Per Vognsen

I like programming and understanding how stuff works. My background is in systems programming and game development.

Public Key

npub152ta9mm2x6nk4ffyrn7ycz444vf7x743t24yddyd40vvcwkw39wqv3v7uc

NIP-05 Address

pervognsen@mastodon-social.mostr.pub

Profile Code

nprofile1qqs2997jaa4rdfm255jpelzvp266kylr02c442jxkjx6hkxv8t8gjhqpz4mhxue69uhhyetvv9ujuerfw36x7tnsw43qzrthwden5te0dehhxtnvdakqcdnd7j

Publishing to

relay.ditto.pub

nos.lol

nostr.wine

Author Public Key

npub152ta9mm2x6nk4ffyrn7ycz444vf7x743t24yddyd40vvcwkw39wqv3v7uc

Show more details

Last Notes

2026-05-06 11:54:09 UTC

Apparently wolves are back in Danmark and a quick search shows some hilarious news headlines. Politiken: Er der virkelig nogen, der render rundt og fodrer ulve med godbider? "Is anyone really running around and feeding treats to wolves?"

2026-04-21 10:38:36 UTC

- reply

@nprofile…h6ty I'm a fuzzy on what exactly you're trying to accomplish. Are you just saying that you're okay with having a very delayed copy of the counters?

2026-04-09 21:20:25 UTC

In case you're not already using it, SponsorBlock is really good: https://addons.mozilla.org/en-US/firefox/addon/sponsorblock/

2026-03-30 05:19:32 UTC

- reply

@nprofile…zdn5 @nprofile…valu The specific repo is new and Codeberg itself is new. So this is the absolute worst it'll ever be in terms of search ranking.

2026-03-30 05:15:02 UTC

- reply

@nprofile…zdn5 @nprofile…valu It's on the first page of a Kagi search but definitely a bit buried. It's disappointing to see a bunch of random GH forks with higher ranking, but the Codeberg repo is also new.
https://files.mastodon.social/media_attachments/files/116/316/418/871/426/150/original/d6dde06ed851fff4.png

2026-03-26 04:28:27 UTC

Yesterday I was thinking about serialization of fiber-based state machines and remembered Near (RIP) had written an article exploring the problems and potential solutions: https://saveweb.github.io/near.sh/articles/design/cooperative-serialization.html.

2026-03-24 20:47:23 UTC

I've seen several independent, complete misconceptions of the Curry-Howard isomorphism just in the last week. Where are people getting these ideas, I wonder. Today's specimen, courtesy of the Hegel thread on HN: "Given Curry-Howard isomorphism, couldn't we ask AI to directly prove the property of the binary executable under the assumption of the HW model, instead of running PBTs?"

2026-03-21 03:28:30 UTC

My favorite weird physical scaling limit in computing is that you can't tile a 3D networked grid of computers indefinitely because you eventually create a black hole: https://en.wikipedia.org/wiki/Schwarzschild_radius#Calculating_the_maximum_volume_and_radius_possible_given_a_density_before_a_black_hole_forms. Whereas you can in theory tile the computers in a 2D network indefinitely.

2026-03-20 01:52:03 UTC

My recollection is that people implementing infinite smooth scrolling on the Amiga dealt with the wrap-around issue by replicating the streaming writes so that when you eventually reset the window to the opposite edge the data there is ready as an identical copy. On the NES it's usually assumed you do this via mappers in the carts (so you don't have to replicate the writes but can rely on address mirroring) but were there non-mapper-based smooth scrolling games on NES that replicated writes?

2026-03-05 17:36:39 UTC

- reply

@nprofile…valu One of the nice things about https://phanpy.social (which is purely client side) is how it handles that. I remember having that problem you mentioned with the original Mastodon web UI.

2026-03-02 11:33:13 UTC

- reply

@nprofile…wxct Cool demo. That reminds me, did you ever see https://rerun.io/viewer?

2026-02-27 11:13:17 UTC

I guess this deserves to be posted on a regular cadence for the benefit of anyone who hasn't seen it before: https://www.dkriesel.com/en/blog/2013/0802_xerox-workcentres_are_switching_written_numbers_when_scanning

2026-02-09 21:49:04 UTC

- reply

@nprofile…mv7s Like https://github.com/KhronosGroup/Vulkan-ValidationLayers/blob/main/docs/gpu_validation.md?

2026-02-07 03:21:12 UTC

There's long-running debate in programming about brevity vs descriptiveness for names, but something I don't see pointed out is that strong conventions (either universally or in your code base) like 'ptr' for memory pointer and 'idx' for sequence index are _more_ descriptive than 'pointer' and 'index'. A 'pointer' could be a mouse pointer and an 'index' could be a database index.

2026-02-05 13:54:54 UTC

I almost can't believe this thing is on the Archive with ISOs and everything: https://archive.org/details/ldr_0496_6cd/linux%20%20back.jpg. This specific CD-ROM set is how I got started with Linux (Slackware 3.0 for the win). Back when FTP mirrors were things you shipped on CD-ROMs.

2026-01-30 01:31:38 UTC

- reply

@nprofile…mv7s That cuts both ways. If 2^53 bits of integer precision aren't enough then 2^64 bits might not be enough either. E.g. some of the int128 proposals for timestamps (which strikes as overkill but I'm sure it can be justified).

2026-01-17 23:14:26 UTC

ASCII characters are not pixels: a deep dive into ASCII rendering. https://alexharri.com/blog/ascii-rendering

2025-08-13 11:00:12 UTC

The 9800X3D + RX 9070 XT system arrived. Might have gone slightly overboard with the fans in this case.
https://files.mastodon.social/media_attachments/files/115/021/106/946/450/356/original/8181e95976c5b21c.jpeg

2025-08-11 10:46:14 UTC

Random musing: If T is an immutable (non-interior-mutable), copyable type then it would be nice if r: &T and let t = *r; &t were indistinguishable. But you can cast references to pointers and observe their pointer identity (via address equality), so they are distinguishable. That seems like a tangible way in which the value semantics picture is nicer. Letting a shared reference's observable identity change non-deterministically (even if only one at function boundaries) is too icky.

2025-08-08 11:47:06 UTC

It feels like a troll that p => q (implication) is equivalent to p <= q (less than or equal).

2025-07-07 06:43:29 UTC

- reply

@nprofile…tslx That's the kind of thing @nprofile…2key's link to GetFrameStatistics, etc, is about. Unfortunately I don't think there is a reasonable cross platform option currently.

2025-07-07 02:53:51 UTC

- reply

I know there's the frame timing Vulkan extensions, there's mailbox present mode (but that isn't supported widely, e.g. not on NVIDIA, and I don't know how much it would help), there's probably a bunch of compositor-specific timing hooks, etc, but it seems nearly impossible to find useful, actionable information on how to proceed in the best way.

2025-07-07 02:50:06 UTC

Has anyone done a recent (cross platform) deep dive on how to do minimum latency UI? Especially with regard to stuff like mouse dragging widgets where you want the hardware mouse cursor to stick tightly. I was writing an imgui app and immediate present mode (not fifo/vsync) and event-driven presents with manual frame limiting to the refresh interval produces sufficiently low latency even with a 60 Hz refresh monitor. But that's going to be super janky if you want to do smooth animation as well.

2025-07-06 08:01:30 UTC

I was doing some experiments and you have to admit this is pretty funny:
Summary
tcc sqlite3.c -c -o sqlite3.o ran
1.25 ± 0.05 times faster than tokei sqlite3.c
14.02 ± 0.45 times faster than cloc sqlite3.c

2025-07-04 08:42:48 UTC

You know things have gotten bad when the Social Security Administration is sending out email blasts with partisan propaganda. Just got an email from them with the subject line: "Social Security Applauds Passage of Legislation Providing Historic Tax Relief for Seniors"

2025-07-01 13:55:19 UTC

- reply

@nprofile…tslx Aka blocking (a term I try to avoid since its default meaning in programming is confusingly different). In this case it's representing a table as a variable-length array of fixed-size tiles where a tile is internally a structure-of-arrays (so the arrays are fixed length).

2025-07-01 13:22:24 UTC

One nice thing about tiling compared to fully columnar/structure-of-arrays is that the fixed-size arrays within each tile are at a constant bytewise offset so you don't need a separate variable/register for each array base. You just need the tile base and the per-array offsets can be folded into the memory operands.

2025-06-17 16:31:52 UTC

If Apple ever moves away from ARM and designs their own instruction set, they should call it Apple LISA.

2025-06-17 10:20:27 UTC

- reply

@nprofile…e2ev What is an even number, even?

2025-06-07 02:08:02 UTC

- reply

@nprofile…59k6 Portmanteau of chick and shiksa?

2025-05-28 00:34:34 UTC

There are two reasonable justifications for encapsulation. The only fundamental justification is preserving data invariants, especially when safety is involved. So it's very strange to me that the OOP people never talk about them and only talk about reducing low coupling to make it easy to reduce the long-range impact of changes.

2025-05-24 00:15:28 UTC

- reply

@nprofile…59k6 Was AD niche back then in computer graphics? It must have been; at least I don't remember it being much of a thing when I was still paying attention to SIGGRAPH in that period (well, a little after).

2025-05-23 11:43:05 UTC

Re: boost, what the fuck is going on in Windows land. Has it really gotten this bad over there?

2025-05-22 00:08:10 UTC

Building a deterministic hypervisor for fuzzing: https://www.youtube.com/watch?v=0E6GBg13P60

2025-05-14 08:32:39 UTC

- reply

@nprofile…gvnm A practical joke? You mean a pranck?

2025-05-04 01:44:10 UTC

- reply

@nprofile…zsct Yeah, I didn't switch from Amiga to PC until the middle of the 486's lifespan and for games specifically it didn't really feel Amiga got overtaken by PC until Wolf3D and especially DOOM.

2025-05-04 00:01:53 UTC

- reply

@nprofile…zsct I always had the impression Amiga never got very big in the US and it was more of a European phenomenon. What was your perspective?

2025-05-01 22:01:12 UTC

- reply

@nprofile…59k6 Personally love this graphic design style.
https://files.mastodon.social/media_attachments/files/114/434/825/888/855/891/original/d486f63519af8315.png

2025-03-02 07:50:47 UTC

The frequency response of my laptop speakers is so bad that hihats are completely inaudible even at a volume high enough that other bands produce distortion.

2025-03-02 01:04:12 UTC

- reply

@nprofile…59k6 (This is very tangential but I first came across the name 'Gauss' as a kid because of the ultra-powerful Gauss Gun in Bullfrog's game Syndicate for the Amiga. At the time I didn't know it was a German name so I also butchered the pronunciation until I finally learned where the name came from.)

2025-02-25 05:28:18 UTC

A random comment I made reminded me that half-integers naturally show up when you represent an open or closed integer interval as a (center, radius) pair. It seems like many of the natural occurrences of half-integers in math/physics might be secretly related to the same kind of thing with centers on grids? https://en.wikipedia.org/wiki/Half-integer#Uses. Never made that connection, so it might be nonsense.

2023-08-31 03:52:37 UTC

- reply

@npub1y0e…mm9n The 18 MB is referring to the generated Rust code, not the size in the static data segments when compiled/linked. I agree that's a large blow-up but you do see a significant blow-up when trying to convert from a raw byte array to an equivalent C static initializer using hex literals or whatever.

2023-08-31 03:35:35 UTC

- reply

@npub1y0e…mm9n Yeah, I've used the fixed-address trick before although usually only as a fast path, i.e. I would still have the relocation information, I just don't need to apply if I can mmap at the preferred base address. But I usually just use position-independent representations these days if this is something I care about.

2023-08-31 03:00:52 UTC

While this has nice ergonomics, I take exception to the idea that it's zero-copy: https://manishearth.github.io/blog/2022/08/03/zero-copy-3-so-zero-its-dot-dot-dot-negative/
As soon as your types have references (e.g. the Person struct example), the operating system loader has to do pointer fix-ups on load. And because this usually happens with the additional overhead of copy-on-write soft page faults, I could see this being significantly slower than equivalent non-copy-on-write pointer fix-up code. I'd have to do proper testing.