heads

What I realized after 25 years is that configuration comes in three parts:

1/ the defaults, either built in or read from /etc;

2/ my defaults, included in each file (or with ssh, at the bottom) with that particular config’s native version of #include; and

3/ local specifics that are rarely if ever used anywhere else, or trivially short as to be copy-paste-able.

Almost everything I want to customize goes into (2) so I wrote a single Python function that manages a block at the top (or with ssh, at the bottom) of each config file:

  # BEGIN my foo stuff
  include = /my/repo/foo/config
  # END my foo stuff

That way foo starts out with (1) the system defaults; then adds (2) my personal foo defaults as defined in a working copy at /my/repo; (3) anything else I insert in the file after that which isn’t centrally managed and that’s ok.

I haven’t ever needed anything more complicated. I do not have any work specific configs that I need to gate. I no longer have to manage different configs based on whether I am using Debian, Debian (old), Debian (very old), SunOS (very very old), or AIX (very very very old) because those days are behind me.

If you do still need to manage slightly different but ethereally different configs on different hosts then I’m sorry to hear that. Rationalising my computing life so that I use the latest version of some Linux distribution everywhere has been very helpful!

North America has lots of land and oil. Timber (Canada), plastics, and corn fertilised by nitrates that were made using fossil fuel energy. Corn probably doesn’t ship in containers but corn-fed beef and poultry do?

In the sense that violence is coercion they have a lot of purely electronic means of persuasion:

  update account
  set balance = 0
  where user = ‘steve1977’;

We used to have an ecosystem like this. In our case it reflected an entrenched set of divisions between warring teams. In some ways it may have then enhanced those positions and we still bear a few of the scars today.

A lot of the old guard have left the company though and our main product moved from four repos to just one. The threat from the legal team to have enforced OWNERS files — essentially replicating the divisive politics of the old repos but in the monorepo — thankfully withered on the vine. We still audit what goes into each release but it’s no longer part of any active permissions thing. We trust our developers but verify, for legal reasons, that nothing went wrong.

You either want one engineering team to act in unison behind your company’s mission, or you want to live a divisive narrative that you are actually multiple teams “working” together with none of the advantages of living under one roof and all the disadvantages of hard repository boundaries crisscrossing your intellectual property.

So many factors threaten to curdle your team dynamic: multiple offices, multiple floors, work from home hermits, bad management, etc. It’s simply org entropy and it takes much effort to keep the weeds out of the garden. Multiple repositories is one less bullet you can keep out of your feet while fighting all the other battles that threaten to turn your team from 1990s Sun Microsystems into 2010 Sun Microsystems.

Put wings on it and go up sideways. It solves the thrust/weight issue the same way a hill climber does: instead of going directly up they zig zag.

You need the lateral velocity anyway, for orbit.

And at the command line: headers to stderr, body to stdout.

Without it you get this behavior:

  > heights | sort -k2 -nr
  George  186
  Wally   173
  NAME    HEIGHT

But if the headers bypass stdout then you get this instead:

  > heights | sort -k2 -nr
  NAME    HEIGHT
  George  186
  Wally   173

I’m sure this is real but just for fun I’m entertaining the idea that someone is trolling — either with a fun dialog title or the app name.

I worked at a company in the late 2000s that moved to an office with automatic urinals. These were fairly novel at the time and had a matte black plastic unit with a flush handle that also had a shiny window made of dark, IR transparent plastic. It was clearly some kind of proximity sensor for the autoflush but some joker made an official “do not touch the cameras” sign that wound a few people up.

I feel like my memory (and that article) aren’t very clear on this, but Backbone in spirit felt quite different to the series of high bandwidth towers of which Telecom Tower was one part.

Regular lets-call-grandma telecommunications would be handled by your well know, friendly, cuddly round towers like Telecom Tower and Stokenchurch just next to the Chiltern cutting on the M40.

Sinister everyone-has-died communications would be handled by the mysterious, dangerous looking towers like Highgate and Kelvedon Hatch. These towers resemble high tension power lines and growl at you if you look at them for too long.

This site has a lot more detail on how the two ended up as one, over the years:

https://www.subbrit.org.uk/features/backbone-microwave-relay...

Parliament is all the players from all 12 teams including all those on the bench who never even played.

Government is the members of the winning team who were on the field when the whistle blew. They get to lift the trophy.

In real competitions the subs get to lift the trophy too so maybe a better analogy is that the UK Parliament is 12 teams, all their subs, the coaching staff, club management, and the people who run the official fan club. A subset of these from the winning club get to lift the “I am the government now” trophy but not all of them do.

Oh, and of course while the Commons are playing [association] football, the Lords are playing charity match celebrity cricket, and yet some of them (e.g David Lord Cameron) get their hands on the FA Cup too.

In the US, by contrast, it’s more like the guy who owns the team is the one who lifts the trophy. They do so by winning a “whose team do you like the most?” competition, all while a separate match with actual players is going on, none of whom will win the executive trophy.

There are also only two official USA Sports League teams but they split into cliques that play together behind the scenes. Kind of like the midfield caucus doing a deal with the forward caucus to give them the ball in exchange for extra oranges at half time.

What a fun read! Your “how to get shot in the airport” caption nearly made me spit out my coffee.

I feel like there are metric lurking in here somewhere. Something to do with the following concepts:

(1) the number of documented atomic changes to the codebase, be these merges or pull-requests or changesets or patches — the thing with a title and description that was code reviewed;

(2) the number of tasks in your task management system that might be bugs or feature requests — these are large and must be implemented in pieces, with testing and reviewing of assumptions along the way; and

(3) the time period on which you are required to be held accountable for actually finishing stuff.

I do maybe 30 commits for a big task and will spend six weeks on it. There might be small bugs along the way that get closed out with a 1:1 mapping between task and commit. The majority of the work is a giant slowly but surely moving set of iterations towards a top level goal. The cadence is to review on a quarterly basis.

Not so in the past though. I’ve had to do one commit per task. Multiple tasks in a big bushy tree of verbiage that overlaps too much with the actual commits (the meaningful bits!) and all done on two week cycles. It felt exhausting and piecemeal, as if software engineers were a fungible resource there to further marketing goals instead of innovate the next big things. Nowadays I demand to work for a true technology company, not a marketing company that sells tech.

Speechmatics.com in London and Cambridge, UK. We build audio and language models that perform the most accurate speech recognition available.

https://www.speechmatics.com/company/careers/roles

So negative! Where I work this tool is called “new-hires”. It uses a restricted read-only API key to our third-party people tool. It was given to me by our People Director. Sometimes there are lines beginning with - but the tool is named for the lines beginning with +.

new-hires is built on top of the “people” python module / cli in our monorepo. That tool is so much more useful than just a way of diffing the org chart. Who is in what team, where are they, are they working today, is it time to celebrate their anniversary, etc. It also follows what I coin the “ZFS litmus test” for good CLI tools by providing -pH for parseable, headerless output.

Treasure such places indeed.

Are there any examples of this kind of assertion turning out to be incorrect? I’m happy to be told this kind of libation analysis is (ahem) rock solid but it’s not easy to tell from the article.

My understanding of the article and the plot is that the authors modelled how the moon ought to wobble if it has an ocean versus it does not have an ocean, and if it did then how does it wobble with either a thick crust or thin crust on top. The actual, observed wobbling is in the “has an ocean” part of the plot therefore, assuming the model is correct and there isn’t some other cause of wobble, it probably has an ocean. This is the only evidence so far though — other than this data there is “no hint of an ocean underneath”.

It would also wobble if it had a moon-mouse dance party at each pole. Modelling for different sized discos and different tempo dance music, we see that the observed wobbling is consistent with 4 billion mouse / 185bpm portion of the plot.

I could also plot the hypothesised size of the kraken mouth which swallows ships as they vanish over the Earth’s horizon. The plot shows the rates at which the ship sinks from view based on kraken mouth size. The observed sink rate is consistent with a 12m wide kraken mouth which is a cool result except that ships sink over the horizon because the Earth is round. There are no krakens!

I’m British with, funnily enough, a Duke’s County cricket ball on my desk.

My nephew is Californian and I gave him one for his sixth birthday. Mine felt a bit lighter so we did an experiment over Zoom together to measure the density. He’s certainly a bit young for that — “it’s the same size but yours is heavier” — but there’s no harm in influencing them from an early age.

Of relevance to your comment: I made a point of doing it in grams and millimetres.

Wasn’t one of the great puzzles of the 20th century exactly this — will the universe collapse back into a singularity or not?

The numbers are big but they don’t seem “big big” like Sagittarius A* is big or the distance to Andromeda is big.

If you’re doing 2e8 m/s, your ship is long and thin with a 1m3 nose cone, and space has 100 baseballs per m3 then you’re being hit by 2e10 baseballs a second. How do I get an idea for what 20 billion baseballs (2TJ) feels like?

Apparently bullets are about an order of magnitude more energetic than baseballs (800J vs 80J) so I guess I could try to build my intuition based on being shot ~2 billion times a second instead. A kiloton of TNT is 4GJ, so it’s also like a 500 kiloton bomb going off every second.

Dropping 5e5kg of rock into the worlds biggest dumpster truck at 10m/s yields 25MJ of chaos so it’s also like a parking lot of 80,000 of those being filled with a continuous stream of rubble. That’s probably the best analogy given that we’re talking about machinery — your spaceship needs to have the build resilience of tens of thousands of dumpster trucks but condensed into the cross sectional area of a dinner table.

Configuration languages are an interface to the API presented by the software you are using. There can be a hard boundary between the two — a C codebase for a web server and the XML-like file that configured it — or there can be an imperceptible transition between the two where the configuration is just another module in the code base.

As one of the authors working in a 2000+ module Python codebase, nothing gives me greater pleasure than to drive one part of the codebase by creating another module. Nothing gives me greater sadness than to be forced to interact with something through a YAML config file, command line flags, or launching GitLab pipelines. All three of those boundaries interrupt the powerful electromagnetic force fields that pervade the system: type checking, linting, and symbol finding (IDE integration). In time, I vow to destroy every last one of these non-code boundaries. Another unwanted-boundary demon on the exorcism list is polyrepos.

There once was a time when the world worked like this but instead of Python it was C. It wasn’t as rich as the dynamic world we have today — I’m certain I don’t want to go back to configuring my softest by recompiling it — but it did have a lot of the advantages of working in one language environment across all the things.

To tack one additional problem onto your excellent list: the commit message is usually only the start of a conversation about why a change should be made. The rest of that discussion is whether it meets the bar and what needs to be adjusted before it can land on the collaborative trunk. Done well, that is valuable reading.

Git was designed with the distributed viewpoint. A commit message, as written by the author, is necessarily correct: I’ve decided this is right, and it’s on you to decide if you want to merge it into your history too.

In our current systems we usually have a URL in the commit message that links to the actual story behind the commit — the discussion on the pull request, merge request, or code review. I rarely see the results of these discussions being amended into the commit message. If the repo lives forever but the database behind the code review tool gets toasted then something just as important is lost forever.

(I come from a background of one idea equals one amended, fast forwarded commit to master. It’s possible other people rely on branch history to reflect the evolution of ideas and how they go from a request for review to approved code. In my experience branch histories tend to have very low quality commit messages and even then they only show one side of the conversation — the author’s responses to their reviewer’s and their own critiques.)

Thanks, that’s an important detail. I guess it boils down to how much private wealth is too much, and what if anything should we do about it?

An argument on Harvard’s side is that, in the grander scheme of things, they are part of a process that generates phenomenal tax revenue. The state takes their 40% cut off the top of every $150k CS and IB graduate salary after all. Is that a good enough benefit to let us tolerate Harvard’s pile of gold?

There is a related argument that if society is to be the ultimate recipient of the majority of benefits of a university system, then it should probably be state funded. Just like roads (public) trains (ish) electricity (not any more) and food (private but subsidised).

On the one hand money in an endowment has presumably already been taxed at least once. Taxing it again is double taxation and that’s bad because once you allow money to be taxed twice, you have to be ok with it being taxed many times over — O(1) to O(n) if you will. The moral position of government would be that taxing the outcome of growth and wealth creation is fine, but retaxing existing money is the government breaking its covenant with the people.

On the other hand I would accept the government’s position if it were to say aloud that there goal is for all large pools of money to be reclaimed by society. Double taxing wealth is then ok because it will, eventually, redistribute the wealth. In such a moral code, a $51bn pile of gold is quite an aberration.

It is classic right versus left: natural hierarchy versus we’re all the same.

This a product update we’ve been working on. Our multi-headed model is accurate in 49 languages and the full pipeline has really good normalisation (“$10”), abbreviations, and punctuation too. It makes a big difference when you’re feeding this, say, to an LLM in real time.

(Should we add oxford_comma as a config option?)

That is brilliant, thank you so much for sharing. The galaxies’ enormous size is truly jaw dropping. I had absolutely no idea they were this large, in the night sky.

Have you ever seen them live, though an image intensifier / night vision system? I feel there would be something profound about seeing a galaxy with my own (albeit enhanced) eyes.

The gaming paragraph seems a little too underwhelming. Steam Deck is a perfect implementation of a Linux computer. I adore using it just to glow in the polish and expertise with which it has been assembled. Kudos, Valve. I don’t even have a Windows install anymore.

Tilt it? That is to say, to some extent, could you repurpose gravity to give a sense of acceleration?

Well if you were working in “endless cubicle farms” and “wasting your lives in traffic” then, yes, removing those two things from your life is a smart thing to do. Being socially isolated from the other members of your team and only having VC is, in my mind, just as poor of an alternative.

Imagine instead if you and (the majority of) your colleagues came to work by cycling for 20 minutes through green pastures, parks, under willow trees along the river, and through the centre of one of the most famous historic towns in England. That’s what we do here in Cambridge.

We also do not operate a cubicle farm. Our office is spacious, filled with breakout spaces, relaxation spots, a library (aka quiet room), VC equipped rooms of all sizes, with a mini- and a maxi-kitchen and free snacks and meals.

We come and go as we please with two days a week all together and frequently other in-office days for sub teams too, agreed by consensus not fiat. I often work at home in the morning or go home after lunch for the afternoon if I need uninterrupted focus / maker time. We have a personal IT budget for our home offices so I don’t need to cycle with my laptop (which btw is an Ubuntu Dell XPS, because our IT team know what they’re doing — you can have a MacBook if you insist.) Cambridge has gigabit fibre which is handy when pushing and pulling models from the A100 machines in our remote data centre.

We still allow people to be as remote as they want and we also have a satellite office in London. It is rare though for the core eng team to not see each other in person for more than five days on the trot. It’s just so much more productive to build and plan stuff in person around a whiteboard, with pair programming, or in a war room.

Cool idea! Something about this makes me think about chaining moves, like function calls. Is there a version of chess where you get to keep moving if you take a piece?

It’s always baffled me how the regard that the average Englishman holds for itinerant folk is wrapped in a big if statement:

  if (boat)
     … # All good
  else:
     raise TravellerAlert()