Globally well-known chatbots displayed a drastic failure to understand the parties running in the Hungarian election.

Given that ChatGPT hid the main opposition to Orbán, I have to worry that this was intentionally set up to help reelect him.

Congress Should Reject Bill That Would Block a Federal Workplace Heat Standard. Republicans are trying to do just that.

Cutting down the Amazon forest for raising cattle has massively spread screwworm, as well as perhaps dooming what remains of the forest.

The deportation thugs have adopted a dress code and require wearing of badges.

This is a change for the better, but can't overcome what's wrong with the mission they have been directed to carry out.

Italian thugs killed immigrant Abderrahim Fakir in the process of "restraining" him after he became somehow upset.

This has stimulated large protests, to which right-wing hate leaders respond by saying it is unthinkable to doubt the rightness of the thugs' actions.

Many young people are giving up hope on close relationships with other people and turning to chatbots for an imitation or substitute.

Putin and Musk are organizing to destabilize Britain by supporting the right-wing extra-extremist politician, Yaxley-Lenin.

I can suggest one possible countermeasure: prohibit antisocial media platforms from operating a recommendation engine. Ex-Twitter would have less power without that, and so would Tiktok and others.

A henchman says that deportation thugs will enable body cameras when stopping a car.

That is a step forward — it should apply to their other dealings with the public.

Robert Reich: *No one can be honest in a regime based on big lies. As the poet, philosopher, and statesman Václav Havel put it, "If the main pillar of the system is living a lie, then it is not surprising that the fundamental threat to it is living in truth." So a choice must be made — personal integrity or loyalty to [the corrupter].*

Many countries that are more or less democratic are suffering from a flood of political violence against officials.

While you (yes, you! no, not you, the one behind you) have been sweltering in the heatwaves of the northern hemispheres (Assisted-by: AI), I've been busy adding graphics tablet support to libei. This is scheduled for the soon to be released libei 1.7.0.

The initial work was done by Jason Gerecke and Josh Dickens from Wacom, I've been extending, polishing and testing it for the last few weeks.

Also, upfront: this only covers the stylus part of a tablet, we do not yet have an implementation for the "pad" part (the buttons, dials, rings, strips).

libei is, of course, the library for Emulated Input, a good-enough transport layer for sending logical input events between processes. We're already using libei as part of the XDG Portal Remote Desktop and Input Capture portals where we've been busy hurtling key and pointer events between the participating parties (and soon gesture events and text).

In the next release of libei, we will now also have "ei stylus" capabilities, i.e. the ability to send tablet stylus events. Getting pointer, keyboard and touch events supported was a long undertaking, everything was new and shiny and needed to be added everywhere in the stack. Now that all this is in place, scuffed and scratched, adding tablet events will be quite simple.

The ei stylus interface

Here's a short outline of how libei handles tablet events because it is, of course, different to how libinput handles them. Logical events are much nicer after all than physical hardware events.

First: we have a new interface: "ei_stylus". An EIS implementation (e.g. your compositor) may provide you, the libei client, with a device that supports this interface and one or more associated regions (typically representing the available screen areas). Typically this will be a separate device to the pointer devices or the keyboard devices but it's not a requirement. The ei_stylus interface comes with a bunch of capabilities you'd expect from a stylus (tilt, pressure, distance, ...) that you can selectively enable to emulate the stylus you want to. So basically, EIS will say "here's a stylus device, I support pressure, tilt, rotation, ..." and then the libei client says "This stylus should have pressure and tilt but nothing else". And then you do the normal thing: send proximity events, send tip down/up events, send data for the various capabilities you've enabled.

Happily for the EIS implementation, libei forces the client to take the guesswork out of everything: if you select the pressure capability, you must send a pressure value when coming into proximity. Where libei is used to forward data from a physical stylus (e.g. via some remoting protocol) it is up to the client to deal with firmware bugs that e.g. won't send data until a few frames in.

Note that there is no "tablet" anywhere. The tablet is represented by the region that the device may interact with. So in some ways every tablet is an on-screen tablet (which makes sense since we have logical events).

Multiple styli

The only quirky thing is how to request multiple styli[1]: libei 1.5.0 has added a "request device" request that allows a client to say "hey, EIS, I want a new device with capabilities pointer, keyboard, ...". And, if you've been a nice client, minding your own business, the EIS implementation may just create such a device for you.

So for the case of multiple styli: if the default stylus (if any) isn't good enough, you can now tell EIS that you want a(nother) device with stylus capability, configure the stylus capabilities once the device shows up and voila, you now have a normal pen, an art pen and maybe even an airbrush represented as logical device in libei. And since they're all separate devices in the protocol, they can be individually tracked and used, much like libinput tracks individual styli.

[1] For the "lots" of users that actually use multiple styli...

/me gestures vaguely at everything

Oh, hey, this works now? Great!

I'm not going to go too deeply into how pinch, swipe and hold gestures work, suffice to say we've had those in libinput (for touchpads) for years now so compositors and toolkits should already support those. And since libei and libinput have vaguely equivalent API layers integrating gestures for libei devices in compositors should be fairly straightforward.

The plumbing layers in the portals exist already too, so adding gestures to libei means that - once the compositors support it - we can have gestures support in remote desktop and input capture implementations without needing to update anything else. Hooray! Join in with me. Hooray! Louder! HOORAY!

For testing I had a (vibe-coded and thus immediately abandoned once testing was complete) gesturemouse utility which translates input events from a mouse into gesture events (depending which button is down). But don't let my lack of be a limit to your imagination, I'm sure you can come up with good use-cases for this.

If you've been paying attention (and I know you have, because it'd be embarrassing for you if you didn't) you'd have noticed that libei 1.6 (May 2026) added support for keysym and text events.

libei sends logical events between a libei client and an EIS implementation (typically: a Wayland compositor) but the keyboard interface it had was designed like real keyboards: key codes together with an (XKB) key map. You press one key, the keymap decides what that key means on the compositor side and off we go. This is easy but not always useful.

As of 1.6.0 libei now also supports an "ei_text" interface. A compositor may choose to provide you[0] with a device that supports this interface and that gives you two really nice opportunities.

First, you can now send a key sym. Instead of sending the KEY_Q key code and hoping it actually translates to 'q' (and if there's e.g. a frenchman^Wfrenchperson lurking behind the keyboard it may mean 'a'), you can now send 'q' as actual keysym. Or 'Q' instead of sending shift+q and hoping for no french influence in the process. It becomes the EIS implementation's job to handle that keysym - if it's a shortcut it may handle it directly, otherwise it may pass it on via Wayland to an application[1]. This centralises the keysym to keycode handling in the EIS implementation which is a pain for compositor authors (though they likely have that code already for e.g. RDP support) but reduces the variety of differently-wrong implementations in clients and of course makes it so much simpler to write clients.

Second, a client can send UTF-8 text to the compositor. So instead of emulating shift, keycodes, etc. you can literally send "Hello World" and expect the EIS implementation to pass that one. Again, makes a bunch of utilities a lot simpler to write and I mostly leave it up to your imagination to figure out what to do with that.

Notably for both cases: libei is about logical events that have a specific meaning that do not need further interpretation. If a client sends 'Q' that means it is supposed to be an uppercase Q. Sending keysym Shift_L and Q makes little sense. And for the utf8 text events: how the text comes to be matters doesn't matter for libei so you may use an IM to make up the text to begin with and send it, once committed, to EIS. It's not for sending partial strings.

As mentioned in the previous post: the plumbing for this is already in place so both clients and compositors can add support for this new interface without having to bother the rest of the stack (e.g. portals). So, hooray I guess.

The text/keysym support is relatively recent so expect this to hit the next compositor version (or the one after that).

[0]: the EIS implementation decides which devices are available and arguing about this is even less useful than arguing with a world cup ref
[1]: after converting it to a key code with possible keymap changes... but hey, such is life

Turns out it's been years since I've talked about eggs, so let's change this. libei is, of course, the library for Emulated Input[1].

This post is mostly a refresher because it's been so long and a short summary of some of the work we've done so far, in preparation for some more posts that come soon.

libei is a transport layer for logical input events, unlike libinput which is a hardware abstraction layer. In libinput's case the device's firmare/kernel pass events that are somewhere on the sanity spectrum, libinput tries to make sense of those and then we convert those to logical events to be consumed by the next layer (typically the Wayland compositor or Xorg). This is how e.g. "touch down at position x1/y1, touch up at position x1/y2" is converted into a button click event if touchpad tapping is enabled. Or maybe into nothing if we find it was an accidental palm touch.

libei works purely on the logical level - you as the libei client pass logical events to the EIS (Emulated Input Server) implementation (typically the compositor). No guesswork, you say button click, EIS gets a button click. libei supports a "sender" and "receiver" mode, depending on whether events are sent to the EIS implementation (input emulation) or receive from the EIS implementation (input capture). libei is designed for the Wayland stack but there are zero requirements for Wayland on either the client or the EIS implementation.

Core to libei's design is that the EIS implementation is in control of virtually everything, it decides which devices are available to the client, when those devices can send events, etc. Much like the compositor is in charge when it comes to physical devices - if a compositor decides a physical device doesn't exist, a Wayland client cannot get events from it.

Since the original proposal (again, [1]!) we've been busy bees and libei is now a part of the XDG Remote Desktop portal and the XDG Input Capture (both since version 1.17, mid 2023). In both cases the portal is for the negotiation and initial agreement of what should happen, libei is then used as the transport layer between the two processes [2].

More recently we also added session persistence support so you don't have to allow access on every connecton. Much of the work enabling this was done by Jonas Ådahl, it is now in the portals since version 1.21.0 and should be in the major compositors in the current or next versions.

Plumbing the Pipes

Getting all this into place was a huge amount of work across several pieces of the stack. This isn't exciting in the same way as laying plumbing pipes isn't particularly exciting but much like regular plumbing: once it's in place you can change your diet without severely impacting everyone again. Try get that analogy out of your head now. You're welcome.

In libei's case this means three things:

• if you have a client that uses the XDG portals to send/receive events they will now work with any compositor that implements the portal. No need for GNOME/KDE/... specific APIs.
• if you have a compositor that implements EIS you have all the infrastructure in place to talk to libei clients from somewhere else, if need be. The use-cases for this aren't fully scoped yet (assisitive technologies, virtual keyboards, touchpads, etc?) but the piping is there and ready to be (ab)used .
• since the actual events back and forth don't affect the layers in between, we can now add new events to libei without having to change everything else again.

The XWayland XTEST use-case

An example for such a case where we can now abuse the piping is Xwayland support for XTEST. XTEST is the protocol that everyone uses to emulate input under X but in Wayland it's not hooked up to anything so those APIs simply won't work.

But what we can do in Xwayland is translate XTEST to libei events and facilitate the portal interaction. This means our stack looks roughly like this:

    +--------------------+             +------------------+
    | Wayland compositor |---wayland---| Wayland client B |
    +--------------------+\            +------------------+
    | libinput |   EIS   | \_wayland______
    +----------+---------+                \
        |          |           +-------+------------------+
 /dev/input/       +-----------| libei |     XWayland     |
                               +-------+------------------+
                                                |
                                                | XTEST
                                                |
                                         +-----------+
                                         |  X client |
                                         +-----------+

What's next

So we have a lot of the plumbing in place, or in another anology: we have a hammer, let's go looking for nails. And right now the nails we can see are sending text, gestures, and tablet support. And those will be the subject of the next few posts.

[1]: 6 years ago?! whoah...
[2]: in Remote Desktop's case replacing the DBus emulation APIs which were a Newton's Cradle of wakeups for at least 4 processes per event

Happy July 4th! For those of us around the world contemplating independence, it's a good day to think about how we came to rely on expensive cloud infrastructure for our fundamental computing needs.

With that in mind, here is my latest toy project: an open source tool that makes replicating, forking, sharing, and running container snapshots fast and easy across cloud and personal devices.

It's fun to play with, especially on bare metal hardware you run at home, or rent from a provider like Hetzner or OVH. Or, because it uses Tailscale, why not all of them in a single mesh?

There's a lot more to say but I don't have time right now. Details are in the README.

I will say this: humans and AI agents both want the same things when they're trying to get work done. Ephemeral containers aren't really it. But how about unlimited disk space, fast CPUs, an undo button, and the ability to move to whatever provider offers the best hardware at the best price? That's more like it.

Go visit thundersnap on github and tell me what you think!

Over the years I’ve occasionally noodled on what might be a better working fluid for supercritical turbines than Carbon Dioxide. It turns out the main fixed parameter is the critical temperature, because there’s a strong nonlinearity in density with it going up rapidly as that temperature is approached. The other parameters of note are thermal conductivity, specific heat, and density, with more being better. There’s a very short list of possible fluids to mix which aren’t horribly corrosive, thermally unstable, or otherwise problematic. I’ve put together a tool to play with all the possible options here. You should go play with it. The short of it is that a mix of Neon and Perfluorobenzene tuned to the desired critical temperature is probably optimal, but if Perflueropropane’s decomposition problems aren’t too bad or Titanium Tetrachloride mixes with other things well then combining with some of those may be beneficial. This approach to visualization is probably equally applicable to conventional refrigerants but with the fixed parameter being boiling point rather than critical temperature. I don’t know if it’s standard there. If it isn’t it should be.

Years ago there was this insane academic idea that the isotope Thorium-229 might have a metastable isomer whose energy state is so close that it could be flipped into that state using a laser. In principle this worked on paper but so completely goes against the fundamentals of chemistry that it has to be assumed that it won’t work. Now it’s actually been made to work. It’s a little hard to convey how bonkers this is. A truly herculean effort was necessary to find out what the extremely precise wavelength of the laser has to be. The chemistry actually matters. The chemical which the Th-229 is embedded in matters for how precise the laser has to be. The laser is pushing on the nucleus, which is pushing on an electron, which is in turn pulling on the nucleus, which is pulling on the laser. This is not how chemistry works. But it does have directly applications to making yet even more insanely accurate clocks than we have currently, with possible applications to things like measuring fluctuations in the dark matter passing over the earth.

Here’s a crazy new idea of mine: It would be very convenient if there were some isotope which absorbed neutrons and then turned into something with an insanely high cross section similar to Xenon-135 but a half-life on the order of minutes. That could be left in a reactor core to to provide a passive negative feedback loop which operated on flux instead of temperature. Since flux is leading and temperature is trailing this could react more quickly and reliably. The downside would be losing some neutrons to the passive buffer. The funny thing is we have no idea if such unobtanium exists: The neutron cross sections of things with short half-lives are largely unknown and hard to predict. But we have some data already! If this process is already happening accidentally from something in existing nuclear reactors then there should be a resonance in the time series data for temperature measurements in them which is very precise and consistent across reactors. A lot of such data for many different reactors already exists. Checking for that would be an experiment worth doing.