Hoshino Lina (星乃リナ) 🩵 3D Yuri Wedding prep!!!

Hacker & Mad Scientist VTuber!

💍 Cyan Nyan 星乃しお 🩵✨ 3D Yuri Wedding 2026 (npub1rzf…w4kn) 🩵 | We are Tokyo Hacker Girls (npub1k6r…8c3g) ✨

ママ : https://bsky.app/profile/heripiro.bsky.social | パパ : https://twitter.com/otozuki_teru
All the links: https://lina.yt/me

YouTube: https://lina.yt
Twitter: https://lina.yt/twitter
Bluesky: https://lina.yt/bsky
GitHub: https://lina.yt/gh
Sponsor Me!: https://lina.yt/sponsor
Buy me a Ko-Fi!: https://lina.yt/kofi

Public Key

npub1cxqje22r8ewnr30rt625d58cr8gpdjnw405cq5765m2qsvghnmpquz6wtz

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsvrqfv49pnuhf3ch34a92x6rupn5qkefh2h6vq20d2d4qgxyteasspqqcrz0ta

Publishing to

Author Public Key

npub1cxqje22r8ewnr30rt625d58cr8gpdjnw405cq5765m2qsvghnmpquz6wtz

Show more details

Last Notes

2026-07-24 14:31:41 UTC

Looks like Hugging Face scraped ~all of GitHub into an AI training dataset...
https://huggingface.co/spaces/HuggingFaceCode/in-the-stack
There is an opt-out link. You might want to use it.

2026-06-29 09:46:57 UTC

RE: https://vt.social/@lina/116822363262715333
Aaaaa I went over 12 hours a bit and I thought YT would cut off the beginning wait screen, but it cut off the end!!!
I'll upload the last bit from my local recording later ;;
#note1x7d…w8lp

2026-06-26 07:06:19 UTC

BTW, the open source Live2D renderer I'm building is fully cross-platform!
Thanks to wgpu-rs, it runs on Vulkan, Metal, DirectX 12, OpenGL ES, and even inside a browser with WebGL or WebGPU!
The power of Rust 🦀
https://media.vt.social/media_attachments/files/116/815/139/839/622/664/original/cf969506384cb5ea.png

2026-06-14 03:22:17 UTC

- reply

"What about projects like Wine?"
Wine has a test suite, with large numbers of tests they do not pass. That's how you know it's not copied from Windows code.
"Emulators?"
Emulators go to great lengths for compatibility, which is needed for things like game replays not to desync and cross-netplay. This takes a HUGE amount of work over many years to achieve legally, with people devoting months or years just to crack one single mystery.

2026-06-11 05:45:24 UTC

Is there any way to make macOS play well with Linux Samba? I'm getting all kinds of weird behavior like `mv` not working and files ending up with mode 000 that can't be changed.
I'm about to start tcpdump... but I can't be the only one running into this right???

2026-06-10 01:27:29 UTC

So I'm looking at what a "production" Live2D renderer would look like, and I'm torn because I think I want to use wgpu, but I need some features it doesn't support to do some things efficientlly.
To do 5.3 blend modes efficiently, I need either VK_EXT_blend_operation_advanced or VK_EXT_rasterization_order_attachment_access or [subpass self-dependency](https://developer.arm.com/community/arm-community-blogs/b/mobile-graphics-and-gaming-blog/posts/framebuffer-fetch-in-vulkan ), but WebGPU has none of those.
For efficient masking, I think it's basically going to have to be subpasses (i.e. something that lets me render a mask to an attachment, then sample it in another subpass, that stays in tile memory/cache).
(Metal has equivalents for all this too)
Another option for both is VK_EXT_fragment_shader_interlock, but that will probably suck on tilers...
How hard is it to get stuff like this added to wgpu as an extension?

2026-06-02 10:46:46 UTC

- reply

I hope VTubers and artists can work together and help each other, with mutual respect, giving each other the benefit of the doubt and some leeway when not everything is agreed upon in writing.
And I think 2D riggers and artists should look towards the 3D world, where this sort of model modification and self-expression is much more common!
People sell 3D models with edit rights all the time. It's essentially the norm.
If you're an illustrator, please consider allowing your clients to modify the artwork themselves in the future.
If you're a rigger, please spend some time to consider whether that cmo3 really is "liquid gold", or whether perhaps it's okay to be a bit more open.

2026-03-15 14:37:39 UTC

- reply

To put it mildly, I completely disagree with his take on this.

2026-03-09 01:17:50 UTC

Out of curiosity, I wanted to see if ChatGPT noticed the bug/mistake I mentioned here:
https://github.com/obsproject/obs-studio/pull/13198#issuecomment-4019545931
I pasted the relevant function in and... it nitpicked all sorts of irrelevant stuff about error handling (allocation failures etc), types, and API contracts, and completely missed the actual issue. It even mentioned "the use of CLOEXEC is reasonable" and the fact it conflicts with some logic branches was completely missed.
This is a function using standard POSIX APIs. This isn't even obscure code, it's doing something (spawning a process with a pipe) that has been done a million times before.
It just didn't catch this bug because this bug / a similar shaped bug wasn't in its training data. Because AI isn't intelligent, it's still just fancy autocomplete and can't think.
("But you didn't catch it either" - I wasn't reviewing the whole function when I wrote that PR, I was fixing a different bug and had some tunnel vision since I was just normalizing/extending/fixing a pattern that was already there, so I didn't notice the use of CLOEXEC earlier. But I did ask ChatGPT to review the whole function, not just the PR change.)

2026-02-26 01:55:34 UTC

- reply

SHE IS LINING THEM UP AND PROUD OF IT
I cannot bellieve this girl!!!!!!!! @npub1rzf…w4kn
https://media.vt.social/media_attachments/files/116/134/438/297/127/134/original/c2049e94e70e01f4.jpg

2026-02-21 14:37:06 UTC

Me: Move over, you're on top of the futon
Cyan: *moves over just a bit*
Me: *pulls hard* nngngnngngng *gives up*
Me: Why do you always do this???
Cyan: It's hard getting under~
Me: But you have to fix it later anyway!!
Cyan: *I* don't 😏
Me: 🤨🤨🤨

2026-02-12 12:51:36 UTC

- reply

Also I checked my desktop client and it also broke, just differently. It does not agree on the number of people in the room. It also seems to think I'm not myself, or at least my avatar is gone (even though it displays on other rooms).
Jank, jank everywhere...
https://media.vt.social/media_attachments/files/116/057/747/160/329/713/original/fd4c2d567bc9ad2c.png

2026-02-12 12:30:43 UTC

Speaking of Matrix jank... a 7000+ person room that I think is a Fedora room just randomly started notifying me for every message and became an "Empty chat" in my client (definitely not empty).
I'm so sorry but... Matrix really isn't ready for prime time. Please think twice before moving there.
https://media.vt.social/media_attachments/files/116/057/665/355/360/235/original/7801b92459c06f25.png

2026-02-12 09:07:31 UTC

- reply

You'd just have to migrate the account to have non local scope (presumably set a scope/domain field). Whether this breaks past mentions depends on whether those are stored by name or by ID... (I hope id)
But the more interesting thing is designing a flow to "take over" a local account and migrate to remote identity without risk of ending up with two unmergeable accounts too easily...

2026-02-12 05:46:47 UTC

Honestly Matrix E2EE and federation is a bit like cryptocurrency (without going into all its other social problems).
It's clever, it's an impressive technical achievement that it works at all... and it completely ignores all the practical reasons why centralized systems exist.
You don't solve Visa and Mastercard censorship with crypto, you solve it with payment platform diversity. And you don't solve Discord with Matrix, you solve it with something with a design closer to IRC or XMPP.
(Reposted & lightly edited from a DM reply I wrote)

2026-02-12 00:53:46 UTC

Posted my thoughts on Matrix over on Bluesky in a thread:
https://bsky.app/profile/lina.yt/post/3memjfjwzzs26
TL;DR no, it's not going to replace Discord. Matrix is too full of jank due to early poor design decisions to be widely usable by non-technical people.
The Fediverse and Bluesky are different takes on federated Twitter with different pros/cons, and federated Twitter is a much harder problem (because it's "global"), but Matrix is basically just all cons. It's a failed experiment, and you can simply do much better starting over rather than trying to fix it. Just because it's relatively widely used in FOSS spaces doesn't mean it's the way to go. Sunk cost fallacy etc.

2026-01-29 07:10:57 UTC

RE: https://vt.social/@lina/115976596526464609
Brought to you by "reverse engineering" the Unity Mono dll to match it up with the public bdwgc source code, and then patching the symbols back in with objcopy --add-symbol, to get proper stack traces and watchpoints to work in MinGW gdb.exe ^^;;
Also had to patch the dll to enable GC debug logs and warnings (which are hardcoded off in the build...)
https://media.vt.social/media_attachments/files/115/977/117/297/845/324/original/7ff72d17b23f5b91.png
#note13zp…4ay5

2026-01-27 04:36:58 UTC

Cyan: I want to have がきごり
Me: ???
Cyan: やきとり？
Me: ????
Cyan: やきごり？
Me: ??????
(She means かきごおり)

2025-12-08 03:21:27 UTC

I knew this was going to happen, but it's still extremely funny having to write this update to my [original doc](https://asahilina.net/luna-story ).
I wonder if any of her users/contributors noticed any of this?
https://vt.social/media/media_attachments/files/115/681/778/291/913/053/original/c01153d8d4ac999f.png

2025-10-24 23:49:12 UTC

- reply

You know it's beyond weird when I have people asking me "Are you SURE this is from FDo? Have you checked the email headers?"
Yeah. I checked. DKIM pass from lists.freedesktop.org. They used an anonymous CoC sender list as From: this time (previously when CoC emailed me it was individuals, but I guess they're going for "CoC as a whole" now).

2025-10-13 04:30:37 UTC

Feeling sad about Asahi Linux again this morning... I wish I could still work on that stuff but there's no way I can feel safe in the Linux GPU community after everything that happened. There's just... no closure about anything.

2025-10-08 05:09:03 UTC

- reply

I don't know if they support it in the app yet... the update you saw could just be some preliminary display support.

2025-10-04 17:34:30 UTC

- reply

I think the idea is one is for your account proper and the other is for the frontend, but yeah it's weird...

2025-09-12 09:07:11 UTC

- reply

Translation is basically the prime example where this angle becomes really gray, yes...

2025-07-27 10:07:59 UTC

So I had a feeling, but... after today, I can confirm that Spout2 is VERY poorly designed, in ways that affect performance and stability which can't be fixed without a complete redesign.
It's a real shame the entire Windows VTuber world relies on this mess of a system...
It uses a *single* shared texture, and uses CPU side synchronization to mutex updates and reads. You end up doing CPU blocking copies on both ends.
This is horrible for GPU performance, it's basically the worst way you could do this short of full CPU data copying.

2025-07-14 23:32:22 UTC

Cyan's iPhone woke me up with a scary unknown beeping sound and the only thing I saw on the screen was "Liquid in USB port" before it shut down out of battery (apparently it hadn't charged due to this).
What??? I get the liquid thing, but the sound?
Also I don't see any liquid and it charged fine to 37% after booting again, but apparently it has now decided there's liquid in the USB port again and won't charge with two different cables. ???

2025-06-23 13:35:36 UTC

- reply

I think it has it? https://invent.kde.org/plasma/kwin/-/merge_requests/7475/diffs

2025-04-02 06:46:19 UTC

- reply

They keep crafting new accounts...

2025-03-20 10:30:42 UTC

- reply

If you don't believe me, look at what happens when you configure an EQ with a reeeeally high Q (super steep/complex frequency response), and play back a click through it!
https://vt.social/media/media_attachments/files/114/194/293/504/019/267/original/159fd79ec11da4c4.mp4

2025-02-03 16:36:54 UTC

How do you change the frame rate to a simple constant and properly regenerate timestamps with ffmpeg, without transcoding or duplicating or dropping any frames?
Every time I try to do this, the only way that works is converting to a raw bitstream format and back. Is there really no way to tell ffmpeg to just throw away all timestamp and FPS information and just recreate it from scratch?
The closest I got is -bsf:v setts=ts="N/TB_OUT/30" but that only changes the timestamps, not the container timebase/frame rate, so it still shows up with weird numbers like 17.42 fps, 17.42 tbr, 1k tbn. I tried -r, -video_track_timescale but they don't work.
I ended up doing it with mkvmerge, which is super easy: mkvmerge -o out.webm --default-duration 0:30fps input.webm. This does what I want, which is clean timestamps with no weird rounding errors or duplicated frames at the beginning/end and a proper timebase 30 fps, 30 tbr, 1k tbn. But it means I need to round-trip through mkv if the container is something else...

2024-10-22 16:10:15 UTC

I just hit 1K followers!!! Thank you everyone!!!!! 🩵🩵🩵
It's crazy how fast things are growing here on Bluesky!! How long will it take to hit 2K? 🚀
https://vt.social/media/media_attachments/files/113/351/938/349/457/310/original/c24cc33856c25019.png

2024-10-17 19:29:52 UTC

Almost every thread about Rust for Linux ends up with someone saying "why not Zig instead"? And usually the answer is just "it's less mature" or "nobody pushed it".
I didn't know anything about Zig, so I decided to take a look today... and I'm not very impressed ^^;;
I learned that Zig does not have:
- Destructors
- Macros
- Lifetimes
- Operator overloading
Those are major reasons why I chose Rust for the drm/asahi driver...
It sounds like Zig is trying to be "modern C"... but the whole point of R4L is to not get stuck with C!
All those things Rust has that Zig doesn't are important for the things I'm doing.
Destructors/RAII are fundamental to how the driver tracks and cleans up firmware structures safely and reliably when needed. If I had to write "defer" everywhere it would be a bug-prone mess...
Rust's amazing macros are how I deal with the firmware versioning differences while keeping the driver maintainable. In C you can sort of do that (poorly) with the preprocessor... but Zig doesn't have that either, so that's actually a step back from C, as far as I can tell...!
I don't think I really need to explain lifetimes/memory safety, that's the one thing everyone knows is different between the languages... I'll just say, I am certain I wouldn't have gone from "first render" to "stable accelerated Linux desktop" in 2 days without memory safety.
(continued)

2024-10-07 13:56:52 UTC

From a Discord comment about this post:
> you know, i'm surprised lina and cyan talk about something so personal publicly
I think people don't write enough about healthy relationship dynamics. Sure, it's personal, but how are others supposed to learn what is healthy and what isn't if we don't talk about it?
The internet is full of horror stories about abuse, neglect, weaponized incompetence, cheating... but you almost never see long posts about positive relationship situations. The bad stories are a good reference for what not to do... but how are people going to learn what to do?
The way relationships are presented in media is very stereotyped, dry and often ritualistic (and sometimes actually veering on abuse or toxicity even when that is not the intent), but it's the only reference a lot of young people have for this stuff. Is that really okay?
I get the feeling a lot of people fall into "tolerable" relationships because they feel obliged to, or due to peer or family pressure, and they never realize that's not how it's supposed to be, because they just don't know. They think that as long as you aren't having regular fights with your partner and there is no physical abuse, that's a healthy relationship. But that's just the lowest bar...
I waited many years until I found Cyan, but I'm so glad I did. 💙
https://vt.social/users/cyan/statuses/113256887195850285

2024-09-11 06:20:36 UTC

Cyan: 4000 messages that day... how many seconds per message is that?
Me: *Thinks for 5s* about one message every 20 seconds
Cyan: *ignores me, uses a calculator*
Cyan: ~20 seconds per message...
Me: Told you
Cyan: How did you do that??
Me: 86400 sec per day / 4000...
Cyan: o_o

2024-09-09 22:28:39 UTC

4 days until our stream~ ✨
Each day we'll share a secret slide from our presentation! 💙
https://vt.social/users/lina/statuses/113104093914716150
https://vt.social/media/media_attachments/files/113/109/953/223/865/861/original/5423efbdfd656708.png

2024-09-09 10:52:45 UTC

- reply

Maybe you can script it somehow?

2024-08-31 22:56:20 UTC

- reply

The claim is that the hw_fence needs to keep the scheduler alive. So it has to hold a reference. If the driver no longer needs the scheduler, then it would drop its top-level reference, leaving only the hw_fence reference, and then that would deadlock when it is dropped.
> If the driver dropped its scheduler ref, wouldn't the scheduler only be dropped if there was nothing left in the workqueue ("But the scheduler also keeps track of jobs, which reference their completion fences" -> nonzero job_cmp refcnt)?
The jobs hold a reference to the job_completion_fence but *not* the only reference. Queueing a job returns a reference to the completion fence, and that fence lives on and possibly gets sent to other drivers entirely. The job_completion_fence has to be able to outlive the jobs since it continues to exist after jobs complete (its job is to signal that completion). So if the driver no longer needs the scheduler and drops its main ref, then later the last job_completion_fence to be dropped would actually trigger the scheduler to be destroyed per the above mechanism, and the deadlock.