My week at Recurse Center
In late April I served a residency at Recurse Center, formerly known as
Hacker School. I want to write up what I did before I forget.
Recurse Center bills itself as being like a writer's retreat, but for
programming. Recursers get better at programming four days a week for
three months. There are some full-time instructors there to help, and
periodically a resident, usually someone notable, shows up for a week.
It's free to students: RC partners with companies that then pay it a
fee if they hire a Recurser.
I got onto the RC chat system and BBS a few weeks ahead and
immediately realized that it was going to be great. I am really wary
about belonging to groups, but I felt like I fit right in at RC, in a
way that I hadn't felt since I went off to math camp at age 14.
Recurse Center isn't that different from math camp now that I think
The only prescribed duty of a resident is to give a half-hour talk on
Monday night, preferably on a technical topic. I gave mine on the
history and internals of lightweight hash structures in programming
languages like Python and Perl. (You can read all about
that if you want to.)
Here's what else I did:
I gave a bunch of other talks: two on Git, one on calculating with
continued fractions, one on how the Haskell type inferencer works,
one on the topology of data types, one on the Unix process
model, one on Alien
Horrors from the Dawn of
Unix. This was too many
talks. I didn't have enough energy and time to prepare all of them
properly. On the other hand, a lot of people were very
complimentary about the talks and said they were very glad that I
gave so many. Also, giving talks is a great way to get people
familiar with you so that they won't be shy about talking to you or
asking you to work with them. But I think I'll cut it down to one
per day next time.
Alex Taipale was inspired by my
hash talk to implement hashes synthetically in
Python, and I paired with
her on that for the first part and reviewed her code a couple of
times after. It was really fun to see how she went about it.
Libby Horacek showed me around the text adventure game she wrote in
Haskell. I had the first of several strokes of luck here. Libby
had defined an input format to specify the room layout and the
objects, and I observed that it was very similar to
Asherah, a project that
another Recurser, Michelle Steigerwalt, had done a couple of years
before. I found this out because I read everyone's self-posted bio
ahead of time and browsed the interesting-sounding links.
Aditya Mukerjee was
implementing Git in Go.
He wanted help deciphering the delta format. Later I paired with
Aditya again and we debugged his implementation of the code that
expanded the deltas back into complete files. I hadn't known any
Go but it's easy to pick up.
Geoffrey Gilmore had read my
ancient article on how to write a regex
matcher. He had written his own
implementation in Scala and wanted to show it to me. I didn't know
any Scala but the code was very clear. Geoffrey had worked out a
clever way to visualize the resulting finite automaton: his
automaton object had a method that would dump out its graph in the
"dot" language, and he could feed that to
Graphviz to get it to
draw the graph.
I had a conference with Ahmed
Abdalla and Joel
Burget about SML. The main question they
wanted me to answer: Why might they want to look at SML instead of
Haskell? This was a stroke of luck: I was unusually well-prepared
to answer this question, having written many thousands of lines of
SML since about 1993. My answer was unequivocally that there was
no reason, SML was obsolete, because it had big problems which had
never been solved, and Haskell had been introduced in part to
solve, avoid, or finesse these problems.
For example, nobody knows how to incorporate references into a
Hindley-Milner type system. SML has tried at least three methods
for doing this over the years. They all suck, and none of them
work right. Haskell avoids the whole issue: no references. If you
want something like references, you can build a monad that
simulates it locally.
I could probably write a whole blog article about this, so maybe
Libby wanted to pair with me again. She offered me a choice: she
was building an e-reader, controlled by a Raspberry Pi, and mounted
inside an antique book that she had hollowed out. I would have been
willing to try this, although I didn't know anything about
Raspberry Pi. But my other choice was very attractive: she was
an ancient Windows paper-doll app that had been current in the
1990s. people had designed hundreds or thousands of dolls and
costumes, which were all languishing because nobody wanted to run
the app any more. She wanted to reimplement the dress-up program
Here I had another stroke of luck. I was already familiar with the
program, and I think I have even been into its source code at some
Libby had found that Gimp could load a KiSS cel, so we looked at
the Gimp source code to figure out the file format. She had been
planning to use
libpng to turn the cel into a PNG, but I showed
her a better way: convert it into a PPM file and feed to to
ppmtopng. This saved a lot of trouble! (I have written a little
bit about this
approach in the
past.) Libby hacked in the Gimp code, grafting her PPM file
writing code into the Gimp cel reading code in place of Gimp's
internal pixmap operations. It worked!
I talked to Chris Ball about his GitTorrent
Chris wants to make a decentralized github that doesn't depend on
the GitHub company or on their technical infrastructure. He spent
a long time trying to make me understand why he wanted to do the
project at all and what it was for. I think I eventually got it.
It also transpired that Chris knows way more about BitTorrent than
I do. I don't think I was much help to Chris.
Jesse Chen paired with me to fix
the layout problems that have been troubling my blog for years. We
redid the ancient table-based layout that I had inherited from
Blosxom ten years ago, converting it mostly to CSS, and fixed a
bunch of scrolling problems. We also fixed it to be legible on a
phone display, which it previously wasn't. Thanks Jesse!
I had a discussion with Michelle
Steigerwalt about big-O notation
and how you figure out what an algorithm's big-O-ness is, either
from counting lines in the source code or the pseudocode, or from
running the algorithm on different-size inputs and timing it. It's
fun that you can do the static analysis and then run the program
and see it produce the results you predicted.
There was a lot of other stuff. I met or at least spoke with around
90% of the seventy or so Recursers who were there with me. I attended
the daily stand-up status meetings with a different group each time.
I ate lunch and dinner with many people and had many conversations. I
went out drinking with Recursers at least once. The RC principals
kindly rescheduled the usual Thursday lightning talks to Monday so I
could attend. I met Erik Osheim
for lunch one day. And I baked cookies for our cookie-decorating
It was a great time, definitely a high point in my life. A thousand
thanks to RC, to Rachel Vincent and Dave Albert for essential support
while I was there, and to the facilitators, principals, and especially
to the other Recursers.
[Other articles in category /misc]