. Some opinions if relatively little knowledge. YMMV. Etc.</p> (PS. If I know you, I’ll also be happy to give recommendations of solicitors, builders and other house-associated professionals too. And a couple of warnings.)</p> Of Mice and Keyboards 2015-11-21T00:00:00+00:00 A bit of a delay since the last post – lots going on! But anyway: I (relatively) recently upgraded my old skool Macbook Pro (look! built-in Ethernet port! DVD drive!) to El Capitan. This was generally rather less faff that the previous upgrade, though it did seem to take rather more reboots than might have been assumed to be strictly</em> necessary before it settled down, and I’d remembered to fix up permissions for Homebrew with sudo chown -R $(whoami):admin /usr/local</code>. So that was ok.</p> </p> Except… I have a slightly odd keyboard and mouse setup. It’s a UK Macbook which means a slightly tweaked keyboard layout compared to the standard US Macbook keyboard. At my desk, I also use a Microsoft Digital Media Keyboard</em> – nice action (for me!) plus some handy shortcut keys – and a Microsoft 5-Button Mouse with IntelliEye</em>. Now, until El Capitan I’d happily been using the Microsoft provided software to make use of the extra mouse buttons and shortcut keys, coupled with a Ukelele-generated</a> keymap to handle the oddities of the UK laptop keyboard (like, who in the world really needs §</code> at the top-left key, below escape</code> rather than `</code>; and doesn’t need an easily accessible #</code>?).</p> This had never been entirely satisfactory – I had to have a standard keymap installed in addition to my modified one, and some apps (all of Microsoft Office, I’m looking at you) liked to intermittently flip the keymap away from my keymap to the standard issue on, including undoing my remapping of caps lock</code> to ctrl</code>. This was annoying, but having it completely break was intolerable.</p> So I went hunting for alternatives and am now very happy with Karabiner.app</a> for standard keyboard remappings, and fairly happy with USB Overdrive</a> to handle the mouse and the special Microsoft Digital Media Keyboard shortcut keys.</p> USB Overdrive</a></h3> USB Overdrive seems to do the mouse mappings correctly, having detected the device as a “Microsoft 5-Button Mouse with IntelliEye(tm), Any Application” – Button 4</code> and Button 5</code> can be remapped to forward</code> and back</code>, just as I like it.</p> </p> It also allows me to repurpose some of the extra keys on my Microsoft keyboard that Karabiner</a> doesn’t seem able to see– so I get one touch play/pause of iTunes and other such delights.</p> Karabiner.app</a></h3> Karabiner</a> took a bit more setting up but does a very nice job. I needed to remap certain keys differently on the two different keyboards to make both consistent and to fix some of the weirder (to my mind!) decisions both Microsoft and (particualrly) Apple have taken with their layouts. The result is an XML configuration file</a>, symlinked by ~/Library/Application Support/Karabiner/private.xml</code>. This applies two keymaps based on the detected device, using product ID codes determined by the EventViewer</code> app that comes with Karabiner</a>:</p> <</span>deviceproductdef</span>></span></span> <</span>productname</span>></span>MACBOOK_PRO_UK_KEYBOARD</span></</span>productname</span>></span></span> <</span>productid</span>></span>0x0253</span></</span>productid</span>></span></span> </</span>deviceproductdef</span>></span></span> </span> <</span>deviceproductdef</span>></span></span> <</span>productname</span>></span>DIGITAL_MEDIA_KEYBOARD</span></</span>productname</span>></span></span> <</span>productid</span>></span>0x00b4</span></</span>productid</span>></span></span> </</span>deviceproductdef</span>></span></span> </span> <</span>deviceproductdef</span>></span></span> <</span>productname</span>></span>FIVE_BUTTON_MOUSE_WITH_INTELLIEYE</span></</span>productname</span>></span></span> <</span>productid</span>></span>0x0039</span></</span>productid</span>></span></span> </</span>deviceproductdef</span>></span></span></code></pre> There are then two <item></item></code> stanzas that configure the two different keyboards, e.g.,</p> <</span>item</span>></span></span> <</span>name</span>></span>Keyboard mappings for Microsoft keyboard</span></</span>name</span>></span></span> <</span>identifier</span>></span>private.io.mort.microsoft_keyboard</span></</span>identifier</span>></span></span> <</span>device_only</span>></span></span> DeviceVendor::MICROSOFT,</span></span> DeviceProduct::DIGITAL_MEDIA_KEYBOARD</span></span> </</span>device</span>></span></span> ...</span></span></code></pre> Each of these contains a number of <autogen></autogen></code> stanza mapping specific keycodes for that keymap. For example, I want the top-left key on the main block to be `</code> and, when shifted, to be €</code>. This leads to the following on the Microsoft keyboard:</p> <!-- shift-` to € --></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::BACKQUOTE, ModifierFlag::SHIFT_L \| ModifierFlag::NONE,</span></span> KeyCode::KEY_2, ModifierFlag::OPTION_R \| ModifierFlag::SHIFT_R</span></span> </</span>autogen</span>></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::BACKQUOTE, ModifierFlag::SHIFT_R \| ModifierFlag::NONE,</span></span> KeyCode::KEY_2, ModifierFlag::OPTION_R \| ModifierFlag::SHIFT_R</span></span> </</span>autogen</span>></span></span></code></pre> …but to the following on the Macbook built-in UK keyboard, to take account first of the different keycode it generates but also to ensure that when used with command and command-shift, the standard behaviour of cycling between windows works:</p> <!-- top-left § to ` --></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::DANISH_DOLLAR, ModifierFlag::NONE,</span></span> KeyCode::BACKQUOTE</span></span> </</span>autogen</span>></span></span> <!-- ...with shift, to € --></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::DANISH_DOLLAR, ModifierFlag::SHIFT_L \| ModifierFlag::NONE,</span></span> KeyCode::KEY_2, ModifierFlag::OPTION_R \| ModifierFlag::SHIFT_R</span></span> </</span>autogen</span>></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::DANISH_DOLLAR, ModifierFlag::SHIFT_R \| ModifierFlag::NONE,</span></span> KeyCode::KEY_2, ModifierFlag::OPTION_R \| ModifierFlag::SHIFT_R</span></span> </</span>autogen</span>></span></span> <!-- ...with COMMAND/SHIFT, so that cycle-window-{forward,back} work --></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::DANISH_DOLLAR, ModifierFlag::COMMAND_L \| ModifierFlag::NONE,</span></span> KeyCode::BACKQUOTE, ModifierFlag::COMMAND_R</span></span> </</span>autogen</span>></span></span> <</span>autogen</span>></span></span> __KeyToKey__</span></span> KeyCode::DANISH_DOLLAR, ModifierFlag::COMMAND_L \| ModifierFlag::SHIFT_L \| ModifierFlag::NONE,</span></span> KeyCode::BACKQUOTE, ModifierFlag::COMMAND_R \| ModifierFlag::SHIFT_R</span></span> </</span>autogen</span>></span></span></code></pre> There are a number of other mappings made in karabiner.xml</a>: shift-'</code> is @</code>, shift-2</code> is "</code>, shift-3</code> is £</code>, and resolving general confusion among #</code>, \</code>, ~</code>, and \|</code>.</p> Emacs</a></h3> That fixed things for the terminal and for most apps – the only remaining sticking point was Emacs. I don’t pretend to understand the entire chain of event processing but suffice it to say that Emacs was receiving shift-@</code> and shift-3</code> without knowing what to do with them. Fortunately, when coupled with my hacks to enforce a my-keys-minor-mode</code> to override everything</a>, the fix was pretty straightforward:</p> (define-key my-keys-minor-mode-map (kbd </span>"s-@"</span>) </span>"€"</span>)</span></span> (define-key my-keys-minor-mode-map (kbd </span>"s-3"</span>)</span></span> '</span>(</span>lambda</span> () (interactive) (insert-char </span>#x00A3</span>))) </span>; £</span></span></code></pre>Result?</a></h3> A significant</strong> decrease in the need I feel to curse because my keyboard has changed in the middle of typing! It seems that keyboards remain, like time and terminals, one of those Really Hard</em> things for computers/manufacturers to handle…</p> Note: Thanks to http://www.amp-what.com/unicode/search/</a> for an easy way to hunt down some of the unicode symbols used above!</em></p> Fame and Fortune! 2015-06-24T00:00:00+00:00 After a slew of HDI related items, a quick MirageOS</a> note. I was invited to give a presentation on MirageOS at the TNG</a> Big TechDay 8</a> conference. This was an interesting, and very broad based, event. It brought together about 500 people with about 25 speakers over one day, on topics from cognitive science to AI to chess playing algorithms to obviating testing through proof reasoning. And, of course, operating systems and MirageOS</a>! If you’re interested, the slides used are available at decks.openmirage.org</a> in the usual way, and if you really want to hear me droning on, TNG are making the video (with synchronised slides) available</a>.</p> As ever, comments welcome!</p> Actual HCI Papers! 2015-06-17T00:00:00+00:00 I have to confess to being quite pleased to having a couple of HDI</a>-related papers accepted recently (even if we can’t even get proper reviews on the original HDI paper</a> – recently judged out-of-scope for the third time, even though the Special Issue in question seemed bang on target!).</p> The first is a full paper to ECSCW’15</a> titled Human Data Interaction: Historical Lessons from Social Studies and CSCW</em></a>. A collaboration with Dr Andy Crabtree</a>, it examines particularly the role of interaction</em> in HDI and explores how past technical approaches, such as Dataware</a>, to the challenges posed by HDI haven’t fully considered the inherently social nature of data.</p> The second is a short paper to the decennial – seriously: every ten years! – Aarhus 2015</a> conference titled Personal Data: Thinking Inside the Box</em></a>. This sets out a vision for an embodiment of a Databox</em>: a physical device supported by online services that empowers us to take back control of our online lives. Building a Databox, using MirageOS</a>, is hopefully going to be a key activity for me in the coming months…</p> Rescuing a Shattered Nexus 4 2015-05-21T00:00:00+00:00 A little while ago, before I’d done the smart thing and got myself a case for my Nexus 4, I dropped it a couple of inches onto a hard surface at the wrong angle. The screen promptly shattered – and this was bad because without the touch screen, I couldn’t interact with it, I had some photos on it from son#1 birthday party that hadn’t been copied off, and I hadn’t got round to enabling USB access to the filesystem or any of the debug/developer options.</p> So what to do? I really</em> didn’t want to lose those photos. A couple of hours searching the Interwebs and a little bit of experimentation later, and I managed it. Basically, download and apply the clockwork mod bootloader, and this turns on the developer options that allow access to the filesystem via the Android SDK tools. To find out the details, read on…</p> First, download the recovery image:</p> $</span> wget http://download2.clockworkmod.com/recoveries/recovery-clockwork-touch-6.0.3.1-mako.img</span></span></code></pre> Next, install the Android SDK – I’m on OSX using [Homebrew][] so I do:</p> $</span> brew install android-sdk</span></span></code></pre> Now, power off and disconnect the phone! Then boot it into fastboot mode by holding down power</code> and volume-down</code>. Once it boots you should be in the fastboot list – the volume keys will cycle you through the list. You should now also be able to see the device once connected to USB, and you can then OEM unlock it:</p> $</span> sudo fastboot devices</span> -l</span></span> 04f02d4bdcd3b6e2</span> fastboot usb:FD123000</span></span> $</span> sudo fastboot oem unlock</span></span> ...</span></span> OKAY</span> [</span> 17.937s]</span></span> finished.</span> total time: 17.937s</span></span></code></pre> Having unlocked it, you can now install the clockwork recovery bootloader you downloaded (assuming it’s in the local directory):</p> $</span> sudo fastboot flash recovery recovery-clockwork-touch-6.0.3.1-mako.img</span></span> sending</span> 'recovery'</span> (7560</span> KB</span>)...</span></span> OKAY</span> [</span> 0.526s]</span></span> writing</span> 'recovery'...</span></span> OKAY</span> [</span> 0.448s]</span></span> finished.</span> total time: 0.975s</span></span></code></pre> When you now use the volume keys to cycle through the list, you should now see recovery mode</strong> as an option – select it, and you should be able to see the device listed in the usual way via adb</code>:</p> :</span> mort@greyjay:phone</span>$;</span> sudo</span> adb devices</span> -l</span></span> List</span> of devices attached</span></span> 04f02d4bdcd3b6e2</span> recovery usb:FD123000 product:occam model:Nexus_4 device:mako</span></span></code></pre> Finally, pull all the contents off the sdcard:</p> $</span> adb pull /sdcard/0 ./sdcard/</span></span> $</span> adb pull /data/ ./data/</span></span> $</span> adb pull /system/ ./system/</span></span></code></pre> …and that’s it – you should now have a local copy of everything off the phone, and you can send it away for repair (or whatever you feel like otherwise), possibly while sobbing quietly.</p> Moving Onto Mirage 2015-05-20T00:00:00+00:00 For a little while I’ve had this site</a> running as a MirageOS</a> unikernel, shadowing the main site hosted on GitHub</a>. I’ve finally decided to make the switch, as part of moving over to take advantage of Mirage’s DNS and TLS libraries.</p> Following the usual pattern, as previously explained by Amir</a>, Mindy</a> and others, the process is:</p> Construct a static Jekyll</a> site.</li> Write a Travis</a> YAML file to cause Travis</a> to build the unikernel image and commit it back to the deployment repository.</li> Write a Git post-merge</code> hook for the deployment repository, so that the latest unikernel is automatically booted when a merge is detected, i.e., there is a new unikernel image.</li> Write a cron</code> job that periodically polls the deployment repository, pulling any changes.</li> </ul> Building a Jekyll</a> site is well-documented – I did find that I had to tweak my _config.yml</code></a> so as to make sure my local toolchain matched the one used by Github, ensuring consistency between versions of the site. For convenience:</p> make</span> site</span></span></code></pre>Bringing up the network</a></h2> The .travis.yml</code></a> file then specifies the three main targets for the CI test build to carry out: Unix with a standard sockets backed (MIRAGE_BACKEND=unix</code>, MIRAGE_NET=socket</code>) and with the Mirage network stack (MIRAGE_BACKEND=unix</code>, MIRAGE_NET=direct</code>), and with the Xen backend (MIRAGE_BACKEND=xen</code>). For the latter case, we must also specify the static IP configuration to be used (MIRAGE_ADDR</code>, ..._GWS</code>, and ..._MASK</code>). The .travis.sh</code></a> script then calls the standard skeleton .travis-mirage.sh</code></a> script after first building the site content using Jekyll.</p> This tests the three basic combinations of network backend for a Mirage appliance:</p> $</span> make configure.socket build</span></span></code></pre> UNIX/socket</strong> requires no configuration. The network device is configured with the loopback address, 127.0.0.1</code>. Appliances can be run without requiring root</code> privileges, assuming they only bind to non-privileged ports.</li> </ul> $</span> make configure.direct build</span></span></code></pre> UNIX/direct/dhcp</strong> requires no configuration if a DHCP server is running and can respond. The appliance must be run with root</code> privileges to use the new network bridging capability of OSX 10.10, whereupon the DHCP client in the appliance follows the usual protocol.</li> </ul> $</span> make configure.xen build</span> \</span></span> ADDR="46.43.42.137" GWS="46.43.42.129" MASK="255.255.255.128"</span></span></code></pre> Xen</strong> uses the Mirage network stack and expects static configuration of the network device.</li> </ul> Using Travis CI</a></h2> Of course, all that is for local development – for the live site, this is actually all wrapped up using Travis CI</a>. Due to a small pull request waiting on the OCaml Travis CI skeleton scripts</a> and a few Mirage releases currently being readied, this looks a little more complex than it needs to (the FORK_USER</code> and DEV_REMOTE</code> variables shouldn’t need to be specified in the long run) but anyway:</p> language</span>:</span> c</span></span> script</span>:</span> bash -ex .travis.sh</span></span> env</span>:</span></span> matrix</span>:</span></span> -</span> FORK_USER=mor1 DEV_REMOTE=git://github.com/mirage/mirage-dev</span></span> OCAML_VERSION=4.02 MIRAGE_BACKEND=unix MIRAGE_NET=socket</span></span> -</span> FORK_USER=mor1 DEV_REMOTE=git://github.com/mirage/mirage-dev</span></span> OCAML_VERSION=4.02 MIRAGE_BACKEND=unix MIRAGE_NET=direct</span></span> -</span> FORK_USER=mor1 DEV_REMOTE=git://github.com/mirage/mirage-dev</span></span> UPDATE_GCC_BINUTILS=1</span></span> OCAML_VERSION=4.02 MIRAGE_BACKEND=xen</span></span> MIRAGE_ADDR="46.43.42.137" MIRAGE_GWS="46.43.42.129" MIRAGE_MASK="255.255.255.128"</span></span> XENIMG=mortio MIRDIR=_mirage DEPLOY=1</span></span></code></pre> This uses the local .travis-sh</code></a> script to build the three versions of the site, using the Mirage development OPAM repository</a> so as to pick up the latest versions of all the various packages, and updating the Travis gcc</code> and binutils</code> to ensure the stubs for a couple of packages (notably mirage-entropy-xen</code>) build.</p> Next stop: adding TLS and DNS support…</p> Diamonds are a Chap's Best Friend 2015-04-09T00:00:00+00:00 As noted previously</a>, this site is basically a Github</a>-hosted Jekyll</a> site at present, though one that can be built as a Mirage</a> unikernel. Part of the Mirage</a> workflow to publish a new post involves using Travis CI</a> to build and then commit back a new unikernel image. Thus it is currently necessary to run Jekyll</a> in the Travis</a> build scripts, and the dynamism of the Ruby environment meant that this broke (again) recently as one of the github-pages</code> gem’s dependencies now depends on Ruby >= 2.0</code> while the default Rubies on the Travis</a> Ubuntu image for C</code> language builds is 1.8</code> (via Ubuntu packaging) or, if you remove that one, 1.9</code> (via rvm</a>). Read on to find out how to fix this…</p> The fix that currently works for me turns out to be relatively simple: remove all the rubies installed as Ubuntu packages, and then invoke rvm</a> to set the default ruby to something reasonable – in this case, 2.1.</p> ## remove old ubuntu rubies</span></span> sudo</span> apt-get</span> -y</span> remove ruby ruby1.8</span></span> sudo</span> apt-get</span> -y</span> autoremove</span></span> </span> ## use rvm and a modern-ish ruby</span></span> source</span> ~/.rvm/scripts/rvm</span></span> rvm</span> --default</span> use</span> 2.1</span></span> </span> ## check that all worked...</span></span> which</span> ruby</span></span> ruby</span> --version</span></span> </span> ## install jekyll and github-pages</span></span> gem</span> install jekyll</span></span> gem</span> install github-pages</span> --no-rdoc --no-ri</span></span> jekyll</span> -v</span></span></code></pre> And that’s all there is to it – you should now be able to call jekyll</code> in your Travis</a> environment as you’d expect…</p> Taking it to the People People 2015-04-08T00:00:00+00:00 Looks like I get a chance to run my mouth off again :) Upcoming HDI</a> research seminar</a>, organised by Cambridge Big Data</a>/Digital Humanities</a>. In short, details are: 20th April 2015, 14:00–16:00</strong> in S1, Alison Richard Building, West Road, Cambridge</strong>. If you’d like to attend, please do register at http://www.eventbrite.co.uk/e/human-data-interaction-cambridge-big-datadigital-humanities-seminar-tickets-16337148852</a>.</p> And just because pixels are, in some loose sense, nearly free, here’s the abstract from the seminar link above:</p> The increasing generation and collection of personal data has created a complex ecosystem, often collaborative but sometimes combative, around companies and individuals engaging in the use of these data. We propose that the interactions between these agents warrant a new topic of study: Human-Data Interaction (HDI), that sits at the intersection of various disciplines, including computer science, statistics, sociology, psychology and behavioural economics._</p> </blockquote> In this brief presentation I will pose some of the challenges that HDI raises, organised into three core themes of legibility, agency and negotiability. I will also outline some of the technical work we are currently undertaking that attempts to address some of the underlying platform problems. My hope is to elicit discussion of both the HDI framework and the technical solutions we are pursuing, as well as to engage in a broader conversation about the ways we should approach the personal data ecosystem with other interested parties._</p> </blockquote> I Think Someone Else Is Also Following Me... 2015-03-16T00:00:00+00:00 Turns out others were listening too – notably the USA’s DARPA</a>. The recent announcement of the Brandeis</a> programme makes explicit reference to HDI</a> and our website</a>. This has been picked up by GCN</a>, USAToday</a>, NBCNews</a>, Engadget</a> among others. With $60M potentially on the table, I hope that there’ll be many more who get interested in pushing HDI forwards now :)</p> I Think Someone's Following Me... 2015-03-02T00:00:00+00:00 It’s always1</a></sup> nice when someone notices what you’re doing, and I was pleasantly surprised recently to find that someone had indeed been watching.2</a></sup></p> Specifically, it appears that someone at MIT Technology Review has noticed the work of me and some of my pals: our work on HDI</a> and a recent sketch of some follow-on work we’re pursuing around building a personal Databox</a>. Independently of that, the latter also happened to get picked up by John Naughton</a> in the Guardian</a>, which ended up with a pretty active comments thread, including one from Richard Stallman himself…</p> Guess we’d better get on and deliver it now then :)</p> Well, usually. Unless you’re doing something embarrassing like, e.g., dancing. ↩</a></p> </li> Though in fact, given the pattern involved, I suspect it may be my colleague, Dr Hamed Haddadi</a> who’s actually being followed. To whit, the fact that they also picked up on a study</a> of third-party web-tracking from last year. ↩</a></p> </li> </ol> </section> Looping the Loop 2015-02-17T00:00:00+00:00 In a fit of blogging mania, here’s another one literally barely days after the previous one. Maybe I’ll crack this yet.</p> Anyway, this is just a short one with what verges on a Technical Contribution. To whit: I recently sorted out this domain</a> and was having some issues getting some consistency between what dig</code>, Chrome and my domain provider</a> believed to be the correct state. In particular, I was switching over to make the domain properly live rather than simply a 301 Moved Permanently</code> redirect to my old pages at Nottingham.</p> It turns out this was probably mostly Chrome being confused. It seems that it caches 301 Moved Permanently</code> redirects fairly aggressively and the cached entries are not</strong> discarded when you go through the standard mechanisms to clear caches.</p> After a bit of experimentation and browsing, it seems that one way to clear this is to view-source</code> on the page but pass a spurious parameter to defeat the cache. So, to force the browser to fetch http://mort.io</a> properly, all I had to do was view-source:mort.io?spurious=parameter</code>. And lo! All was well.</p> Talks, Old and New 2015-02-14T00:00:00+00:00 Thanks to an invitation from Hitesh</a>, I recently got the chance to revisit my old stomping ground at Microsoft Research Cambridge</a>. Well, I say “old” – in the intervening 7 years, they’ve moved to a rather splendid new building at the other end of Cambridge, just next to the station. (And improved the coffee too, not that it wasn’t pretty good to start with!)</p> Anyway, this was a pleasant chance to catch up with some old colleagues, meet some new ones, and even speak to my most recently graduated Ph.D. student, Dr Ewa Luger</a> – and who’d’ve thought that I’d ever end up supervising someone coming from the discipline of Political Science too!</p> The ostensible reason was to talk about the Homework</a> project – a talk I’ve given a few times now</a> – and to lead from that into discussing some of my current agenda, around Human-Data Interaction</a> and User-Centred Systems</a>. It seemed to go pretty well – a few questions, and some lively discussion over lunch followed.</p> Happily, the fine folk at MSRC</a> recorded it and have made it available, so if you’re really interested, you can take a look – direct your browser of choice to Interacting with Infrastructure: Home Networking and Beyond</a>.</p> I have to say, when I checked that link in the course of writing this, one thing that did come as something of a surprise was to notice that a talk I didn’t remember giving – it was 8 years ago to be fair! – is also available. So if you’ve a burning desire to find out about Measuring and Monitoring Microsoft’s Enterprise Network</a>, it turns out you can do that too. The past haunts us they say. In my case, it seems that the haunting has happened already, and turned my hair mostly white in the process… :)</p> Begin, Again! 2015-01-15T00:00:00+00:00 Specifically, I’ve left Horizon</a> and the School of Computer Science</a> at the University of Nottingham</a> to (re-)join the Cambridge University</a> Computer Laboratory</a>. In celebration, and frankly because it was long overdue anyway, I’ve reworked my website. What do you think?</p> For the curious, or the technically inclined, the site now uses ZURB Foundation</a> 5.5.0 (the current downloadable release as of yesterday), with some slightly customised CSS. The site itself is largely written in Markdown</a> and currently generated using Jekyll</a> to be hosted on Github</a>.</p> It’s actually gone through an interim phase where it was parsed by the OCaml OMD</a> parser before being crunched into a Mirage KV_RO</a> filesystem which is then compiled into a type-safe, self-contained web appliance that serves these pages and no other using the OCaml Cowabloga</a>, COW</a> and CoHTTP</a> libraries. This could either be run as a POSIX binary</a> or a self-contained Xen VM</a> depending on what I felt like. Neat eh? (And for the sceptical among you, yes, a thing can</em> be neat and yet appear curiously over-engineered at the same time… :)</p> For the time being however, I’m using it as an excuse to think about what I might do to better support site generation like this in Cowabloga</a> so that I can more seamlessly switch between Jekyll</a> and Mirage</a>.</p> Longshaw Lodge to Higger Tor via Carl Wark, and back via Burbage Moor 2014-10-05T00:00:00+00:00 A family walk in the Peak District on a rather fine mid-autumn day after what looks like a relatively dry summer.… </p> Monsal Hotel to Monsal Head Viaduct and back via the River Wye 2014-06-08T00:00:00+00:00 Another family walk in the Peak District.… </p> A 21st Century IDE 2013-10-13T00:00:00+00:00 I finally decided to sit down and get the shiny new merlin</a> mode for OCaml working with my emacs configuration. Basically, really rather simple in the end although (in the usual fashion!) I did end up spending considerable time tweaking various other customisations…</p> Most of the information below is based on the following sources:</p> http://github.com/def-lkb/merlin#emacs-interface</a></li> http://zheng.li/buzzlogs-ocaml/2013/08/23/irc.html</a></li> http://www.ocamlpro.com/blog/2013/03/18/monthly-03.html</a></li> </ul> Before we begin, install merlin</code>:</p> $ opam install merlin</span></span></code></pre> The complete commit</a> change is in my github</a> account (combined with a large cleanup of various other aborted OCaml configurations). Breaking it down a bit, first setup some paths: where to find ocp-indent</code>, merlin.el</code> for merlin-mode</code>, and the ocamlmerlin</code> command itself. Note that this relies on the current state of opam</code>, so when you start emacs</code> be sure to have selected the opam</code> compiler-switch that you installed the merlin</code> package into, above.</p> ;; ocp-indent</span></span> (load-file (concat</span></span> (substring (shell-command-to-string </span>"opam config var prefix"</span>) </span>0 -1</span>)</span></span> "/share/typerex/ocp-indent/ocp-indent.el"</span></span> ))</span></span> </span> ;; merlin-mode</span></span> (</span>push</span> (concat</span></span> (substring (shell-command-to-string </span>"opam config var share"</span>) </span>0 -1</span>)</span></span> "/emacs/site-lisp"</span></span> )</span></span> load-path)</span></span> </span> (</span>setq</span> merlin-command</span></span> (concat</span></span> (substring (shell-command-to-string </span>"opam config var bin"</span>) </span>0 -1</span>)</span></span> "/ocamlmerlin"</span></span> ))</span></span> (autoload </span>'</span>merlin-mode </span>"merlin" "Merlin mode"</span> t</span>)</span></span></code></pre> Now the meat: when we select tuareg-mode</code>, use ocp-indent</code> to indent lines, turn on merlin</code> auto-complete, and finally set a couple of local key bindings so that I can fix up merlin</code> to not conflict with my now-neurologically-hardwired navigation keys.</p> (add-hook </span>'</span>tuareg-mode-hook</span></span> '</span>(</span>lambda</span> ()</span></span> (merlin-mode)</span></span> (</span>setq</span> indent-line-function </span>'</span>ocp-indent-line)</span></span> (</span>setq</span> merlin-use-auto-complete-mode </span>t</span>)</span></span> (local-set-key (kbd </span>"C-S-<up>"</span>) </span>'</span>merlin-type-enclosing-go-up)</span></span> (local-set-key (kbd </span>"C-S-<down>"</span>) </span>'</span>merlin-type-enclosing-go-down)</span></span> ))</span></span></code></pre> Finally, do the usual to use tuareg-mode</code> for OCaml/F# editing.</p> (push'(</span>"</span>\\</span>.ml[iylp]?"</span> .</span> tuareg-mode) auto-mode-alist)</span></span> (</span>push</span> '</span>(</span>"</span>\\</span>.fs[ix]?"</span> .</span> tuareg-mode) auto-mode-alist)</span></span></code></pre> And that’s it!</p>

mort’s mythopœia

Updating Cambridge VPN configuration

Sabbatical Diary: Week 18

Sabbatical Diary: Weeks 16/17 / Months 2---4

Some Remarks on the Risks of Lawful Access

Restarting audio

Sabbatical Diary: Week 15

Extracting ramblings

Pinning packages

Sabbatical Diary: Week 14

Updating `eduroam`

Sabbatical Diary: Week 13

Sabbatical Diary: Weeks 11---12 / Month 3

Re-connecting Wi-Fi

Bowlees & High Force

Sabbatical Diary: Week 10

Displaying Wi-Fi configuration

Sabbatical Diary: Weeks 6---9 / Month 2

Sabbatical Diary: Week 5

Back to my Roots

Sabbatical Diary: Week 4 / Month 1

E(pi)glot-tal start

Sabbatical Diary: Week 3

Tree-sitting Emacs

Sabbatical Diary: Week 2

Sabbatical Diary: Weeks 0 & 1

All Change!

A wander around Vienna

Systematising the obvious

Orating honours

Resizing swap partition on NixOS

Re-key-remapping

Whither AI?

Artificial Intelligence, but at what cost?

Rusty Nu POSIX

Coping and Capping

Rewarding influence

Not one but two accepted papers and a Ph.D.

`nmcli` connection to internal WiFi

`just` OCaml

Reverse `find`

PhD viva, UK style

Fontsizing in LaTeX

`just` LaTeX it!

Introducing `bibtox`

Retreat from TUM

Workshopping, across the universe

Part II Projects

Post-COVID19 TPC experience

Stop, Start

Discovering `restic`

Debugging Electron apps

Reinstalling `maestral`

NixOS channels

OneDrive and NixOS

WSL2 / Windows 11

OCaml operators

Software folklore

Hosting HotCRP

Tarballing some files

Data visualisation links

Renovating an old media PC

Discord discord

Setting up a Macbook

Keyboard</a></h2> Set the touchbar to expanded</li> Turn off various emacs conflicts (e.g., S-M-6</code>)</li>

Falsehoods

Wipeout

Back to the Future

Quelle dommage

Spring Loading Karabiner in the Autumn

Platforms, Packaging, Progress— Addendum

Platforms, Packaging, Progress

Redwood Trail, Big Basin Redwood Park, USA

May Lake, Yosemite National Park, USA

ARMing LinuxKit

Docker Docker

Bootstrapping Docker for ARM64 (aka AARCH64)</a></h1> Basic process is:</p> bootstrap ARM64 go</code> toolchain on x86, and install</li> build ARM64 go1.7.5</code> toolchain needed for docker</code> build</li> bootstrap ARM64 docker</code> v1.10.3</li>

Lab GitLab