@pervognsen@mastodon.social

Per Vognsen @pervognsen@mastodon.social · Jul 28, 2026

Replying to @jplebreton@mastodon.social

@jplebreton@mastodon.social The future is high carb.

3

0

Per Vognsen @pervognsen@mastodon.social · Apr 29, 2026

With the DRAM shortage, maybe it's time for RAM Doubler and other assorted scamware to make a comeback. On that note, https://devblogs.microsoft.com/oldnewthing/20211111-00/?p=105897

The Old New Thing

The inside story of the outside investigation of SoftRAM 95 - The Old New Thing

It had a plan, but didn't quite execute on it.

7

0

Per Vognsen @pervognsen@mastodon.social · Apr 28, 2026

Replying to @zeux@mastodon.gamedev.place

@zeux Sure, in this case I'm just trying to describe how the two cases (assignment vs expression-oriented initialization) aren't any better or worse for this kind of example, for the small subset of people who care about such things (a non-empty set that includes me).

0

Per Vognsen @pervognsen@mastodon.social · Apr 28, 2026

Replying to @pervognsen@mastodon.social

Example: In C, ? : is annoying in a single-pass compiler because you have to go back and insert an int-to-double conversion at the end of the basic block, or more likely you use a shim basic block just for that conversion (which I believe is what tcc does), once you determine that the overall expression has type double. You have the same problems (and same solutions) when single-pass compiling the equivalent case with let assignments.

2

0

Per Vognsen @pervognsen@mastodon.social · Apr 28, 2026

Replying to @pervognsen@mastodon.social

The simplest version of the single-pass algorithm (with no support for subtyping) is that the first (in source order) assignment to a let variable fixes the variable's type and any other assignments to the same variable (necessarily on disjoint paths) check against that type. Anyway, I used to think that one objective advantage of expression-oriented initialization was its simpler interaction with type propagation, but this shows that isn't actually a real difference.

1

3

0

Per Vognsen @pervognsen@mastodon.social · Apr 28, 2026

I realized this some months ago but don't think I mentioned it here:

Let variables with exactly-once assignment analysis can support simple type inference just like expression-oriented initialization. You don't need HM-style constraint solving. It works in a single-pass compiler assuming your subtyping rules are simple enough in their interaction with multiple branches, but that same caveat applies to expression-oriented initialization.

1

4

0

Per Vognsen @pervognsen@mastodon.social · Apr 26, 2026

Replying to @steve@discuss.systems

@steve@discuss.systems @joe@f.duriansoftware.com I'm more interested in the SIGBOVIK version of this, inspired by those terrible underdetermined sequence extrapolation puzzles. On a side note, when I first was exposed to Haskell in the late 90s and saw the syntax [a, b..c] I was excited for a second that it might be more general (i.e. not be limited to arithmetic progressions).

4

0

Per Vognsen @pervognsen@mastodon.social · Apr 24, 2026

Replying to @pervognsen@mastodon.social

I guess in the case of airlines they don't even need to appeal to force majeure because they already cancel underbooked and otherwise unprofitable flights all the time, so this is an application of existing practices on a larger scale.

2

1

0

Per Vognsen @pervognsen@mastodon.social · Apr 24, 2026

As airlines are starting to mass cancel "unprofitable flights" I wonder if I'll be a victim of that and my early booking back in March for my June trip will turn out to be useless. I understand the moral and legal justification for force majeure in general, but the concept of specifically cancelling _unprofitable_ flights feels predatory.

5

3

0

Per Vognsen @pervognsen@mastodon.social · Apr 22, 2026

Replying to @joe@f.duriansoftware.com

@joe Rolex Cenobite

3

0

Per Vognsen @pervognsen@mastodon.social · Apr 21, 2026

Replying to @zeux@mastodon.gamedev.place

@zeux@mastodon.gamedev.place It's funny to think back to how much confusion there was about tangent space generation in the early days of the normal mapping era. It's not something I've thought about in approximately 20 years, so I'm looking forward to what you're cooking.

4

6

0

Per Vognsen @pervognsen@mastodon.social · Apr 13, 2026

The NBA Jam repo on historicalsource has a bunch of doc files that are a time capsule from 1992. Here's a fun excerpt, relative to what actually happened: "The NBA has been cautious about entering the European market, it's understandable. The NBA, it is believed, will grant franchises and expand into Europe within the next few years." https://github.com/search?q=repo%3Ahistoricalsource%2Fnbajam-complete%20doc&type=code

GitHub

Build software better, together

GitHub is where people build software. More than 150 million people use GitHub to discover, fork, and contribute to over 420 million projects.

2

1

2

0

Per Vognsen @pervognsen@mastodon.social · Apr 12, 2026

Replying to @joe@f.duriansoftware.com

@joe Throw in a Game Boy Printer and now we're talking.

3

2

0

Per Vognsen @pervognsen@mastodon.social · Apr 12, 2026

Replying to @joe@f.duriansoftware.com

@joe I'm going to bet this is closer to "random junk in the attic" than "divorce asset".

2

4

0

Per Vognsen @pervognsen@mastodon.social · Mar 29, 2026

Replying to @dougall@mastodon.social

@dougall@mastodon.social Return of the king! I was so sad that uops had gotten stale.

4

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 27, 2026

Replying to @adrian@discuss.systems

@adrian@discuss.systems This reminds me, I currently have separate Firefox add-ons for old.reddit.com and nitter.net redirects. There should really just be a generic regex-based redirect add-on. I just checked, and indeed there is: https://addons.mozilla.org/en-US/firefox/addon/redirector/?utm_source=addons.mozilla.org&utm_medium=referral&utm_content=search

3

0

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 27, 2026

Replying to @MikkoMononen@mastodon.gamedev.place

@MikkoMononen@mastodon.gamedev.place Nice article! Even augmented with some of the additional information (e.g. the override flags), the derived data still feels more like a snapshot representation than a delta/change-based representation. My intuition is that a delta representation would be better at capturing the user's intention and therefore be better for rebasing, but I haven't thought deeply about it.

0

4

0

Per Vognsen @pervognsen@mastodon.social · Mar 14, 2026

Replying to @siguza@infosec.space

@siguza BTW, I should have clarified (as mentioned later in the linked thread) that this feature was actually removed in response: https://github.com/llvm/llvm-project/issues/156571 So all is well, for now.

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 14, 2026

Replying to @lenary@types.pl

@lenary Yeah, found the comments on that in the proposed LangRef changes. It's good that this is at least getting nailed down.

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 14, 2026

Replying to @pervognsen@mastodon.social

The ultimate reductio ad absurdum would be if we start putting machine code bytes in inline asm blocks only for compilers to eventually start "helpfully" disassembling them to perform intrinsic substitutions.

7

3

0

Per Vognsen @pervognsen@mastodon.social · Mar 14, 2026

TIL about LLVM's ExpandInlineAsm via https://discourse.llvm.org/t/rfc-constant-time-coding-support/87781/37 (@lenary@types.pl) and I'm more than a little horrified. Is nothing sacred?

LLVM Discussion Forums

[RFC] Constant-Time Coding Support

Aside: LLVM has looked inside inline assembly since ~2010 iirc. The hook is called ExpandInlineAsm: LLVM: llvm::TargetLowering Class Reference It is implemented for x86 (which looks at the asm string

10

14

3

0

Per Vognsen @pervognsen@mastodon.social · Mar 09, 2026

Replying to @ocornut@mastodon.gamedev.place

@ocornut@mastodon.gamedev.place Well deserved. Congrats. :)

2

0

Per Vognsen @pervognsen@mastodon.social · Mar 09, 2026

Replying to @steve@discuss.systems

@steve@discuss.systems It sounds like you're asking for _woke_ self-driving.

0

2

0

Per Vognsen @pervognsen@mastodon.social · Mar 08, 2026

Replying to @bal4e@tech.lgbt

@bal4e You're catching up. :)

1

2

0

Per Vognsen @pervognsen@mastodon.social · Mar 08, 2026

Replying to @pervognsen@mastodon.social

Also, I didn't know that VisiCalc literally did brute-force fixed-point iteration without any sorting, so that was a neat bit of history on its own.

2

0

Per Vognsen @pervognsen@mastodon.social · Mar 08, 2026

How to recalculate a spreadsheet, https://lord.io/spreadsheets/. Not just about spreadsheets, but any structurally similar system. A great, accessible overview if you don't feel like diving through the literature.

lord.io

How to Recalculate a Spreadsheet

25

0

8

0

Per Vognsen @pervognsen@mastodon.social · Mar 08, 2026

How to recalculate a spreadsheet, https://lord.io/spreadsheets/. Not just about spreadsheets, but any structurally similar system. A great, accessible overview if you don't feel like diving through the literature.

lord.io

How to Recalculate a Spreadsheet

25

5

8

0

Per Vognsen @pervognsen@mastodon.social · Mar 05, 2026

Replying to @adrian@discuss.systems

@adrian@discuss.systems Boo! I'm sure they were great.

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 03, 2026

Replying to @regehr@mastodon.social

@regehr@mastodon.social @zwarich@hachyderm.io @joe@f.duriansoftware.com Heh, speaking of 0.0f/0.0f crashing the tested compiler (one of the table 1 examples), you can still crash tcc today (last I checked) with INT_MIN/-1 on x86. I like to test any program that has a built-in expression evaluator with that and it's scary (but many not surprising) how many you can crash with it.

7

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 01, 2026

Replying to @pervognsen@mastodon.social

@harold I always think of it like flying around the Earth to chase the Sun--ultimately you have to "pay off" your time debt in a feedback circuit, but you can shift around where it is in the multi-cycle feedback loop depending on timing closure requirements.

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 01, 2026

Replying to @harold@mastodon.gamedev.place

@harold The dark arts: https://en.wikipedia.org/wiki/Clock_skew#Beneficial_skew

2

1

0

Per Vognsen @pervognsen@mastodon.social · Mar 01, 2026

Replying to @soller@floss.social

@soller@floss.social Yup. Even my most optimistic prediction involves it getting a lot worse before it starts getting better again.

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 27, 2026

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

D. Kriesel

Xerox scanners/photocopiers randomly alter numbers in scanned documents

Xerox scanners/photocopiers randomly alter numbers in scanned documents Please see the “condensed time line” section (the next one) for a time line of how the Xerox saga unfolded. It for example depi

19

1

12

0

Per Vognsen @pervognsen@mastodon.social · Feb 26, 2026

Replying to @pervognsen@mastodon.social

@shachaf The ability to split on basic block boundaries like that does assume you have some sort of static or dynamic CFI validation so untrusted code can't jump into the middle of basic blocks.

0

2

0

Per Vognsen @pervognsen@mastodon.social · Feb 26, 2026

Replying to @shachaf@gts.y.la

@shachaf Yeah, I was using it for basically that; I mentioned in a subthread that you just have to split on basic block boundaries to make it amenable to random access execution. Of course it still has some issues with locality for random access execution though you can somewhat address that with an AoSoA-like layout.

0

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 26, 2026

Replying to @glassbottommeg@mastodon.gamedev.place

@glassbottommeg A musou game with mice? A mousou game?

2

0

Per Vognsen @pervognsen@mastodon.social · Feb 20, 2026

Replying to @vurtun@mastodon.gamedev.place

@vurtun@mastodon.gamedev.place @sschoener@mastodon.gamedev.place Also, it's not "just" documentation and tests he's writing with coding agents. Just look at his GH output. It's hundreds of thousands of lines per month for dozens of new projects. Here's just one initial commit for one of his random projects that has 300,000 added lines.

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 20, 2026

Replying to @sschoener@mastodon.gamedev.place

@sschoener@mastodon.gamedev.place @vurtun@mastodon.gamedev.place There's ZLUDA as well in the "CUDA on AMD" category and it's much further along: https://vosen.github.io/ZLUDA/blog/zluda-update-q4-2025/ If you write stuff like "LLVM is NOT required. BarraCUDA does its own instruction encoding like an adult" (which the README is full of) any criticism of your effort seems fair game. It would definitely be nice to have something standalone, though.

2

0

Per Vognsen @pervognsen@mastodon.social · Feb 19, 2026

Replying to @pervognsen@mastodon.social

@zeux My immediate use-case is actually a lot like that but rather than chunking by a fixed quantity like K it's chunking on basic block boundaries for a variable-length instruction format.

0

Per Vognsen @pervognsen@mastodon.social · Feb 19, 2026

Replying to @zeux@mastodon.gamedev.place

@zeux Yes, although I tend to think of that as a separate framing issue--any multi-stream format can be (usually is) chunked into aligned frames.

0

2

0

Per Vognsen @pervognsen@mastodon.social · Feb 19, 2026

Once again reminding you of The Trick to reduce the critical decoding path of any off-the-shelf variable-length data format. Split each variable-length item into a small fixed-length part that uniquely determines the length of the remaining variable-length part. Then split the fixed-length and variable-length parts into two streams and use prefix sums: https://gist.github.com/pervognsen/547113f6b368317d86110a6bb01ce314

Gist

dual_streams.md

GitHub Gist: instantly share code, notes, and snippets.

19

7

8

0

Per Vognsen @pervognsen@mastodon.social · Feb 14, 2026

An interesting example of parallel discovery: the zero memory widget (ZMW) library released in 2003 (discovered in 2001-2002) vs Casey's first presentation of immediate-mode UI in 2005 (discovered in 2002). I assume approximately no-one has heard of ZMW. https://perso.univ-lyon1.fr/thierry.excoffier/ZMW/zmw-0.2.4/zmw.html

perso.univ-lyon1.fr

15

1

3

0

Per Vognsen @pervognsen@mastodon.social · Feb 11, 2026

Replying to @wolfpld@mastodon.gamedev.place

@wolfpld@mastodon.gamedev.place The paradox of tolerance but for coding agents: if you use coding agents, should you use coding agents written by coding agents?

2

6

0

Per Vognsen @pervognsen@mastodon.social · Feb 10, 2026

Replying to @chandlerc@hachyderm.io

@chandlerc@hachyderm.io @meowray@hachyderm.io The review culture thing also came up in https://www.npopov.com/2026/01/11/LLVM-The-bad-parts.html

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 04, 2026

Replying to @shanselman@hachyderm.io

@shanselman@hachyderm.io 'Stochastic parrot-crow-magpie' would be a mouthful, I guess.

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 04, 2026

Replying to @shanselman@hachyderm.io

@shanselman@hachyderm.io Parrots are really getting a bum rap from their stochastic cousins.

0

2

0

Per Vognsen @pervognsen@mastodon.social · Aug 25, 2025

Replying to @pervognsen@mastodon.social

@moonchild@social.applied-langua.ge @corbin@defcon.social Looks like there's this on defunctionalization in dependent type theory: https://arxiv.org/pdf/2304.04574 They also mention in the introduction that GADTs are sufficient for defunctionalizing non-dependent but polymorphic function types, referencing this paper: https://dl.acm.org/doi/10.1145/964001.964009

0

Per Vognsen @pervognsen@mastodon.social · Aug 24, 2025

Replying to @moonchild@social.applied-langua.ge

@moonchild@social.applied-langua.ge @corbin@defcon.social I'm still booting up from sleep but I think it's https://en.wikipedia.org/wiki/Defunctionalization See also section 2.3 (page 13) of https://www.brics.dk/RS/01/23/BRICS-RS-01-23.pdf Danvy in that example restricts himself to pairs where the component types are the same, but that's because he's using ML and doesn't have dependent types.

0

2

0

Per Vognsen @pervognsen@mastodon.social · Feb 14, 2023

Replying to @zwarich@hachyderm.io

@zwarich@hachyderm.io @stoklund@oldbytes.space I'd totally forgotten about the work on transducers with tagged transitions. Do you know any regex engines that use that for submatches?

0

1

0

Per Vognsen @pervognsen@mastodon.social · Feb 14, 2023

Replying to @stoklund@oldbytes.space

@stoklund@oldbytes.space The left factorization that's inherent to determinization doesn't play well with capture groups. AFAIK, everyone switches to a non-DFA method once a match has been confirmed and the match boundaries found, e.g. https://swtch.com/~rsc/regexp/regexp2.html

1

0

Per Vognsen

Posts

The inside story of the outside investigation of SoftRAM 95 - The Old New Thing

Build software better, together

[RFC] Constant-Time Coding Support

How to Recalculate a Spreadsheet

How to Recalculate a Spreadsheet

Xerox scanners/photocopiers randomly alter numbers in scanned documents

dual_streams.md

Remote instance