The Browser War Is Over. The Desktop War Just Started | Source Code #14

Añadido: 2026-05-05

[00:00:00]Before we talk about the most important story of the week, at least for me, the one that gives this episode its title, I want to walk you through everything else first.

[00:00:10]And I am starting with something that is not exactly mainstream news, but that touches me personally because I covered this very topic recently in my dedicated kernel video. I am starting with something that almost nobody is going to report on because it is technically obscure and involves the removal of a feature rather than the addition of one.

[00:00:30]And yet for me, someone who covered the Linux memory subsystem in a dedicated video not long ago, this is the kind of story that makes you appreciate how the kernel actually works as a living evolving organism.

[00:00:43]In 2019, a kernel developer named Songlu added support for readonly transparent huge pages for the page cache. THP transparent huge pages is a mechanism that allows the kernel to use large memory pages, typically 2 megabytes instead of the standard 4 kilobyt to reduce TLB pressure and improve performance on memory intensive workloads. The basic idea is elegant.

[00:01:07]Instead of tracking 512 individual 4K byte pages, you track one 2 megabytes page. Fewer TLB entries, fewer misses, faster execution. Songlu extended this to the page cache, the intern kernel cache for fileback data, but only in readonly mode. And in the K config entry he wrote at the time, he added a comment writable huge page support for the file cache would arrive quote in the next few release cycles. That comment is still in the kernel source today, 6 years later.

[00:01:37]And this week, LWN reported that the readonly option is about to be removed entirely. Not extended, not fixed, removed. Why? Because in those six years, the colonel's memory subsystem evolved in a completely different direction. The modern approach uses what are called large folios, a more general and architecturally cleaner abstraction for managing multi-page memory blocks.

[00:02:01]and MADV collapse, a more recent mechanism, allows user space to explicitly request that filebacked memory be collapsed into huge pages in a more controlled and correct way. The original config readonly THP for FS feature became a stranded island surrounded by infrastructure that had moved on without it. It was never wrong.

[00:02:21]It just became unnecessary and then became technical debt. I want to dwell on this for a moment because I think it says something important about kernel development that doesn't get said enough.

[00:02:32]The instinct in software, in most of software, is to accumulate, to add, to extend.

[00:02:40]The colonel resists this instinct better than almost any other large code base in existence. Partly because of Lionus and the culture he has cultivated, and partly because the cost of carrying dead weight in the kernel is paid by every system that runs it. Removing code is an act of discipline. Admitting that a promise written six years ago is better broken than kept. That is intellectual honesty. And a comment in a K config file is a form of technical debt, too.

[00:03:06]Even words cost.

[00:03:09]This is a small story, but it is a good one. Let me say something that will annoy approximately half of the Linux internet. A is the best package manager on Linux. Not the most powerful in every dimension, not the fastest in every benchmark, but the most complete, most battle tested, most trusted piece of package management infrastructure that exists in the free software ecosystem.

[00:03:31]And this week, it got better. Apt 3.2 has landed in DBN Unstable, and the headline feature is something that users of Pac-Man and Zepper have had for a while, and that Debian users have quietly wanted for years. Transaction history with roll back. specifically apt history roll back, apt history list, apt history info, apt history undo, and apt history redo. These are not cosmetic additions. The ability to say undo what I just did to your package manager on a production Debian stable machine is genuinely useful. It is the kind of feature that saves you at 2 a.m. when you realize the upgrade you just ran pulled in something it shouldn't have.

[00:04:13]But the part of this release that deserves more attention than it will get is the dependency solver improvements.

[00:04:19]Propagation, provider handling, source package upgrades, backtracking, all improved. The solver is the heart of AP.

[00:04:28]It is the thing that decides how to satisfy a complex web of dependencies without breaking your system. And every improvement to it is invisible when it works and catastrophic when it doesn't.

[00:04:39]The release also adds JSONL performance counter logging, structured logging for debugging and analysis of package management behavior, which is the kind of infrastructure work that makes the next 5 years of development easier. It is boring. It is essential. A few other details worth mentioning. The system now stays awake while DPKG is running, which prevents a class of infuriating interruption bugs on laptops.

[00:05:05]An apt daily is triggered when connecting to AC power, which again sounds small and is genuinely useful in practice. For now, AP3.2 lives in Debian side. Stable users will need to wait for the normal propagation cycle. But this is the direction of travel and it is a good one. APT3.2 is not a revolution. It is something harder to achieve. A mature trusted tool getting meaningfully better in exactly the ways its users needed it to. Miracle WM is not Hyperland. It is not sway. It is not trying to be either.

[00:05:37]Miracle WM is a Wayland compositor built on top of Mir. The display server library originally developed by Canonicle for Ubuntu Touch and Ubuntu's failed Unity 8 desktop.

[00:05:48]MIR survived those projects. It is actively maintained. It is solid and it gives Miracle WM a different kind of foundation than most compositors in this space. Version 0.9 was released this week and it marks what I think is a genuine philosophical shift for the project. The headline is the plug-in system. Miracle WM0.9 introduces Web Assembly plugins and a new Rust API. Web Assembly is an unusual choice for a window manager and I mean that as a compliment. It means you can write extensions in any language that compiles to WASM. It means the extension sandbox is well defined. It means the project is thinking seriously about extensibility as a first class concern, not as an afterthought bolted on later. The Rust API completes this picture. Rust and Web Assembly have a natural affinity, and offering both says something about the project's long-term ambitions.

[00:06:43]Beyond the plug-in story, 0.9 adds cursor theme support, improves performance, and changes the config reload behavior. Instead of reloading automatically, users now press meta shift R. There is also a breaking change. Keybinding configurations must now use human readable XKB keys names instead of raw Linux key codes. This is the right decision even if it breaks existing configs. XKB names are portable, readable, and maintainable.

[00:07:09]Raw key codes are not. The bug fix list includes several things that were embarrassing in earlier versions.

[00:07:16]dialogues and pop-ups being incorrectly tiled, inconsistent pointer-based window resizing, and timing issues with open animations. Miracle WM will not become your daily driver next week. The project is still finding its audience. But in a space dominated by Hyperland's aesthetics and sways minimalism, a compositor that leads with hackability and WAM plugins is at minimum an interesting experiment and it is built on infrastructure MIR that has survived longer than most people expected it to.

[00:07:47]I sometimes wonder whether the real differences between mirased compositors and the rest of the Wayland ecosystem deserve a dedicated video of their own.

[00:07:56]That is a conversation for another time.

[00:07:58]For now, worth watching.

[00:08:01]Every year or so, Feronx runs a fresh benchmark comparison between Firefox and Chrome on Linux. Every year, I read it hoping for good news. Every year, I come away with the same feeling. Quiet, persistent disappointment.

[00:08:15]This year is no different. In the Jetream 3 benchmark, one of the most comprehensive modern JavaScript tests available, Chrome came in at 1.47 times the performance of Firefox on the same hardware on Linux. That is not a rounding error. That is Chrome being nearly 50% faster at JavaScript execution on the platform that is supposed to be Firefox's home. Firefox wins motion mark and stylebench. There is some competitive ground on web assembly. Power consumption is essentially identical. Chrome at 11.44 watts, Firefox at 1174. Memory usage is close. CPU utilization is close. But JavaScript performance is the engine of the modern web. And on that metric, Firefox is losing badly. Now, I want to be fair here because the easy narrative, Google has infinite resources, Spider Monkey can never beat V8, case closed, is only part of the truth. Yes, Google employs an army of compiler engineers.

[00:09:13]Yes, V8 is one of the most aggressively optimized pieces of software ever written. Yes, the structural advantage is real and it is large. But Mozilla's problem is not purely a resource problem. It is a focus problem. In the last several years, Mosilla has launched a VPN service, a relay email service, an AI assistant, a social platform experiment. They have reorganized, cut engineers, laid off entire teams, including people who worked on Firefox's core rendering and JavaScript engine and redirected money toward products and initiatives that have not succeeded and have distracted from the one thing that Mozilla exists to do. Firefox is not losing to Chrome because Chrome is inevitable. Firefox is losing to Chrome because Mosilla has spent years making decisions that prioritized revenue diversification over technical excellence in the browser itself. I am still on Firefox. I will probably stay on Firefox. Not because I think it is the fastest browser. It clearly is not.

[00:10:13]Not because I think Mozilla is blameless. I clearly don't. But because Firefox is the last mainstream browser engine that is not Blink, if Firefox disappears or if it becomes a Chromium wrapper as some have speculated, the web becomes a Google monoculture in the most literal sense. One rendering engine, one set of design decisions, one company deciding what the web can and cannot do.

[00:10:38]For those of us who use Linux, who have always understood that the tools you use are political choices as much as technical ones, using Firefox is a statement. It is imperfect. It is declining and it still matters. The benchmark results are what they are. I am not going to pretend otherwise, but I'm also not going to let a Jetream score tell me which browser to trust with my web. I'm still waiting for a truly free and independent alternative, something like Ladybird, which I find genuinely exciting as a long-term prospect. But in the meantime, I'm stuck with the Mozilla Foundation, which is not exactly the kind of collective I have a lot of admiration for. And that tension between the tool I need and the organization behind it, is one I live with every day. This one is shorter and it is lighter, and I want it to be both of those things.

[00:11:30]KDE Plasma's two classic themes, oxygen and air, are being restored by a group of KDE contributors, and they are targeting the plasma 6.7 release scheduled for June 16th of this year. If you have been using Linux since the KDE4 era, you know these themes. Oxygen was the default from KDE 4.0. Dark gradients, a glassy depth, the aesthetic that said, "This is a serious desktop."

[00:11:56]Air replaced it as the default in KDE 4.3. going the other direction. Light, transparent, almost ethereal. Both defined what a KDE desktop looked like for millions of users through the late 2000s and early 2010s.

[00:12:11]Oxygen limped into plasma 5 and six in increasingly broken condition. Air was dropped from plasma entirely. Now both are being brought back properly by contributor Philip Filela alongside Nuno Panero, the original oxygen designer and several other KDE developers. The panel has been reworked and is now correctly orientation aware. Adaptive opacity is supported. Air's transparency has been restored. Both themes are being rebuilt to work correctly in the modern Plasma 6 environment, not just patched to barely function. The timing is deliberate.

[00:12:44]KDE's 30th anniversary coincides with the Plasma 6.7 release. These are not just pretty themes. They are historical artifacts, and treating them as such is the right call. I will be honest about why this one caught my attention. I loved Oxygen. I genuinely did. I found that theme iconic. It had a presence, a weight, and identity that felt right for what KDE was trying to be. And watching it decay through successive plasma releases was a small but real loss.

[00:13:13]Because in an era where every design trend points toward flat, minimal, reduced to zero, where breeze is the aesthetic ideal precisely because it disappears.

[00:13:24]There is something honest about wanting depth back. Oxygen and air were not subtle. They were proud of being there.

[00:13:30]They had visual weight. They said this desktop has presence. I am not suggesting it was better. I'm suggesting it was different. And that difference is worth preserving. If you are running KDE and you want to try them now, the files are already available from Philip Filela's blog. Link in the description.

[00:13:48]Here we are. I told you at the start of this episode that there was one story this week that I had been waiting for and that when it arrived, I felt something. Let me try to explain what I mean by that. Hyperland has been in Debbian 13 stable backports since April 10th. Not in SID, not in testing, in the official backports repository for Debian 13 Trixie, the current stable release available right now to anyone running Debian stable with a single apt install Hyperland. And not just Hyperland itself, the full ecosystem. Hyperdle, Hyperlock, Hyperpulcanet, Hyperland Guutils, Hyperland Protocols, Hyperwalan Scanner, XDG Desktop Portal, Hyperland, multiple plugins, all the supporting libraries. And crucially, at version 3, which is the current upstream release, not an ancient snapshot, the real thing.

[00:14:41]To understand why this matters, you need to understand the history. Hyperland first entered Debian's unstable branch in July 2024.

[00:14:50]That was already meaningful. Sid gets a lot of things that never make it further, and Hyperland making it that far was a signal. But then in June 2025, ahead of the DBN 13 release, Hyperland and its related packages were removed from the testing branch. The reason given by Debian developers was straightforward. The packaged version was too far behind upstream and Debian could not commit to supporting it over the lifetime of a stable release. That is the correct decision. Dian is not Arch. Dian stable is a commitment not just to shipping software but to maintaining it for years. If you cannot maintain a package responsibly for that time frame, you do not put it in stable.

[00:15:34]That is the Debian way and it is one of the reasons Debian stable is worth running. But it meant that Hyperland, which was already the most interesting thing happening in the Whan compositor space, was locked out of the distribution that underpins more of the Linux ecosystem than any other single project. Ubuntu descends from Dian. Mint descends from Ubuntu. The Raspberry Pi OS, KI, Arbian, dozens of distributions that collectively represent the majority of Linux users in the world all trace their lineage to DBN. Being in DBN backports is not the same as being in DBN proper. I want to be precise about that. Backports require the user to explicitly add the source and opt in.

[00:16:16]They are not enabled by default, but they are official. They are DBN's way of saying we have evaluated this software.

[00:16:23]We believe it can be packaged responsibly and we are making it available to users who want it without requiring them to abandon stable. That is the Debian seal of approval.

[00:16:34]And for a project like Hyperland, which has been controversial, which has had community friction, which carries a reputation in some quarters as something that works beautifully but lives in the Arch or Nothing universe. That seal means something. I have used Hyperland.

[00:16:51]When I was running Debian, I used it and it was stable for me in ways that the Debian developers caution did not quite predict, but I understood their caution.

[00:17:00]The packaging situation was genuinely problematic. It was not prejudice against the project. It was responsible stewardship. Now that situation has resolved and the result is that Hyperland is for the first time a first class citizen of the broader Linux world. Not just the world of rolling release distributions and power users who compile things from source. The world where most people actually live.

[00:17:26]There is a principle I come back to often on this channel. Availability determines adoption. A compositor that requires Archer void to install comfortably will reach one kind of user.

[00:17:37]A compositor that is in Debian stable backports at current upstream version with the full supporting ecosystem that reaches a different kind of user entirely and reaching that user is how projects grow from communities into infrastructure.

[00:17:52]Hyperland has been maturing rapidly. The development pace has been extraordinary.

[00:17:57]the plug-in ecosystem, the configuration language, the integration with tools like Hyperlock and Hyperl, it is a complete coherent environment. And now Debian has looked at all of that and said, "Yes, this is ready. I think they are right." To everyone who has been waiting for this moment, who wanted to run Hyperland on their Debian machine without leaving Stable, without adding third party repositories, without trusting a PPA or a random flatp pack, the wait is over. one apt install hyperland. That is it. That is source code for this week. We covered a kernel feature that proved the value of knowing when to delete. A package manager that grew up a little more. A Whan compositor that wants to be a platform. A browser benchmark that hurt but didn't surprise.

[00:18:44]A pair of old KD themes that deserve to exist again. And a compositor that finally got the stamp it always deserved. A dense week, a good week mostly. I will see you in the next video. And as always, may Linux be with you.