Dig Dug at the local arcade is now filled out with my high scores. I’ve still got it.
Remote
Thalia Archibald
@thalia@discuss.systems
mastodon 4.7.3Computer science PhD student at the University of Utah, researching compiler verification with @regehr@mastodon.social.
UNIX historian and archivist.
Occasional Rust standard library contributor.
702 Followers
218 Following
50 Posts
Joined January 18, 2026
Site:
GitHub:
Might be getting a PDP-11/23 soon!!! :)
Replying to
When I read UNIX-HATERS a few years ago, it unexpectedly helped me fill a gap in regular expression history.
In its chapter on Usenet, they complain about Larry Wall's rn newsreader. It introduced killfiles, which filter emails using regular expressions.
Well, I had incidentally just been researching historical regular expression engines and trying to trace back the development of regular expressions in Perl 1.0. Since Larry Wall created both, it seemed likely that rn influenced Perl. It turned out that Perl adapted the engine from rn and ultimately from James Gosling's Emacs. Since Perl removed attribution comments, I don't think this had been established before.
When I revisit that project, I'll need to update it for what I learned from James Gosling at SIGGRAPH.
https://github.com/thaliaarchi/regexp-museum/blob/main/viewers/rn.md
#unix #vintagecomputing #retrocomputing
Replying to
For context, this is a draft copy of The UNIX-HATERS Handbook from two years before publication. It was presumably sent to Jay Lepreau for feedback and I think this is the only surviving copy. Jay worked with UNIX throughout his career and it was found among his papers, along with the UNIX V4 tape. I will be scanning it.
Personally, I found the book to be quite a fun read. It's a bunch of sore losers to UNIX's success. The anti-forward by Dennis Ritchie is a treat.
https://web.mit.edu/~simsong/www/ugh.pdf
Replying to
Do you know anyone in Nebraska who worked for Union Pacific in the '70s–90s or is a ham radio or vintage computing enthusiast? I'm following leads on Teletype equipment used there.
Boosts appreciated.
#vintagecomputing #retrocomputing #teletype
In reviewing my research notes for a SIGGRAPH interview, I'm surprised at how accurately I determined the who and when of the UNIX V4 tape.
Before we had any supporting documents with the tape, I had to look elsewhere. My best source was a list of the first UNIX users outside of Bell Labs, which lists Martin Newell of the University of Utah as #19, but does not have any dates. To estimate when the tape would have been made, I tracked down the other institutions' histories and UNIX license agreements from their archives. This was tedious, but it was the only way I could narrow the timeframe. I determined that UNIX was likely licensed for the U in June 1974.
After determining that, a letter was found among the same papers as the tape, from Ken Thompson to Martin Newell, apologizing for their delay in sending the tape. It was dated 31 May 1974. Then, once we recovered the tape's data, the file system showed the tape was copied on 12 June 1974. Wow.
#unix #retrocomputing #vintagecomputing
Replying to
A simplified history of regular expressions:
-
In 1951, Stephen Kleene invented regular expressions.
-
In 1966, Ken Thompson built the first software implementation of regular expressions, in the QED text editor for CTSS. UNIX ed is a cut down version of QED, popularizing the tradition of regular expressions in editors. grep started as a standalone version of the engine from ed. em, ex, vi, and vim evolved from ed.
-
In 1987, Larry Wall spread regular expressions to the next generation of programmers through Perl. Gravely, it uses backtracking, spreading terrible performance for decades. It derives its engine from the rn newsreader, which derives it from Gosling Emacs. James Gosling claims he wrote it from scratch.
-
In 2010, Russ Cox, having spent time at Bell Labs, rejected backtracking and built RE2, a fast and scalable engine, generalized from work on Google Code Search. Its ideas have spread to Go regexp and Rust regex, a new generation of engines.
-
Since 2025, Ian Erik Varatalu has been researching regular expressions for his PhD and has produced a fast engine using derivatives. It implements seldom-included intersection and complement and efficiently implements lookaround. It is faster than all previous engines.
Replying to
I was interviewed by SIGGRAPH to promo my talk! I cover some angles of the story here that haven't fit elsewhere.
https://blog.siggraph.org/2026/07/unix-v4-and-the-utah-teapot-recovering-a-computing-time-capsule.html/
#retrocomputing #vintagecomputing #graphics
Replying to
SIGGRAPH also featured my Q&A on Instagram! Bringing UNIX and teapots to the masses.
Replying to
My SIGGRAPH paper is now published!
https://dl.acm.org/doi/10.1145/3799885.3816010
#retrocomputing #vintagecomputing #graphics
Replying to
Welcome @kishl@discuss.systems, creator of this punch card generator, to the Fediverse!
#retrocomputing #vintagecomputing
Replying to on discuss.systems
My teapots, now signed by the original teapot co-conspirators!
The large teapot is a Melitta teapot of the same kind as the original. The 3D-printed one is Blinn aspect ratio, the one everyone’s familiar with, printed by @Archibald_M@mastodon.social. The small pink set is a children’s coffee set in the same lineup by Melitta (no graphics connection, but it’s cute).
#graphics
Codeberg's recent statement and updated policy on LLM usage on their platform is admirable. They make several well-reasoned points that are often overlooked in the hype and distraction.
https://blog.codeberg.org/protecting-our-floss-commons-from-llms.html
Replying to
Showing Martin Newell and Jim Blinn documents they hadn't seen in 50 years.
Pictured is my copy of Martin's original Utah teapot drawing. The first page is well known, but he had forgotten about the rest.
https://archive.org/details/newell_teapot_drawing
I also showed a list of UNIX users with his name, as well as several UNIX licenses and his Shaded Picture System proposal.
https://archive.org/details/unix_news_july-30-1975_waterloo/page/n9/mode/2up
#graphics
Replying to
My talk went fantastically!
A picture with several Utah folks afterwards.
Replying to
A fun moment during my talk. I showed Martin's name in a list of UNIX users and everyone was amused that the 50-year-old CS department phone number still works.
Replying to
The Utah teapot panel with Cem Yuksel, Martin Newell, and Jim Blinn was recorded!
https://www.youtube.com/watch?v=QhvOm8RrtC8
#graphics
Siggraph: "The Return of the Utah TeaPot" Jim Blinn, Martin Newell, intro by Prof. Cem Yuksel
Replying to
I had intended to do a full thread of teapots around LA, but the Hollywood stars weren't picturesque, the La Brea Tarpits Museum was closed, and I didn't make it to Griffith Observatory, so it was just this one. Oh well.
https://discuss.systems/@thalia/116943048006385061
Replying to
This is surreal. Apparently, an AI-generated version of me holding a magtape (but not the V4 tape) is now the thumbnail on at least the channel listing.
Replying to
A golden teapot for the SIGGRAPH History Archive:
Replying to
Thanks for the great week at SIGGRAPH! As an outsider to computer graphics, I had no idea what to expect, but it was so much fun. Lots of interest in teapots, history, and archival.
Replying to
Pixar’s walking teapots!
Their poster claims Newell modeled the teapot in 1973, which I'd never heard before. The SIGGRAPH trivia slides claim he modeled it in 1975 (when his dissertation was published). Some people claim it was 1976 (actually when Blinn’s paper popularizing it was published). Newell told me it was most likely early to mid 1974. If their 1973 claim isn’t bogus, I want to know what they know!
#graphics
Replying to
Another teapot question! This one’s actually slightly wrong (everyone’s slightly wrong). The teapot was modeled by Martin Newell in early to mid 1974, but his dissertation was completed in 1975.
Replying to
I sent a photo to the reporter to replace the thumbnail and the Google slop engine produced these. Ruined the colors and terrible taglines. "Peak 2026 tech", as the video says. Everything needs A/B testing for maximum clicks.
Replying to
@dahukanna@mastodon.social I'm less familiar with these business case studies, so I did some reading. Kodak's downfall was from not embracing digital photography, despite producing the first digital camera. Boeing's ruination is from lax quality control due to outsourcing and cutting costs.
So the connection you're making to LLM companies is that, despite bringing the technology to the masses, they're falling behind cheaper competitors. And that companies that don't manage quality control in their LLM use, the default state of affairs, are suffering due to mid-level engineer burnout. It's like the Boeing case, but with the effects directed outwards to customers. Is that it?
Replying to
Utah teapot trivia questions from the SIGGRAPH Papers Fast-Forward:
Tonight at Utah Rust, we're going to be building mini async runtimes, led by David Brungardt. Come and learn async with us! 
https://www.meetup.com/utah-rust/events/314696647/
https://github.com/carllerche/tokioconf-2026-workshop-exercises
#rust #async #tokio
https://www.meetup.com/utah-rust/events/314696647/
https://github.com/carllerche/tokioconf-2026-workshop-exercises
#rust #async #tokio
Replying to
The author used flood fill to detect the punched holes in the original image and they manually edited it to produce an unpunched card. Unfortunately, it doesn't type the human-readable text above the column like keypunches did, but I've suggested that improvement. The interactivity is well done!
https://github.com/Kishlay-notabot/web-punchcard
Replying to
@infosecdj@infosec.exchange I believe Teletype made their own ICs, though I haven't disassembled them far enough to get to the ICs yet. Always busy with other projects. But I know someone who's got a 43 with, he expects, bad ICs. I was thinking I might send a set of my ICs to replace his, then he'd send the other set to be imaged. Do most IC faults not interfere with imaging?
Yes, I'd open-source the emulator.
Replying to
@infosecdj@infosec.exchange I'm interested in imaging the ICs from one of my Teletype Model 43s to be able to emulate it. I have three, two of which have issues, so am thinking about sacrificing one to be imaged. But I'm not ready to relegate it to being a parts machine yet; I just want to check if this could be possible.
Replying to
@mrkwlsn@mastodon.social Wonderful! I wasn't aware. Looks like the city has its own connection to the computer graphics industry too.
https://www.sculpturedublin.ie/the-lord-mayor-of-dublin-alison-gilliland-launches-smithfield-utah-by-alan-butler/
Replying to
@ricci@discuss.systems Talks are unfortunately not recorded, due to reduced budget! Very disappointing.
Replying to
@me_@sueden.social Diese Domains sind verfügbare Alternativen zu unix.next:
unix.bayern
unix.jetzt
unix.meet
unix.new
unix.now
unix.today
unix.world
unixnext.dev
unixnext.meet
unixnext.org
Replying to
@hannorein@mastodon.social Apparently ASCII space is encoded as EBCDIC 0x40: https://wiki.osdev.org/EBCDIC I don't know what spacebar on a keypunch would have done, though. ASCII NUL from paper tape did nothing, so is similar.
Boosted by @joe@f.duriansoftware.com
Replying to
@joe@f.duriansoftware.com I have not verified, but that may have originated in Mike Leak’s portable C library, the immediate ancestor of stdio. That would be where I'd check first. A recent discussion:
https://www.tuhs.org/pipermail/tuhs/2026-May/thread.html#33698
Replying to
Replying to
@me_@sueden.social I visited Würzburg for a day on my way back from Rothenburg o.d.T. Beautiful town.
Boosted by @joe@f.duriansoftware.com
Replying to
@joe@f.duriansoftware.com Cursed. But no one will store their strings in low kernel memory and collide with a file descriptor, so it's fine :).
With the original PDP-11 calling convention, printf is really simple, as every function was variadic. Just read words relative to sp. But this was complicated with the addition of long and directly passing structs, and I don't know how it was fixed.
Even earlier than what you're looking at, I've seen some checks in V4/V5 sources for the argument being 0/1/2 in determining whether it's is a (proto version of) FILE or a file descriptor.