Excessive chaos produces a desire for order. The desire for order is both good and evil: cooperation is good, dictatorship is evil. Everything not trivial is complex.

Excessive order produces a desire for chaos. The desire for chaos is both good and evil: self-determinism is good, denying responsibility is evil. Nothing is purely one or the other.

Improving things is not futile, but it can be difficult and requires maintenance and debugging. It is impossible to predict every consequence, but quite possible to predict the likely consequences of most actions. Nothing exists in a vacuum.

Ignoring one or more of the above precepts is a common failure for philosophers, economists, engineers and politicians.

-30-

Whenever a system attempts to predict what a user of the system will want or do next or means, the failure case generates the most noticeable immediate problems. The author(s) will notice them and attempt to fix those problems.

The success case may or may not generate a long-term problem.

Systems in which this applies: autocorrect. Grammar checkers. LLMs. CPU branch predictors. There are others.

-30-

An extract from George O. Smith’s short story QRM INTERPLANETARY (1942), the first of his works about a space station set up to relay radio messages between Earth, Mars and Venus:

On Monday, Francis Burbank sent around a communique removing the option of free messages for the personnel. On Tuesday, he remanded the years-long custom of permitting the supply ships to carry, free, packages from friends at home. On Wednesday, Burbank decided that there should be a curfew on the one and only beer emporium. “Curfew” was a revision made after he found that complete curtailing of all alcoholic beverages might easily lead to a more moral problem; there being little enough to do with one’s spare time. On Thursday, he set up a stiff-necked staff of censors for the moving picture house. On Friday, he put a tax on cigarettes and candy. On Saturday, he installed time clocks in all the laboratories and professional offices, where previous to his coming, men had come for work a half-hour late and worked an hour overtime at night.

and about a month later it escalates to this conversation:

People are very fond of the Shirley argument. The Shirley argument goes like this:

“Surely the people who implement this law will do common-sense things to avoid ridiculous results!”

The Shirley argument is always wrong, usually because of non-systematic yet surprisingly pervasive discrimination… and sometimes because of systematic and intentional discrimination. And occasionally because people aren’t perfect. But it’s always wrong.

Laws can be incredibly pedantic: if you’re not sure, go read some. A law which seems to inadvertently leave loopholes is not a well-written law; anything which you can immediately spot is deliberate.

And don’t call me Shirley.

-30-

Given that:

• No copyright can be claimed on work created by an LLM
• Many open source projects require contributors to sign a release of some form - a certificate of origin or license agreement - that states that the contribution is copyrighted to the contributor and can be transferred to the project
• Many – all? – commercial software projects claim that their work is copyrighted.

Therefore:

• No LLM-produced code or documentation can be used by such an open source project or such a commercial software project without invalidating their copyright.

Notes:

I have a morning routine which would have seemed awfully complicated to the teenaged me, and is probably ridiculously streamlined to the me reading this entry in 2036. One step of it involves recording a number in a spreadsheet.

It has been over 1111 days since we switched from Oracle’s database to Postgresql at work. I have been happy about the switch for every single one of those days. No regrets at all.

Changing your core database is a huge proposition, especially when the one you have is not broken. It’s also a huge amount of work. [Workplace] has been built on Oracle for more than twenty years. Why did we do it?

(This entry is now on its third revision. Not much has changed.)

This is going to be a big one. Here follows a [hopefully factual] comparison and [hopefully useful] subjective review of N terminal emulators. The following three constraints will be in effect:

No, this isn’t an usurper hijacking the blog.

It just occurred to me that this was a good place to put it.

I am good at making ice cream.

I am also rather experimental, so not everything is a big hit.

These flavors work very well:

• multiple vanillas
• Irish whisky
• dark chocolate
• chocolate orange
• chocolate whisky
• Mexican chocolate
• multispicecream (allspice, nutmeg, cinnamon, cayenne)
• peanut butter
• peanut butter caramel
• peanut butter with cayenne
• peach
• peach with cayenne
• plum sorbet
• coconut
• cherry

Nutmeg is apparently popular in Granada, but it didn’t work for me or anyone I tried it on. Could have been me.

Ripe strawberries have better things to be doing.

-30-

This is how I’m dealing with music playback in the house as of July 2025. My goals are reasonably high-quality music at a long-term reasonable price, while avoiding paid streaming services.

Since you’re interested in not breaking things, you should be running Debian Stable.

Debian makes several kinds of package changes available:

Every complex problem is a dependency management problem.

The only robust solutions are the simplest things which can work, created by people who fully understand the problem and have to work with the system regularly.

But the fastest temporary fix for any specific problem is to add a complex layer without fixing the underlying problem.

And every layer reduces your ability to understand the problem.

Every complex solution is a dependency management problem.

-30-

I had assumed that when certbot does an http authentication of a domain name as part of getting Let’s Encrypt to issue or renew a cert, it would do so by touching the appropriately named file in http://domain.name.tld/.well-known/acme-challenge, waiting for the challenge to resolve, and then deleting it.

And sure, that’s what it does with --webroot. And I assumed that if you passed it --nginx, certbot would use that to look for the right config to parse to get the right domain to directory mapping, and then reload nginx afterwards.

It does not do that.

Instead, it inserts new config into nginx to just return a 200 success code for the challenge name as though it were a 0-length file on disk. Nothing hits disk in the .well-known directory.

It turns out that this can backfire in a number of very hard to debug ways.

So, if you are having an authentication problem with certbot, try passing --webroot and the appropriate -w directory.

And if you’re writing some bit of utility code, howzabout not being so clever. Do the expected things.

When 3 of 4 disks in a ZFS pool die at the same time, it is not the disks.

It’s the controller that is feeding them.

(Note for later: move a disk to a different controller. This could have been survived completely if half of each mirror were on different controllers.)

What do you call a household with a backup SAS/SATA controller stashed in the box marked “more disk stuff”?

Mine.

-30-