John Regehr

he/him

professional interests: compilers, software testing, fuzzing, formal verification, lowkey shitposting, mountains, deserts, cats
my page: https://john.regehr.org/

Public Key

npub1er4sdt6lf8959np3qlvv2ev78kyyndxw2lzc7ut6c05upmme757s4kx9ky

NIP-05 Address

[email protected]

Profile Code

nprofile1qqsv36cx4a05nj6zescs0kx9vk0rmzzfkn8903v0w9av86wqaaul20gpz3mhxue69uhhyetvv9ujuerpd46hxtnfduqs6amnwvaz7tmwdaejumr0dsx7zvna

Publishing to

relay.damus.io

nos.lol

Author Public Key

npub1er4sdt6lf8959np3qlvv2ev78kyyndxw2lzc7ut6c05upmme757s4kx9ky

Show more details

Last Notes

2026-04-11 16:26:22 GMT

- reply

actually this file seems like a pretty big tell https://github.com/serjective/aburiscript/blob/main/docs/debugging.md

2026-04-11 16:12:33 GMT

wow-- you don't see someone trying to bring up a C++ compiler very often
https://github.com/serjective/aburiscript/

2026-02-15 02:04:20 GMT

- reply

I think we're of the same mind overall. in fact I'm teaching a "vibe coding" class to a bunch of undergrads right now, and the running joke is that everything I teach them about using an LLM better is actually something that makes them a better software engineer for real

2026-02-15 01:35:36 GMT

- reply

the latest coding models are absurdly good, please believe me that I admit this with a gigantic amount of reluctance

2026-02-14 22:13:20 GMT

program synthesis is a research area that combines searching over program spaces with formal verification. sometimes these are separated, other times we can get an SMT solver to help with the search, as well as doing the verification part.
in any case, it's hard to overstate how cooked the searching part is, due to LLMs. they're just incredibly good at it, and this is nice because our previous solutions were not particularly impressive.
so far LLMs haven't changed the verification part at all

2026-02-11 00:22:40 GMT

today a sysadmin sort of person in the college of engineering that houses the school of computing where I'm a professor suggested that we might need to take down a long series of technical reports, going back to the start of the 1990s, because the PDFs are not accessible, and that it is imperative that we maximize an accessibility score that is being harmed by these pdfs.
it made me think about "So Long and Thanks for All the Fish"

2026-01-14 21:19:43 GMT

if you were writing 1 or 2 slides on designing software so that it's easy (or at least not impossible) to test, what would you list besides what's on this slide?
https://files.mastodon.social/media_attachments/files/115/895/536/405/453/495/original/d90980926d405f18.png

2026-01-13 01:18:50 GMT

does anyone have a pointer to some good reading about practical aspects of software architecture? I'm interested in super nuts and bolts aspects of how we structure large software projects -- material for practitioners, not for software engineering researchers.
looking for some material I can have my class read later this spring...

2025-12-21 03:00:22 GMT

OMG GUYS
https://files.mastodon.social/media_attachments/files/115/755/317/784/604/616/original/15c699aab21a598f.png

2025-12-08 05:03:15 GMT

- reply

I would be remiss to not show this graph when mentioning quicksand
https://files.mastodon.social/media_attachments/files/115/682/190/268/222/793/original/26e3f0c1d739cf24.png

2025-12-08 04:57:45 GMT

in the desert you run into quicksand pretty often, and it's generally kind of fun, but apparently not always
https://www.reddit.com/r/CampingandHiking/comments/1pgxvt5/quicksand_trapped_me_on_the_hayduke_in_arches_np/

2025-11-30 20:52:49 GMT

when I used to teach operating systems, sometimes I'd spend an entire lecture covering the chain of events in between "you move the mouse" and "cursor on the screen moves." but this guy fits it into 7 minutes!
https://www.youtube.com/watch?v=sZC0Biv_Iyk

2025-11-26 23:24:21 GMT

go fuck yourself, program
https://files.mastodon.social/media_attachments/files/115/618/572/337/049/604/original/90008f4bdd7b5beb.png

2025-11-24 03:29:40 GMT

"We conducted a quasi-experimental field study on the Milan metro, observing 138 rides. In the control condition, a female experimenter, appearing pregnant, boarded the train with an observer. In the experimental condition, an additional experimenter dressed as Batman entered from another door. Passengers were significantly more likely to offer their seat when Batman was present"
https://www.nature.com/articles/s44184-025-00171-5

2025-11-10 23:45:08 GMT

I hope that I can ruin your day by getting you to read this map (courtesy wikipedia) of US town names that are portmanteaus of the two (or sometimes more) states they are near the borders of
https://files.mastodon.social/media_attachments/files/115/528/056/087/453/898/original/03f50fc170879462.png

2025-10-18 17:25:01 GMT

- reply

at some level I think this depends on having good seminar leadership to set the tone, etc.
but also, if you have the correct invite list, then there's very little of that work that actually needs to be done

2025-10-18 17:23:41 GMT

- reply

I guess you mean in terms of organization? I think there are two main aspects:
- the invitation list needs to be people who have a lot of interesting things to say, and who you don't mind spending a few days with
- the seminar schedule should not be tightly programmed. there should be opportunities for people to have the floor (if they want this) and then also plenty of time for unstructured discussion

2025-10-17 23:48:31 GMT

- reply

we're not savages

2025-10-17 23:33:43 GMT

- reply

anyway eventually I'll tell bro about the secret to clean shit in the kitchen. but he's gonna suffer a little bit first.
alt: it's "bar keepers friend." that's the secret.
https://files.mastodon.social/media_attachments/files/115/392/119/516/173/054/original/ad0e7c54c798c289.png

2025-10-17 23:31:23 GMT

omg having my kids in college is the best thing ever, my IQ has gone up like 40-50 points just due to them leaving home
https://files.mastodon.social/media_attachments/files/115/392/109/501/173/794/original/fcfa9ad62d8ff9d8.png

2025-10-17 19:26:06 GMT

I was just giving some advice about organizing a Dagstuhl seminar to a junior person who's organizing one. I wasn't sure what I was going to say but I'm happy with how it came out:
the job is not to organize a little conference, it's rather to organize the world's best hallway session about whatever your topic is

2025-10-14 04:37:56 GMT

just noticed that netflix has _Boyz in the Hood_ which I think was the genesis of my Ice Movie Theory, which tells us that movies with Ice Cube are generally good whereas movies with Ice-T are generally bad. I will not be taking questions.

2025-10-10 20:36:43 GMT

me: I have this recurring nightmare
doctor: what is it?
https://files.mastodon.social/media_attachments/files/115/351/787/358/954/419/original/43bc1bb7b5b4d762.png

2025-10-01 19:26:01 GMT

RIP
https://files.mastodon.social/media_attachments/files/115/300/544/563/586/055/original/f867c616166a193a.png

2025-09-23 18:31:49 GMT

did Ken Thompson respond to a question my student asked???
https://www.tuhs.org/pipermail/tuhs/2025-September/032567.html

2025-09-09 00:57:24 GMT

is this bullshit? or does ISA not really matter in some fictitious world where we can normalize for process and other factors?
https://www.techpowerup.com/340779/amd-claims-arm-isa-doesnt-offer-efficiency-advantage-over-x86

2025-08-21 20:50:10 GMT

flew a drone right into a tornado
https://www.youtube.com/watch?v=akQbDX-MMlo

2025-08-21 01:19:40 GMT

- reply

icfPOOPsla

2025-07-16 15:42:25 GMT

I usually think of big and little endian as being as good as each other, with the important thing being that we make a consistent choice. but I wonder how many cycles get used each day swapping byte order in optimized lexicographical order tests because we didn't go with big endian
https://gcc.godbolt.org/z/E5h94eq9d

2025-06-18 18:52:10 GMT

- reply

as far as I know it's the only way to initialize a class member that's a reference, for example

2025-06-18 17:13:34 GMT

I decided to lean into C++ initializer lists
alt: the details don't matter that much, just a kind of long C++ initializer list
https://files.mastodon.social/media_attachments/files/114/705/484/111/803/230/original/d92289a88cfd9242.png

2025-05-26 04:04:19 GMT

unfortunately for the AI bros, MCP only means one thing
https://files.mastodon.social/media_attachments/files/114/572/143/985/045/989/original/6cca6d8b1455214d.png

2025-05-10 22:06:32 GMT

these are my two favorite graphs
https://files.mastodon.social/media_attachments/files/114/485/798/535/369/871/original/82095b99c69e88d5.png
https://files.mastodon.social/media_attachments/files/114/485/799/167/683/553/original/6b0915ba38a67cb9.png

2025-05-03 19:04:31 GMT

if we're gonna be stuck with vibe coding, might as well try to have some fun with it
("anti-vibe coding" would be my standard schtick: stuff like invariant-based programming, assertions, testing, fuzzing, CI -- all the stuff we do to try to make software actually work)
https://files.mastodon.social/media_attachments/files/114/445/451/825/057/935/original/59bfc68c82ce3190.png

2025-04-26 22:51:44 GMT

- reply

some time ago I read four books on debugging and wrote a bit about them
https://blog.regehr.org/archives/849

2025-04-26 21:11:22 GMT

"Why is debugging often overlooked as a critical dev skill?"
think about how much worse this is all going to get with the rise of vibe coding!
https://www.reddit.com/r/ExperiencedDevs/comments/1k8n2c5/why_is_debugging_often_overlooked_as_a_critical/

2025-04-05 20:55:29 GMT

the protest at the Utah state capitol today felt big!

2025-04-04 23:24:39 GMT

appreciate this reddit moderator's energy
https://files.mastodon.social/media_attachments/files/114/282/267/793/094/104/original/1c1803da1322b592.png

2025-03-16 02:27:28 GMT

we took advantage of a break between spring storm systems to do a quick backpacking trip in the San Rafael Swell
https://files.mastodon.social/media_attachments/files/114/169/740/639/696/990/original/487bde7aefeb89b4.jpeg
https://files.mastodon.social/media_attachments/files/114/169/740/672/215/911/original/bffa3bc6b3ff9b77.jpeg
https://files.mastodon.social/media_attachments/files/114/169/740/905/923/177/original/07aef9857bf52961.jpeg
https://files.mastodon.social/media_attachments/files/114/169/740/949/784/763/original/c30beb8898e2cef0.jpeg

2025-03-13 19:46:40 GMT

- reply

https://gist.github.com/regehr/ce90534908085c7a933554b0f739ed92

2025-03-13 19:37:51 GMT

my colleague Pavel is playing with making binary searches and concludes that modern hardware is very very fast
https://pavpanchekha.com/blog/fast-bsearch.html

2025-03-02 00:13:32 GMT

- reply

I cannot, however, get to cancer.gov

2025-03-02 00:11:31 GMT

- reply

it's still up for me. weird. (but I get the same result as you from "downforeveryoneorjustme")

2025-03-01 22:49:05 GMT

- reply

it works for me at present

2025-02-23 02:02:50 GMT

my younger kid attacked my disastrously disorganized spice cabinet and made it easy to find stuff ♥️♥️
https://files.mastodon.social/media_attachments/files/114/050/733/725/313/314/original/5076ac7ea1f264a1.jpeg

2025-02-22 22:52:40 GMT

#caturday
https://files.mastodon.social/media_attachments/files/114/049/987/631/548/794/original/5e2e92810028b44d.jpeg

2025-02-15 01:17:14 GMT

just a heads-up for anyone who works with an ABET-accredited university program: ABET is pathetic
https://info.abet.org/abet-important-message-from-abet-president-and-ceo
https://files.mastodon.social/media_attachments/files/114/005/258/704/319/471/original/8d6f2f29516646b2.png

2025-02-07 05:20:06 GMT

other day I was showing a student something on my machine and he got all quiet. I figured he was digesting the deep thoughts, then he says "so you really just use emacs all the time?"

2025-02-01 21:01:03 GMT

been learning and thinking about the Burrows-Wheeler transform and so far my conclusion is "what the fuck?"
https://en.wikipedia.org/wiki/Burrows%E2%80%93Wheeler_transform

2025-01-25 21:14:11 GMT

- reply

not true. they are only legalizing *shitty* C. in the future please try to be precise.

2025-01-22 04:07:50 GMT

- reply

Orwell was so good!

2025-01-15 00:55:17 GMT

- reply

that's definitely a bit of a different first sentence

2025-01-14 23:40:03 GMT

- reply

nice!

2025-01-13 17:51:07 GMT

- reply

what’s missing from the index for this textbook about low-level C programming?? 🤔
https://files.mastodon.social/media_attachments/files/113/822/311/848/454/179/original/3c1ca85270224826.jpeg

2025-01-13 16:27:30 GMT

- reply

absolutely indefensible code. I mailed them about this years and years ago
https://gcc.godbolt.org/z/W3jdEdnn4

2025-01-13 16:22:31 GMT

Computer Systems: A Programmer’s Perspective, a textbook by Bryant and O’Halloran, is in its third edition and also its third decade of teaching students falsehoods about C
https://files.mastodon.social/media_attachments/files/113/821/958/101/725/134/original/f7365b7cd34e35fb.jpeg

2025-01-11 23:58:51 GMT

- reply

let's say we have 150 million lines in some language and we want to evolve the language. how does a portable IR help with that problem?

2025-01-11 21:00:14 GMT

- reply

the dilemma facing programming languages, as I see it, is that they fall into two categories:
- languages that can be rapidly and radically changed in response to problems (such as failures of abstraction) encountered in real use. but these are the languages without large load-bearing code bases, so they don't get many resources
- languages that have large, load-bearing code bases, which receive adequate engineering power, but not enough to overcome the legacy code problem

2025-01-11 19:39:03 GMT

- reply

I said it in a reply thread but might as well repeat it here: the fix is languages that support proper abstraction of things that you do often, including design pattern sorta things

2025-01-11 18:55:59 GMT

totally agree with this and there are SO many examples of tools where 99% of uses are adapting an existing program/config file instead of creating a new one from scratch
https://blog.yossarian.net/2025/01/10/Be-aware-of-the-Makefile-effect
like think about the last time you started a LaTeX document from a blank file...

2025-01-04 23:43:44 GMT

meta's AI throwing its creator team under the bus. truly, you could not make this shit up
https://www.theguardian.com/technology/2025/jan/03/meta-ai-powered-instagram-facebook-profiles
https://files.mastodon.social/media_attachments/files/113/772/736/323/170/769/original/59ba9e384641abf8.png

2025-01-04 05:14:05 GMT

oh shit— instagram discovered fencepost bugs
https://files.mastodon.social/media_attachments/files/113/768/370/066/793/891/original/d887087cd97d8945.jpeg

2024-12-23 02:57:28 GMT

Fondly recalling this gem, which is one of the things that set me on the path to integer overflow detection, some time ago
https://research.google/blog/extra-extra-read-all-about-it-nearly-all-binary-searches-and-mergesorts-are-broken/

2024-12-22 19:54:39 GMT

from @npub1g4g…9ev3:
"The only difference with C++ is that all fundamental types have the same endianness. Whereas for C, we allow floating point vs. integers to have different endiannesses. That only tends to matter for the tiniest slice of processors and devices, but some people care." 👀 👀

2024-12-21 19:35:02 GMT

this blog post from 2009 is sort of obviously correct and yet I don’t know of any web sites that do this
https://www.evanmiller.org/how-not-to-sort-by-average-rating.html

2024-12-17 15:52:28 GMT

- reply

awesome!!

2024-12-17 15:43:24 GMT

- reply

yeah his brute forcer was getting nowhere and I told him, "look it's just consuming A, three bits at a time" so hopefully he'll get there!

2024-12-17 15:37:14 GMT

- reply

I didn't have much help to give Jonas for part 2, but I didn't look at it all that closely.
he did get to be one of the lucky 10,000 who learned about quines yesterday though

2024-12-17 15:36:22 GMT

- reply

very BF! such busy beaver!

2024-12-16 20:09:33 GMT

- reply

nevermind, I misunderstood what he did. he just ran Dijkstra's rooted at the start and at the finish and that produced enough info that a quick scan of the grid points gave the answer

2024-12-16 18:59:44 GMT

- reply

plenty of repeated work, but whatev!

2024-12-16 18:59:08 GMT

- reply

he built part 1 from a shortest path code he had sitting around from some class, so for part 2 he just ran it in both directions from every grid point

2024-12-16 18:39:02 GMT

- reply

I think we just worked out the right algorithm for part 2 of of last night's problem, he's coding on it now :)

2024-12-16 18:32:51 GMT

I don't have enough spare brain energy to do Advent of Code but my older kid does and I'm having a ton of fun being a sounding board for his ideas. it's a really clever competition!

2024-12-11 21:28:40 GMT

- reply

and now also a CI hook that suggests alternatives when someone wants to add an undef to an LLVM unit test 🔥
https://github.com/llvm/llvm-project/pull/118506

2024-12-11 16:15:09 GMT

this is super nice, Nuno added a guide to undefined behavior in LLVM:
https://llvm.org/docs/UndefinedBehavior.html
we should have done this years and years ago...

2024-12-10 04:17:22 GMT

"Memory-safe PNG decoders now vastly outperform C PNG libraries"
https://www.reddit.com/r/rust/comments/1ha7uyi/memorysafe_png_decoders_now_vastly_outperform_c/

2024-12-10 01:32:16 GMT

https://files.mastodon.social/media_attachments/files/113/625/942/097/681/349/original/f74579fa88999b0d.jpeg

2024-12-06 23:04:09 GMT

I often look at the links at hn but seldom look at the comments. but it's interesting to see their reaction to this Financial Times piece:
"Why America’s economy is soaring ahead of its rivals"
https://archive.is/vt7sY#selection-1757.0-1757.52
basically, even the hn commenters seem to be capable of sensing the enormous divide between the soaring numbers and the on-the-ground experience of being part of this economy. I think that means things are bad.
https://news.ycombinator.com/item?id=42314797

2024-12-01 21:45:33 GMT

- reply

I’m teaching a large systems class in the spring and am trying to think of a pitch that I can use to convince them that actually learning the material is a wise choice, at this point in their lives

2024-12-01 21:35:14 GMT

- reply

now relate this phenomenon to AI coding assistants

2024-11-30 05:00:44 GMT

if you look at what LLVM actually spends its time doing (and I assume probably also GCC?), it's pointer chasing. basic blocks are doubly linked lists, etc. for a while I've had the idea that a from-scratch compiler re-implementation for GPGPUs might work out pretty well. and if not, it would at least be a couple years of super fun hacking. anyhow somehow I only now ran across this:
"Compilation On The GPU? A Feasibility Study"
https://dl.acm.org/doi/pdf/10.1145/3528416.3530249

2024-11-27 17:22:39 GMT

I don't judge but what were the parents thinking here??
https://files.mastodon.social/media_attachments/files/113/556/068/073/850/148/original/c770675cebd1c1c2.png

2024-11-08 16:59:21 GMT

- reply

textual LLVM IR isn't usually produced or parsed other than for debugging or testing

2024-11-08 16:46:23 GMT

- reply

sure, but, like, we mostly don't write LLVM IR by hand. it's not a choice I'd make for a programming language.

2024-11-08 15:31:55 GMT

a design decision about LLVM IR that I like is that it only lets you write down a floating point constant if it's representable in the FP type
https://gcc.godbolt.org/z/76cYfsdqW

2024-11-07 15:20:10 GMT

- reply

I can't believe you made me think about this

2024-11-04 06:14:50 GMT

Lazarus
https://files.mastodon.social/media_attachments/files/113/423/212/653/327/166/original/85911b0cbb630e89.jpeg

2024-11-02 02:57:58 GMT

not quite #caturday yet but I have this tired girl
https://files.mastodon.social/media_attachments/files/113/411/111/631/022/371/original/cc2c3fd14e757ee2.jpeg

2024-10-26 18:49:26 GMT

alt: just a tabbycat in the sun
https://files.mastodon.social/media_attachments/files/113/375/215/937/851/335/original/bcfa4676d5aeb73c.jpeg

2024-10-13 01:42:49 GMT

I ran across a thread on reddit about "very inexpensive food that is delicious" and I've been thinking a lot about it. I've long believed that many of the world's best foods come out of difficult, constrained situations. anyway, here's my highly incomplete list of delicious, cheap foods
- haluski
- frijoles de la olla
- fried potatoes
- mujadara
- polenta
- fried rice
- red beans and rice

2024-10-03 02:30:32 GMT

- reply

don't tempt me!!

2024-10-03 02:24:59 GMT

- reply

that still leaves the problem of synthesizing the closed form, it doesn't help much

2024-10-03 02:20:46 GMT

when compiling Brainfuck, optimizing loops that are actually additions is pretty easy. but I'm stuck on a general way to take this a step further and optimize the next level of loops that are actually multiplications. half baked thoughts at the link if anyone wants to take a look. computing a closed form for this kind of loop doesn't seem straightforward.
https://gist.github.com/regehr/4cb11a3a7911775f43b91971197d9cdd

2024-09-30 23:55:46 GMT

- reply

one of the enterprising students in my class is attacking the problem that ~35% of LLVM unit tests trigger UB. 🤪 for real.

2024-09-30 23:51:56 GMT

- reply

hoo boy. I want to say I'd use this in a performance comparison but LLVM 2.8 or whatev is pretty distant now

2024-09-30 23:39:00 GMT

- reply

but so far in class we've only targeted native assembly. once we're done with this, we'll retarget everything to LLVM and pick up for free all of those amazing low-level optimizations that are too numerous for us to do by hand

2024-09-30 23:38:15 GMT

- reply

ah, impossible to say without having worked on the other thing as well. what I can say about BF is that it's a bit a sweet spot for a compilers class because it has plenty of low-hanging optimizations, but they mostly require BF-specific knowledge, so for example LLVM alone would not be able to pick up most of these benefits.

2024-09-30 23:24:07 GMT

- reply

it was not my intent for this to become an "optimizing Brainfuck" account. this will stop soon.

2024-09-30 23:20:31 GMT

I wasn't sure how much it would matter to vectorize scan loops such as [<] and [>>] in Brainfuck, but on my M1 Mac I'm seeing dbfi.b going about 10x faster due to just this one optimization
http://www.hevanet.com/cristofd/08.html