Kaohsiung on Ivon's Blog

Notes From the KaLuG 2603 Open Source Meetup

infoivonblog.nkfjt@aleeas.com (Ivon Huang) — Sat, 21 Mar 2026 17:00:00 +0800

Today’s theme continued the lightning talk format. It had been two months since the previous meetup.

At Kaohsiung Student Union, we continued borrowing a free venue in the name of a student club (totally real).

Notes are on HackMD.

First, Shawn shared the development prospects of RedHat OpenShift, explaining the upstream and downstream development relationship between OKD and OCD. He clarified the relationship among Fedora, CentOS Stream, and RHEL, which made me understand that CentOS Stream is not actually that unstable. He thinks its positioning is close to Ubuntu LTS. CentOS Stream still has major version numbers and locks the kernel version. RedHat still sends fixes to CentOS Stream, and then the open source community helps test for bugs. Shawn believes this can form a healthier ecosystem. But I feel that if that is the case… then I might as well just use Ubuntu LTS.

He also introduced the benefits of Fedora CoreOS, which heavily adopts bootc and deploys systems in an image-based way. He mentioned that when the time is ripe, rpm-ostree may be replaced by composefs in the future, which can effectively use the Linux kernel’s erofs mechanism to handle system files. But what I want to ask is: now that there are already products implemented with bootc such as uBlue Bazzite, their biggest problem is that it is hard for users to manually install .rpm files locally; they can only rebuild the system image, called local layering. So how will composefs solve this problem? The two of us did not discuss our way to a satisfactory answer, and this technology is not finalized yet either. We can only wait and see in the future. Honestly, I do not want to manually build an image in the cloud and pull it down for deployment every time I want to install some extra package on the system.

Then I, with my shallow knowledge and limited learning, went on stage to demonstrate Phosh, introducing the development process of the Linux phone ecosystem and the possibility of replacing Android. For the slides, I used the Phosh desktop introduction I made a few months ago. Although everyone present was quite interested in the Poco F1 (postmarketOS) and Pinetab 2 (Arch Linux ARM) I brought, as someone who was tortured by these devices for a long time and finally resigned myself to using an iPad, I had all kinds of mixed feelings… Phones and tablets that can run Linux are indeed very interesting toys, but they are only toys.

Hsu Chia-Chun from the Software Liberty Association Taiwan (SLAT) demonstrated the AI features of Nextcloud Office. They had already launched Copilot-like assistance features similar to Microsoft Office at the end of 2023, mainly implemented by connecting to external language model APIs such as ChatGPT. However, he mentioned that this cannot be processed asynchronously, and even running a simple text translation takes a long wait… It seems there are still many rough edges. This kind of web-based AI solution should really learn from Google Docs. It is truly a blessing for lazy people.

Notes from the KaLuG 2512 Open Source Community Meetup

infoivonblog.nkfjt@aleeas.com (Ivon Huang) — Sat, 13 Dec 2025 17:00:00 +0800

For the previous two months, we met at the shared space of Second Space. Today, for the first time, we moved to the larger club meeting room of Xiongxiaolian. From the floor-to-ceiling windows on the 9th floor, you can overlook the scenery of Central Park across the street.

Today’s theme was in the form of lightning talks.

Agenda: 2512- 第一次雄校聯聚會

Shawn shared how he transformed a Pi500 into a genuinely usable computer. Add an external monitor, and the cost is getting close to an Android tablet. Not to mention that the Wayland desktop still has to be configured by yourself. As expected, he also encountered the problem where Fcitx5 cannot input Chinese in certain GTK programs.

A high school student came up to share the 鑽石Minecraft伺服器託管 service they operate. Although part of it is modified from the public Pterodactyl Panel, it is still very impressive. It has a complete management system that can hot-swap Minecraft server cores without typing commands. It reminded me of a feat like Aternos.

The topic YC wanted to solve for the FRRouting Project was too specialized. This is basically at a level only industry insiders can answer, right? I could only keep nodding and pretend I understood. Phew, if we had to do a group discussion, I am afraid I would not be able to help.

Peter seemed very interested in Android flashing and degoogle projects, and even wanted to make a website providing flashing information. Mm-hmm, this kind of website is already rare in the Chinese-speaking world. There is almost no professional discussion forum like XDA. But his understanding of process terminology was not very deep, and after asking, I found out he indeed had never flashed a phone before. I think having actual hands-on experience would make it easier for people to understand the reasons for flashing! If you want to persuade others to flash their phones, start by flashing your main phone to LineageOS! Personally bear the pain of failing Play Integrity! When I said this, Amos, who was next to me holding a Samsung folding phone worth tens of thousands of Taiwan dollars, retreated a little. Well, after all, Samsung is a very nasty company, using the permanent Knox fuse mechanism to threaten people not to flash their phones.

For the first time, someone told me face to face that they read my blog, and there were more than two of them. They thanked me because my Fcitx5 tutorial content had helped them. I felt extremely happy and also bashful, because this blog has too many embarrassing things……

Amos talked about the CC 4.0-licensed open source album project 《時間浸漬》, launched through cooperation between Lin Chiang & Luca Bonaccorsi and the Open Culture Foundation. It is really super interesting. I really like his works after 娛樂世界, and I did not expect that he is still so avant-garde now, willing to explore new business models for the music industry. By the way, when Radiohead released In Rainbows, they also tried an innovative method where buyers decided the purchase amount themselves. Radiohead and U2: making money by giving music away for free - The Conversation

Notes from the KaLuG 2510 Open Source Community Meetup

infoivonblog.nkfjt@aleeas.com (Ivon Huang) — Tue, 28 Oct 2025 17:00:00 +0800

Today, first I listened to Shawn share about “All Systemd Go,” a conference attended by Systemd developers. It turns out there really are conferences for every kind of project. This kind of conference for discussing specialized subjects can only be held abroad.

Agenda: KaLUG meetup 2510 - kernel 遇上 user space

It feels like we have been talking about Linux kernel things for two months in a row. It is truly an honor to invite skilled developers to share.

The Linux kernel feature I heard about at today’s KaLuG that is worth looking forward to is the sched_ext project. It includes results contributed by developers from NCKU in Taiwan and was introduced in Linux 6.12. It allows users to change the scheduler from userspace through eBPF, achieving custom scheduling operations. You can even configure cgroups, with each cgroup independently running a different scheduler. In the past, achieving this required recompiling the kernel; after all, there were only so many built-in schedulers. But if clients want to implement their own schedulers for specific work scenarios, upstream obviously cannot accept merging every scheduler. Therefore, this provides an economical method that lets users define their own.

The most intuitive application of sched_ext is improving game performance. CachyOS was the first to build in a mechanism for switching sched_ext. According to the official documentation, they provide multiple scheduler implementations for users to choose from, such as scx_bpfland, which can reduce latency. But some people say Bore Scheduler performs the best. Hmm… this kind of scheduling play probably really requires running very demanding games to measure the difference; otherwise it is just benchmark scores for your own amusement.

Shawn brought a Pi500 computer. So this computer really does look like this. Hmph, I think if it is going to be used as a server, the case should still be as small as possible.

In the middle, so everyone could get to know each other, we gathered around the round table and introduced ourselves one by one. Hmm… I did not have much to mention, so I simply talked about my current professional identity and the Linux distribution I usually use. That’s it. But the official FreeBSD T-shirt I wore seems to have successfully drawn attention.

Speaking of which, from Amos of OCF I learned that people involved in open source projects are not necessarily full-time programmers. That made this script kiddie feel much more reassured and relaxed.

Sunset mentioned that Facebook’s standards for randomly Zucking people are unpredictable, and I deeply agree. The most infuriating part is that Meta’s button for reporting and requesting review is still broken. Whether on Facebook or Threads, it throws errors and does not even let you fill in a reason. The current mechanism retroactively settles scores: for some unknown reason, an old post might violate the rules one day and get taken down, and it might randomly ban a domain. From survivor bias, when you see those low-quality short-video shitposts and scam ads all getting away with it, it really makes you nauseous. These platforms truly do not value digital human rights. It reminds me of someone at KaLuG saying that they posted a late-night snack post and it was successfully maliciously reported… Facebook’s standards are inconsistent. I sincerely hope the AI reviewers and human reviewers’ entire families drop dead. Recently they even hypocritically sent me a user survey, asking whether they were Zucking people too excessively? No shit. I only treat social media messages as temporary places to store information, intended to share the flow of information and guide traffic to other important websites. I do not believe every post can live as long as a blog article. Therefore, this way of thinking is safe. Even if I get Zucked, the psychological sense of loss will not be too great, because all of this is only temporary. I can just skillfully post the same thing again another day. As for self-hosting a Mastodon instance to avoid censorship, even though he claims you can host one with a single Raspberry Pi… I do not have the guts. I will keep parasitizing mastodon.social.

Notes From the KaLuG 2509 Open Source Community Meetup

infoivonblog.nkfjt@aleeas.com (Ivon Huang) — Sat, 13 Sep 2025 17:00:00 +0800

Today’s topic was writing Linux kernel modules with Rust. For a language that will enter the kernel in the future, it is necessary to understand it early.

Recording: KaLUG meetup 2509 - Rust 的奇妙冒險：Hello Heaven (Rust for Linux)

When attending KaLUG this time, it was also my first time seeing an actual NixOS user in real life… I have always wanted to try this system, but I am very worried about whether I would fall into a cult I cannot escape.

After all, declarative reproducible systems are not mainstream. Are you pretending Ansible does not exist? For cross-distribution package installation solutions, there are currently Flatpak, AppImage, Snap, Nix, Guix, and so on. There are even stranger solutions like Podman + Distrobox. pip and npm are not included because they are not designed specifically for packaging Linux software. Although each implementation differs, all of them can achieve cross-distribution software use. One situation for using cross-distribution package installation solutions is: wanting to install unstable packages on a stable system, such as having Debian Stable’s stable system and Arch Linux’s latest packages. Which solution to choose depends on which functions users care about more. If you do not care about containerization and want contact without a sleeve, use AppImage or Nix. Nix is more suitable for developers, and it feels like the Homebrew experience on macOS. Its package management mechanism also guarantees that software will absolutely not destroy dependencies, as if statically linked, and can 100% roll back. If you can accept some container permission restrictions, there are many choices. There is no need to rush to decide which is better. In the end, it still depends on which format developers prefer, which they are more willing to package, and whether users can easily obtain the software. In my view, Flatpak and Snap are more user-friendly, while Distrobox is purely a developer tool, and like Nix, it requires high-level skill to use.

Hawawa, the laptop brought by sponsor OCF Open Culture Foundation finally let me touch a real Framework laptop. The factory Framework 12 brought by QA is one of the laptop brands with the best Linux compatibility besides ThinkPad, almost requiring no extra drivers. With only one screwdriver, the keyboard can be removed from the front, exposing all internal structures (although it has a bit of FNAF body-horror feeling). All parts are modularly designed and freely replaceable.

Although this machine is a bit thick, considering it can rotate 360 degrees and has a touchscreen, it is actually not heavy. If I have money in the future, I definitely have to buy one. Yes, if I really have money. By the way, after recently listening to the Rust for Linux kernel talk (X) evangelism conference (O), I learned that Rust entering the Linux kernel is already an inevitable trend. From originally only being a language for writing kernel module drivers, to penetrating Rust’s interface layer with C, and counterattacking into the kernel interior!!! Although there is still a lot of resistance.