Rework is a podcast by the makers of Basecamp about a better way to work and run your business. While the prevailing narrative around successful entrepreneurship tells you to scale fast and raise money, we think there’s a better way. We’ll take you behind the scenes at Basecamp with co-founders Jason Fried and David Heinemeier Hansson and bring you stories from business owners who have embraced bootstrapping, staying small, and growing slow.
Bloomberg’s financial columnist Matt Levine joins Recode’s Peter Kafka to discuss his early career in investment banking at Goldman Sachs and that path that led him to becoming one of the industry’s top financial columnists.
Want to attend Code Media in Los Angeles, CA from November 18th-19th? Clic…
The Past and The Future
Tom: It’s interesting that this whole project has really just been a collision of lots of apparently unrelated things that you were interested in at the time, and they’ve all sort of come together in this one project.
Chris: Yeah. I find the most interesting projects are like that. It’s someone who has experienced quite disparate things and different ideas, and just want to mash everything together. I find that really fascinating as a concept for a project — just like, pick ideas out of a bag and see if you can build something that incorporates each of those ideas.
Tom: But in particular, you weren’t pursuing this as a goal initially. It’s not like you learned all of these different things so that you could build this programming language, it’s just that you happened to have already learned these different things following your interests, and then presumably this seemed like the obvious next step, was to make your own programming language?
Chris: I guess so, yeah. It did seem like everything pulled together, and the bits where I was quite fuzzy about “how on earth do I do multiplication in hardware?”, I went and learnt about those things, and it’s been great for just building a really broad understanding of a lot of quite seemingly disconnected things.
Tom: So you built some more of these layers — you had the booleans and then the integers, and you mentioned arrays — how many layers are there here? How sophisticated does this language get?
Chris: The language has four layers, and I think that’s probably where it will stop now. My first three layers in the stack — as you say: booleans, integers, arrays predominantly — those are all stack-based languages. You push variables onto stack, and you’d perform an operation on them, and that was all preparing myself for that layer where I start to actually turn this into a concrete syntax.
Chris: No, at this stage it was really just a really quite primitive stack-based machine. Each layer is almost a completely isolated thing, it has a contract with the layer above and below it. That’s kind of how I tried to architect things.
Tom: But now you’re at the point where you have added a concrete syntax to it, and people can write a Sentient program in something that looks more like a conventional programming language?
Chris: Absolutely. Again, looping back to Ruby and having done a lot of Ruby programming in my time, I tried to model a lot of my concrete syntax on what Ruby can do. It has a really good standard library, and there’s almost no overhead to calling methods on things.
It’s always going to struggle with being Turing complete — that’s a boundary I just can’t cross, as an inherent limitation of the language. But I’ve tried to approximate functions as best I can; you can’t have recursive functions, but you can define functions which are more like macros. It’s a means of calling the same code as you iterate over an array, for example. You can pass function pointers around, within certain constraints.
It’s really getting to that point where I’m able to write programs that really do express how I think about a problem. That original pangram problem, I’ve written it in this language and it’s pretty much exactly how I think about it, line by line. And it’s less than a hundred lines. I think half of that is just this lookup table which is mapping the number words into the digits that appear in those numbers.
Tom: Are there any other notable programs that you have written in it?
Chris: Yeah, I’m trying to capture lots of examples on a web site which I’m building, which is at sentient-lang.org. Things I’ve written as a means of testing the programming language are things like Sudoku solvers, the eight queens problem which is one of the “hello world” programs of declarative programming. Magic squares, these puzzles where all the rows and columns and diagonals add to the same number.
The most adventurous project I’d like to work on is, there’s a game called The Witness, which you and I have both played, where it has these geometric maze-solving problems. On the surface it sounds simple — you have to navigate a maze from start to finish — but slowly over the course of the game they start adding in additional constraints, and each of the different symbols in these puzzles have different rules. I kind of see that as the pinnacle of fiendishly difficult problems to solve with this language; if I am able to express that problem in this language, then I’ve kind of done what I wanted to.
I’m actually using problems as a means of stress-testing the language, as well. So if I’m writing a Sudoku solver, thinking “I really want to transpose this array here” — with a nested array I want to swap columns for rows, and then iterate over it, and make assertions that all of the elements are unique — I’m like “okay, well, I don’t have uniq yet, so let’s add that to the standard library; I don’t have transpose or different types of enumeration yet”. So I’ve been gradually building up this language as I go.
Tom: And just to clarify. For each of those things, if I wanted to write a Sudoku solver, or a magic square finder, or a Witness puzzle solver in Sentient, that challenge is just: how do I express the rules of Sudoku, how do I express the rules of magic squares. So, for magic squares, I would need to write a program that says “when you add up all the values in each column, they all add up to nine” or whatever, and specifically I don’t have to know how to implement an algorithm to find the solution, I just have to know how to explain what success looks like, and then I rely on the SAT solver to actually find a solution that satisfies the rules that I’ve written?
Chris: It almost seems too good to be true. You can write a program that just describes the problem, and then Sentient — by passing it to a SAT solver — will just try and figure it out by itself. If you were to write that Sudoku problem, you would just say like, “all of the columns must be unique, all of the rows must be unique, the boxes must be unique, and the numbers must be within this range”, and that’s pretty much the whole program. You’d then run that, and Sentient — probably at many solutions a second — it would solve that problem for you.
That ties back to a whole other area of computer science, around nondeterminism and NP problems versus P problems, and I find it fascinating that this really big engineering project has so many ties to a lot of the really theoretical stuff that dusty old professors talk about. And it’s actually doing something along those lines, and inventing something from it.
Tom: Yeah, and it’s really very impressive that you have got so far with this. That just by following your interest in various disconnected things, you’ve managed to make it all build up into this thing that — it sounds like — is now capable of actually solving problems in a useful way. And if it had existed when you’d first heard about this self-enumerating pangram problem, it would’ve been the perfect tool to use to solve that, and now, somehow, it exists!
Chris: Yeah, it’s really rewarding to look back at reckons I had a year or two ago about “I wonder if I can run these things backwards?”, or “wouldn’t it be great if there was a high-level language to do this?”, and it’s like, I’m not a hundred percent there yet, but it’s great to reflect on that and see that actually I’ve now got this thing I was craving all that time ago.
Tom: So looking to the future a bit: it’s not at version 1.0 yet. What will version 1.0 of Sentient look like? Have you thought that far ahead, or are you just taking it a day at a time?
So I’d like to build out that web site, and in doing so create a bunch of fun, interesting examples that demonstrate the capability of this language. And while I’m doing that, I’m learning more about what needs to be in the language, I’m stress-testing it. So I want to arrive at a stage where I’ve got a good set of examples, and I’m happy with the way the language solves those examples. I think I will release version 1.0 at that stage.
If you are interested in the language, it is online and I think it would be great to get feedback, especially from people who work with this stuff. If you’ve used Prolog before, compare the “hello world” example in Prolog with what that might look like in Sentient. Feedback on that would be great.
Tom: You mentioned that some of the work involved in doing this has been done at least in the company of other people — you mentioned London Computation Club, which I’m also a member of — and that you read a book, and that there have been other people involved, but it sounds like you’ve mostly done work on this on your own, it’s sort of been a solitary project for you.
I was interested in what it’s like to build something this large and this complex all by yourself.
Chris: It’s really difficult to work on something this big by yourself. If you’re having a bad day and you can’t get something to work, you don’t have anyone who has the context to bounce ideas off, and to go over to a whiteboard and try and figure something out; you have to just find that motivation to do it yourself.
On top of that, it doesn’t help that there’s no obvious way of making this a day job right now. It’s really a passion project that’s gone way out of hand, it takes a lot of my time up, and continually justifying that is an ongoing challenge.
Tom: Would you be working on it differently as part of a team of people?
Chris: I would certainly have worked on it differently at a macroscopic level — I would have been thinking about refactoring a lot more, I would have been documenting things better from the start, rather than trying to play catch-up now. I have a tendency to leave visual clues in the code that mean something to me, but they certainly wouldn’t mean things to other people.
I think there’s also an element of not wanting to be personally shamed if you’re working in a team, and leaving some gnarly method in there. No-one is pushing me to fix technical debt in this thing, it’s only when I actually bump into it again and I have to really do something about it.
Tom: But at the same time, it’s a project where you have been setting the goals. And it sounds as though you may not even have known necessarily what you wanted to do next. If you are building a web application as part of a team of twenty developers, you probably have a little bit more of a sense of where you’re going with it, and what the ultimate business goal is, whereas in this case it sounds like it would have been quite difficult for you to plan ahead, because you’ve really just been following this thread of your own interest for what sounds like years.
Chris: Yeah. It’s probably not at all agile in any way, this project. I haven’t been delivering anything throughout, really. It’s really been a passion project where I’ve gone away in a dark room for a long time and worked on this thing and wanted to get it right.
If I had been working in a team, I’d certainly be doing things differently. If I were working for a client I would be thinking about trying to ship something and seeing what works and getting feedback from people. But yeah, none of that has been happening.
A rare peek into the black hole where some of academia’s best and brightest disappear to make a dent in the universe.
The Vergecast interviews John Gruber of Daring Fireball on how reviewing iPhones has changed over time.
Come for the Super Famicom Mini chat, stay for a really fun interview with our friends working on the upcoming South Park RPG, The Fractured But Whole. Also: Golf Story, Destiny 2 (YES, MORE), Danganronpa V3, and a smattering of news.
Page 1 of 21Older