WEBVTT Supercharging Python Tooling with Charlie Marsh STYLE ::cue(c.speaker) { color: cyan; } 00:00:21.180 --> 00:00:22.490 Tim: Hi everyone, welcome! 00:00:22.870 --> 00:00:32.619 This is Charlie Marsh from Astral and, uh, you, Charlie, have been building something I find slightly incredible. 00:00:32.869 --> 00:00:37.450 So this is my perception of, of, of what you've done. 00:00:37.450 --> 00:00:40.879 You thought, oh look, I've got this linter ruff that's suddenly 00:00:40.879 --> 00:00:45.750o going to make the life of Python developers 50 to 100 times better? 00:00:46.330 --> 00:00:49.770 And then, Oh, it's like, by the way, we're just going to bring out uv, 00:00:49.770 --> 00:00:53.780 hire some of the best Rust programmers in the world to create this amazing company. 00:00:54.259 --> 00:00:59.759 We have adopted this Rye thing as well, which is kind of like adjacent to that. 00:01:00.429 --> 00:01:07.740 And now you have made the lives of millions of Python programmers dramatically better. 00:01:07.740 --> 00:01:09.259 Like what a gift to the world. 00:01:11.585 --> 00:01:12.195 Charlie: Thank you. 00:01:12.215 --> 00:01:15.294 Um, no, I, uh, I, I really appreciate that. 00:01:15.294 --> 00:01:21.615 I mean, I, I don't know if I would, um, say such nice things about myself, but, um, I appreciate you saying them. 00:01:21.975 --> 00:01:30.784 Um, I think, uh, I love working on tools, um, because it's such an amazing, like, platform for impact. 00:01:31.144 --> 00:01:34.125 And, um, you know, originally when we were, when I started working 00:01:34.125 --> 00:01:37.620 on ruff, I, It was really fun, but like no one was really using it. 00:01:38.890 --> 00:01:42.500 Um, and I got, it was very gratifying, you know, in some ways and maybe not in others. 00:01:42.500 --> 00:01:47.609 And now, um, you know, we're kind of at the point now where, you know, if we ship a really big improvement 00:01:47.620 --> 00:01:54.900 to ruff, um, It immediately goes out and impacts, you know, a very large portion of the entire Python ecosystem. 00:01:54.940 --> 00:01:59.720 Um, and new capabilities and new, new powers start showing up in your editor, right? 00:01:59.720 --> 00:02:01.709 And like new workflows become possible. 00:02:02.000 --> 00:02:09.760 So, um, I love working on tooling, um, and I think the idea, That, um, we get to kind of keep in a compounding way, make 00:02:09.760 --> 00:02:12.750 Python better and better and better is really one of the big drivers of, of like 00:02:12.750 --> 00:02:16.280 why I am working on this stuff and why all the people on the team have joined too. 00:02:16.520 --> 00:02:21.570 Tim: Python seems to be going through, I'd say, somewhat of a renaissance or renaissance. 00:02:21.960 --> 00:02:25.080 Uh, the, uh, with like Python 3. 00:02:25.790 --> 00:02:32.709 Uh, where Python's 13 has just been, uh, has landed with sub-interpreters and no GIL. 00:02:32.950 --> 00:02:40.600 And it just feels as though there is, has been this, a reaction to say static 00:02:40.649 --> 00:02:45.600 typing with mypy, for example, originally, and that being pulled into the language. 00:02:45.970 --> 00:02:52.595 And now with tooling such as your own, you have, uh, Essentially, this 00:02:52.595 --> 00:02:55.695 compounding effect, that's a really interesting way to express that. 00:02:55.695 --> 00:03:04.675 Why, why did you think that you could do something different than what had already happened? 00:03:04.704 --> 00:03:09.734 Because when Ruff was being created, there was already Pylint. 00:03:09.765 --> 00:03:14.075 There was already probably, I think Black was also, was quite prominent. 00:03:15.095 --> 00:03:15.435 Charlie: Yep. 00:03:16.335 --> 00:03:16.755 Yeah. 00:03:16.815 --> 00:03:17.975 Tim: And you thought, you know what, we could do better. 00:03:17.994 --> 00:03:18.024 Yep. 00:03:20.234 --> 00:03:26.969 Charlie: Ya, I think, um, you know, one of the big drivers for me, um, in, in working on this kind of 00:03:26.969 --> 00:03:34.039 tooling was, um, I was seeing a lot of what was happening in the, the web ecosystem and the JavaScript ecosystem. 00:03:34.219 --> 00:03:37.519 And I think there are like a couple of interesting. 00:03:38.410 --> 00:03:44.019 Things, you know, properties that the web ecosystem has and things that happened in the way web ecosystem that 00:03:44.019 --> 00:03:52.689 led to this, but, um, they, there's just been a, um, a lot of innovation and investment in, in web tooling. 00:03:53.010 --> 00:03:56.769 Um, and, you know, and people complain about things like, oh, my node, node_modules is too big. 00:03:56.959 --> 00:03:57.250 Right. 00:03:57.250 --> 00:04:04.679 But like, um, uh, there's just a lot of people working on web tooling, um, and doing really interesting things in it. 00:04:05.285 --> 00:04:12.605 And, um, you know, I think a lot of that comes from, like, uh, React and TypeScript being these two, like, 00:04:12.605 --> 00:04:20.265 hugely successful projects that had big corporations behind them, and that, like, attracted a lot, it became very cool, 00:04:20.595 --> 00:04:25.835 at least in, like, my world, I don't know, it became very cool to, like, work on web tooling, whereas I think when I 00:04:25.835 --> 00:04:32.040 was, like, in college, maybe, I think like front end was like not as, um, it didn't seem as like hardcore, right? 00:04:32.040 --> 00:04:35.089 It was like the trope, um, whether it's true or not. 00:04:35.090 --> 00:04:36.700 I started my career doing front end. 00:04:37.390 --> 00:04:43.640 Tim: That's, that's pretty close to being a designer, you know, like, gosh, is that even real engineering? 00:04:44.570 --> 00:04:44.890 Charlie: Yeah. 00:04:44.890 --> 00:04:48.439 And it turns out it like really is, and it's actually very hard and complex. 00:04:48.439 --> 00:04:50.530 And there's like a lot of super interesting work to do. 00:04:50.530 --> 00:04:52.760 Like I've spent a lot of my career doing front end actually. 00:04:52.840 --> 00:04:56.979 Um, but it was interesting to me because there were just all these, all these people going in and 00:04:57.029 --> 00:04:59.580 like, everyone's kind of working on innovating and building tooling. 00:04:59.689 --> 00:05:02.320 Um, and, and, you know, the other thing I noticed there was. 00:05:03.200 --> 00:05:09.780 They, in web, they just ran into a lot of problems, like a lot of performance, especially like local 00:05:09.789 --> 00:05:14.589 driven and development driven performance problems, um, quickly. 00:05:14.599 --> 00:05:18.050 So like apps got bigger, and you have like, bundling and 00:05:18.050 --> 00:05:21.349 transpilation and minification, like all these things going on. 00:05:21.550 --> 00:05:26.320 So like, things are getting slower, and so people started to build like, Faster and faster tools. 00:05:26.350 --> 00:05:29.610 And in particular, they started to build tools that were not written in JavaScript. 00:05:29.630 --> 00:05:32.050 They were like, okay, we're going to build like foundational tooling 00:05:32.060 --> 00:05:35.130 in different languages to enable everyone who's working in JavaScript. 00:05:35.829 --> 00:05:40.050 Um, and the first example I always think of there is like esbuild, which was written in Go. 00:05:40.340 --> 00:05:48.640 Um, Now there's like a whole proliferation of tooling that's written in like Rust and Go and Zig in in the web in the web 00:05:48.640 --> 00:05:55.170 ecosystem And so I kind of like saw all that stuff happening in web and I was like looking at Python And I was thinking 00:05:55.179 --> 00:05:58.809 like why isn't this really is there a reason that it can't happen in Python? 00:05:58.830 --> 00:06:05.135 I guess would have been the question and it was um You know, it was a time in my life when I was kind of trying 00:06:05.135 --> 00:06:09.565 to figure out, I'd sort of, I'd left my job and I was kind of trying to figure out what I wanted to work on next. 00:06:09.915 --> 00:06:19.165 And, um, uh, you know, one of the big pieces from that too was I, um, I wanted to get better at Rust. 00:06:19.474 --> 00:06:22.724 That was like an honest, a very honest motivation for starting to work on Rust. 00:06:22.735 --> 00:06:25.135 Tim: Yeah, I, I, I relate to that very heavily. 00:06:25.195 --> 00:06:31.380 I, uh, sometimes say the reason I wrote Rust in Action was to teach myself Rust 00:06:32.690 --> 00:06:33.490 Charlie: Yeah, exactly. 00:06:33.490 --> 00:06:33.960 I mean I 00:06:34.200 --> 00:06:35.980 Tim: Like take on some big audacious goal 00:06:37.235 --> 00:06:40.825 Charlie: Yeah, and I think I wouldn't recommend this for everyone. 00:06:40.825 --> 00:06:44.415 But like for me, I think I kind of needed to like build something from scratch. 00:06:44.735 --> 00:06:49.854 Um, and you know, in my previous job, like we had a big Python monorepo and 00:06:49.855 --> 00:06:54.175 we started to move towards like a hybrid Rust Python, uh, monorepo over time. 00:06:54.184 --> 00:06:56.515 So, um, and it was, but I did not introduce that. 00:06:56.525 --> 00:07:02.440 It was one of my coworkers, brilliant, engineer named Colin Fuller, um, and he was like, I think we should start 00:07:02.440 --> 00:07:08.210 building some of our, like, performance critical services, uh, in Rust, also get like a lot more platform portability, 00:07:08.250 --> 00:07:14.640 much easier for us, because we were also writing, we were also writing software that we needed to run, um, like on 00:07:14.640 --> 00:07:19.000 Windows machines in particular, because we had a wet, we had a wet lab and we were building software for scientists, 00:07:19.010 --> 00:07:22.835 so we needed to run, to write, you know, code that could be very portable. 00:07:22.915 --> 00:07:27.075 And then, uh, on the server, we were like downloading like many terabytes of images 00:07:27.075 --> 00:07:29.965 and we had like a lot of performance critical routines we were doing there. 00:07:30.655 --> 00:07:32.905 So Rust turned out to be like a really good fit for us. 00:07:32.925 --> 00:07:37.185 And the Rust Python interoperability story was also very good. 00:07:37.805 --> 00:07:39.315 So we were using Py03. 00:07:39.475 --> 00:07:41.495 So we were writing all this stuff in Rust, exposing it. 00:07:42.785 --> 00:07:44.465 over py03 on the Python side. 00:07:44.465 --> 00:07:47.095 So from, from the Python perspective, we were kind of just like, you know, 00:07:47.435 --> 00:07:50.735 calling, calling functions and under the hood, it was all written in Rust. 00:07:50.845 --> 00:07:54.785 Um, and so that was like my first brush with Rust. 00:07:54.835 --> 00:07:58.645 Um, and the thing is like, you're at a startup and you have like so many 00:07:58.645 --> 00:08:03.010 responsibilities, so like, I never really learned Rust in that setting. 00:08:03.030 --> 00:08:06.460 Like I was always, you know, I was going in like making contributions and 00:08:06.460 --> 00:08:10.520 fixing issues, but I was just trying to get in and out as quickly as possible. 00:08:10.869 --> 00:08:15.809 And like the Rust paradigm is just, um, I mean, I continue to believe like 00:08:15.829 --> 00:08:18.859 the learning curve for Rust is very high, especially coming from Python. 00:08:18.859 --> 00:08:24.700 Like most of my career, by the way, like, I mean, professionally, I mean, the 00:08:24.700 --> 00:08:27.980 closest thing I've done professionally is like I wrote Java for a year, right? 00:08:28.020 --> 00:08:35.150 But most of my career has been like Python and TypeScript and I'm not like a I'm not a systems programmer by background at all. 00:08:36.409 --> 00:08:40.559 So it was just a lot of new very new ways of thinking for me. 00:08:41.999 --> 00:08:47.430 I found like again coming back to why I started talking about this in the first place is that Like, one of the 00:08:47.430 --> 00:08:50.890 reasons I wanted to build Ruff was like, I think I need to learn Rust. 00:08:50.910 --> 00:08:53.650 And I think to really learn Rust, I need to build something from scratch. 00:08:53.910 --> 00:08:56.460 And the way that I like to learn is just by doing. 00:08:56.810 --> 00:09:00.790 Um, and, I just know, you know, and then I just spend a few days 00:09:00.790 --> 00:09:03.410 like, banging my head against the wall trying to understand. 00:09:03.780 --> 00:09:09.370 The borrow checker, you know, and like I was working with an AST, like tree structures are like particularly, 00:09:10.060 --> 00:09:13.999 I guess, like, I don't know about annoying, but like, you know, it can be a little bit more challenging. 00:09:14.000 --> 00:09:18.810 And so I just spent a lot of time trying to understand lifetimes and trying to understand what am I doing? 00:09:19.300 --> 00:09:28.480 And, um, uh, you know, I think like at some point, I think like the, one of the first levels of like Rust, one of 00:09:28.480 --> 00:09:37.295 the first milestones I saw in myself was, I knew, um, I wrote some code and I actually knew it wouldn't compile. 00:09:37.620 --> 00:09:38.500 And I knew why. 00:09:38.770 --> 00:09:42.720 And like, I didn't really know how to fix it, but I knew what the, I knew what the error would be. 00:09:43.030 --> 00:09:47.400 Like, that's like a one level of understanding, which is like, you don't quite know like what the solution 00:09:47.400 --> 00:09:50.040 should look like, but like, okay, I know what I'm doing here is not right. 00:09:50.050 --> 00:09:51.420 And I like, I, I know why. 00:09:51.750 --> 00:09:57.564 Tim: I've heard that, I think, in a different context, which is that, um, one of the really big distinguishing 00:09:57.564 --> 00:10:03.515 characteristics between someone that is, I'd say, let's say an intermediate or, you know, starting, very few 00:10:03.515 --> 00:10:06.825 people would be comfortable to say that they are expert Rust programmers. 00:10:07.104 --> 00:10:08.075 I think that's probably the case. 00:10:08.075 --> 00:10:10.015 Charlie: Yes, I would absolutely not say I'm an expert Rust programmer. 00:10:10.025 --> 00:10:14.845 Tim: Um, and, but one of the indicators of someone that has proficiency 00:10:14.964 --> 00:10:22.504 is that they are much faster at diagnosing and like remedying errors. 00:10:22.814 --> 00:10:25.264 Um, it's not to say that they actually have. 00:10:26.130 --> 00:10:36.620 They write fast, or sorry, write better code straight away, it's just that when they encounter problems they, their 00:10:36.620 --> 00:10:42.900 brain is slightly more attuned to the signals that the compiler is going, or it's like something kind of seems 00:10:42.900 --> 00:10:48.290 a little bit iffy with the control flow here, and so I think that the borrow trigger is going to complain. 00:10:49.610 --> 00:10:49.920 Yeah. 00:10:51.670 --> 00:10:55.160 And so that's, that, that it's interesting to see that, that come through. 00:10:55.670 --> 00:11:04.210 I was wondering, and I don't want to put you on the spot too much, but how do you explain lifetimes? 00:11:05.510 --> 00:11:09.640 Like if someone comes to say, Oh, I'm a TypeScript developer, I'm interested 00:11:09.640 --> 00:11:13.820 in Rust, but I'm really, these things don't make any sense to me. 00:11:16.395 --> 00:11:23.475 Charlie: I feel like my answer to that would actually be like I Won't I won't 00:11:23.515 --> 00:11:26.255 actually answer the question because I don't know if I've ever actually had to 00:11:26.264 --> 00:11:26.634 ... Tim: Okay. 00:11:26.955 --> 00:11:31.125 Charlie: ...but I think but one thing I'd think about with Rust and maybe I'll try and tie us back to an answer 00:11:31.655 --> 00:11:41.925 is like I do think that that if you I think if you just like dive in to like expertly written Rust or expose 00:11:41.925 --> 00:11:49.415 yourself actually to like too much of the language, I actually think it's like not that helpful like I wrote just an 00:11:49.415 --> 00:11:58.404 example like I wrote so much Rust code before I actually used even like Box<_>. 00:11:58.724 --> 00:11:58.984 Tim: Yeah. 00:11:59.134 --> 00:12:03.105 Charlie: Like you can just write like a lot of and if you you know If you get dropped into Rust code something you 00:12:03.105 --> 00:12:08.510 see like, okay, I got like box sure fine Then I've got like Rc and I've got Arc 00:12:08.510 --> 00:12:12.220 and I've got like RefCell and like you kind of get exposed to all this stuff. 00:12:12.430 --> 00:12:16.629 Um, even like lifetimes like you can get pretty far explicit lifetimes at least 00:12:16.629 --> 00:12:20.030 like you can get pretty far without like actually using any explicit lifetimes. 00:12:20.390 --> 00:12:31.430 Um, and it, it's funny because I think if you see all that stuff like If you see like Rc and, and I mean it's fine 00:12:31.440 --> 00:12:36.240 to like use this stuff, sure, but like sometimes you see things and you're like, okay, I have a problem in my 00:12:36.240 --> 00:12:41.130 code and like I kind of Google, uh, you know what, what the problem is. 00:12:41.130 --> 00:12:43.749 It's like, I need to be able to mutate this. 00:12:43.790 --> 00:12:46.050 Um, I don't know what's a good example. 00:12:46.069 --> 00:12:49.330 It's like, I need to be able to like mutate this thing without marking it as mut. 00:12:49.685 --> 00:12:50.135 Right? 00:12:50.505 --> 00:12:53.765 And so then people are like, Oh, like you should use like RefCell, right? 00:12:53.765 --> 00:12:54.485 Like blah, blah, blah. 00:12:54.715 --> 00:12:56.625 But, and like, that actually might be right. 00:12:57.025 --> 00:13:00.775 But often when you hit problems in Rust, I find that you're thinking about that. 00:13:00.784 --> 00:13:03.055 You're actually thinking about the problem in the wrong way. 00:13:03.114 --> 00:13:05.194 Like, like someone else should own the data, right? 00:13:05.204 --> 00:13:08.044 Or like the, or like it should be passed in in some other way. 00:13:08.405 --> 00:13:16.504 And so I found that like, it was actually helpful for me to like, ignore a lot of that stuff and like, and it's not like 00:13:16.505 --> 00:13:21.460 I was like that smart about it, but I was like, It was helpful for me to, like, ignore some of that stuff and, 00:13:21.460 --> 00:13:25.570 like, try to, try to, like, uh, force myself to learn things without reaching 00:13:25.570 --> 00:13:31.160 for some of those more advanced tools, um, in order to learn, to learn Rust. 00:13:31.410 --> 00:13:32.710 Tim: Totally agree with you. 00:13:32.790 --> 00:13:41.050 There is a very, I find this, one of the other things is that you encounter an issue and you sort of 00:13:41.050 --> 00:13:43.840 had this decision paralysis, it's like, well, how do I fix this? 00:13:44.150 --> 00:13:49.015 And maybe pausing and thinking about how the code is structured and saying, 00:13:49.015 --> 00:13:53.315 well, actually we want one unique place that has responsibility for cleanup. 00:13:54.485 --> 00:13:54.975 Charlie: Right, right. 00:13:55.315 --> 00:13:56.995 Tim: And how do we enable that? 00:13:57.194 --> 00:14:02.625 Like how we could use a fancy smart pointer or we could tweak our code. 00:14:03.004 --> 00:14:09.515 And, um, yeah, I, I, I like that idea actually of saying, don't be intimidated 00:14:09.545 --> 00:14:16.295 by, um, the very advanced stuff that's, you know, at like the 99th percentile. 00:14:17.574 --> 00:14:22.995 Actually, you could probably stay in the 50th percentile for a long time. 00:14:23.974 --> 00:14:24.704 Charlie: I think you can. 00:14:24.734 --> 00:14:25.134 Yeah. 00:14:25.364 --> 00:14:28.754 And like, and in the context that you're writing code in, like, I don't 00:14:28.754 --> 00:14:32.564 know, I've become like allergic to a lot of things from working on ruff. 00:14:32.584 --> 00:14:34.214 Like, I'm very allergic to cloning. 00:14:34.709 --> 00:14:36.900 Like probably too much, you know what I mean? 00:14:36.910 --> 00:14:42.540 Like now, like we have ruff and we have UV now, which is like a, it's sort of like a pip alternative. 00:14:42.550 --> 00:14:47.310 It's a Python package installer, um, and resolver, uh, written in Rust. 00:14:47.919 --> 00:14:51.959 And like in UV, a lot of the dynamics are really different because there's like way more IO. 00:14:52.419 --> 00:14:56.240 Um, like we're downloading packages and like zipping, right. 00:14:56.240 --> 00:14:57.040 And like blah, blah. 00:14:57.040 --> 00:15:02.435 And so like cloning is actually like, not nearly as much of a big deal. 00:15:02.965 --> 00:15:08.445 Um, whereas in rough, if we're, if we do like a lot of, a lot of cloning or a lot of allocations in a hot, 00:15:08.565 --> 00:15:13.435 in a hot path, like it immediately shows up in all the performance profiles, like it's very obvious. 00:15:14.025 --> 00:15:17.175 Um, and so like ruff kind of trained me in a certain way. 00:15:17.469 --> 00:15:23.269 Um, and it's funny because now I work on UV and like I have people on the team who are like complaining about 00:15:23.279 --> 00:15:28.909 all the lifetimes They're like you should all just be RC or like blah blah blah or like we should just clone 00:15:28.909 --> 00:15:32.180 here or like blah blah And like they're actually they're actually right. 00:15:32.389 --> 00:15:38.249 Like I think I have like over rotated on some of the things I learned from working on ruff But it's been 00:15:38.249 --> 00:15:44.765 interesting but like yeah again, I think Like, you can clone your way out of problems if you're learning 00:15:44.765 --> 00:15:48.715 Rust, right, as a way to get yourself to move along, I think, sometimes. 00:15:48.785 --> 00:15:53.564 Um, and maybe the expert level solution lets you avoid some of those allocations, but like, often it won't really matter. 00:15:54.405 --> 00:15:54.834 Tim: Sure. 00:15:54.984 --> 00:16:03.604 Hey, I was wondering if we could segue slightly back to, uh, your career and some of the things that 00:16:03.734 --> 00:16:08.804 you did beforehand, because there's an overarching suspicion that I have. 00:16:08.804 --> 00:16:13.144 I've got a couple of assumptions about the kind of person that you are. 00:16:13.554 --> 00:16:22.604 Um, so the, uh, and I just want to essentially either validate that, um, Because it seems like a lot of the 00:16:22.604 --> 00:16:31.724 decisions that you've made have been very centric, uh, very, you're, you're, you, correct me if I'm wrong, but you 00:16:31.724 --> 00:16:36.155 worked at a health science, uh, machine learning startup for diagnostics. 00:16:36.245 --> 00:16:41.734 Um, and, uh, before that you worked at the Khan Academy. 00:16:42.584 --> 00:16:42.974 Charlie: That's right. 00:16:43.194 --> 00:16:50.844 Tim: And that implies to me that you've had other people at the front of mind for a long time. 00:16:51.124 --> 00:17:01.194 And, and I don't want to be too prejudiced against, uh, so I'm gonna, and, but you, I saw that your undergraduate at least 00:17:01.194 --> 00:17:05.994 is at Princeton, but you probably made a deliberate choice not to do grad school. 00:17:06.034 --> 00:17:06.805 Was that correct? 00:17:07.865 --> 00:17:08.654 Charlie: Yeah, that's right. 00:17:08.814 --> 00:17:09.024 Yeah. 00:17:10.254 --> 00:17:11.494 Tim: So do you want to talk me through 00:17:11.914 --> 00:17:14.744 ... Charlie: ...that career path and why I made these various decisions? 00:17:14.744 --> 00:17:15.214 Yeah. 00:17:15.885 --> 00:17:16.844 Yeah, yeah, of course. 00:17:16.895 --> 00:17:23.020 Um, well, I, uh, I interned at Khan Academy, um, after my junior year 00:17:23.020 --> 00:17:31.979 of college and, um, that, uh, that experience was just really, really good. 00:17:32.590 --> 00:17:32.910 Tim: Cool. 00:17:32.950 --> 00:17:36.760 Charlie: And so, um, uh, you know, I wanted, it was, it was pretty clear to me that I wanted to go back and 00:17:36.760 --> 00:17:43.210 join full time, but I think the things that made that, um, uh, a really appealing decision to me, you know, 00:17:43.210 --> 00:17:49.120 one was, um, I try and a lot of these career decisions, um, around impact. 00:17:49.405 --> 00:17:55.275 So what are things I can do to have a significant positive impact on, on people in the world? 00:17:55.335 --> 00:17:59.555 Um, and uh, Khan Academy is something that I had used, um, you know, a lot. 00:17:59.564 --> 00:18:01.974 It was really helpful for like, for my own education. 00:18:02.015 --> 00:18:07.765 Um, and so the opportunity to go and work there in terms of like impact, you know, was really, um, really appealing to me. 00:18:08.260 --> 00:18:14.630 Um, you know, the other pieces to it were, um, Khan Academy, um, had 00:18:14.630 --> 00:18:18.070 a really, really strong engineering team and engineering culture. 00:18:18.500 --> 00:18:23.010 And so like most of the things, like, I think. 00:18:23.780 --> 00:18:26.500 Khan Academy is where I learned, you know, at least for one 00:18:26.500 --> 00:18:29.790 definition, like what it means to be like a great software engineer. 00:18:29.820 --> 00:18:34.640 Like, I don't think it was necessarily me at that point in time, but like I got to work with a lot of really amazing people. 00:18:35.070 --> 00:18:39.209 Tim: I can't believe that you are not a person who believes that they 00:18:39.209 --> 00:18:42.820 were, you know, a 10X developer, um, in their very first gig. 00:18:43.200 --> 00:18:46.450 Charlie: I think I was a good, I think I was a good software engineer, but I was also pretty young, right? 00:18:46.450 --> 00:18:49.200 And I was also learning and everything was new. 00:18:49.680 --> 00:18:52.550 But like, you know, the, just the quality of people. 00:18:52.580 --> 00:18:54.360 Tim: So what's it like, the culture? 00:18:54.389 --> 00:18:56.030 So you said they have a strong engineering culture. 00:18:56.040 --> 00:18:57.580 Is it 100 percent remote? 00:18:57.699 --> 00:19:00.560 Is it all, uh, in a building? 00:19:00.680 --> 00:19:01.360 Charlie: This is a while ago. 00:19:01.360 --> 00:19:03.149 This is like 2015, 2016. 00:19:03.150 --> 00:19:04.389 So it's a while ago at this point. 00:19:04.400 --> 00:19:06.899 But at that point in time, it was about a third remote. 00:19:06.950 --> 00:19:09.070 The rest was in Mountain View in California. 00:19:09.580 --> 00:19:14.760 Um, and there were just some really, yeah, some really amazing people who kind of anchored a lot of the 00:19:14.760 --> 00:19:19.230 culture, like, um, like Ben Kamens, who is the VP of engineering. 00:19:19.230 --> 00:19:24.039 And then he went on to found the company that I joined after Khan Academy, which is how I got, how I got sort 00:19:24.040 --> 00:19:27.740 of roped into, um, that, which is, you know, we can talk about that too. 00:19:27.810 --> 00:19:35.440 Um, uh, uh, John Resig, who's like the creator of jQuery, um, still at Khan Academy, I think, um, Craig 00:19:35.440 --> 00:19:41.330 Silverstein, who was like the, The first employee at Google and then joined, he's just like, it's incredible. 00:19:41.340 --> 00:19:47.879 Software engineer, um, Sophie Alpert, who was like for a long time, the number one committer to react outside 00:19:47.879 --> 00:19:50.930 of Meta and then joined React and manage the React team eventually. 00:19:50.930 --> 00:19:53.539 So it was just like a very, very high quality group of people. 00:19:53.899 --> 00:19:59.780 Um, and there was a lot of, um, emphasis just put on, on, you know, on the culture. 00:19:59.790 --> 00:20:02.980 And I think I, I think everything from like our career development guide. 00:20:04.090 --> 00:20:08.610 I just learned so much by looking at like what is this company value and like what are the traits that someone 00:20:08.620 --> 00:20:12.129 has that demonstrates that they are, for example, a senior engineer? 00:20:12.129 --> 00:20:13.999 Like what does it mean to be able to own? 00:20:14.299 --> 00:20:16.899 What is, what is this concept of ownership that people keep talking about? 00:20:16.899 --> 00:20:25.135 And it just really helped me understand, um, uh, you know, the, the, uh, I had a lot of role models, I think, is what I'll 00:20:25.135 --> 00:20:31.475 say, and I had, it was a very good setting for me to kind of grow and, and, and, and learn how to be, um, a better software 00:20:31.475 --> 00:20:34.715 engineer and not just technically, but in terms of taking on more ownership and 00:20:34.725 --> 00:20:39.475 being, um, growing my maturity, um, my ability to execute on large projects. 00:20:39.994 --> 00:20:41.575 Um, but I worked.. 00:20:41.695 --> 00:20:42.355 Tim: Software. 00:20:42.535 --> 00:20:43.615 Oh, sorry, excuse me. 00:20:44.160 --> 00:20:46.120 Charlie: Was just gonna say I worked like all over the stack 00:20:46.120 --> 00:20:50.070 there because I started I started doing like mostly doing front end. 00:20:50.070 --> 00:20:53.410 We were like a really big React company, which is I guess everyone 00:20:53.410 --> 00:20:57.210 is now sort of but like that was a little bit less the case at the time. 00:20:58.499 --> 00:21:02.120 You know, I did like I worked on our first like iOS apps a bit. 00:21:02.130 --> 00:21:04.220 I did like a year of Android, too. 00:21:04.325 --> 00:21:05.885 Tim: Oh, which is where you would have done Java. 00:21:05.905 --> 00:21:06.635 Charlie: Yeah, exactly. 00:21:06.645 --> 00:21:06.905 Yeah. 00:21:06.905 --> 00:21:12.145 And then, and then our, our backend was all Python and we were like, we just had like a huge Python code base. 00:21:12.175 --> 00:21:18.335 Um, uh, and so I kind of got exposed to, you know, all those different uh, parts of the development stack. 00:21:18.889 --> 00:21:25.430 Um, which, so for me, it was like kind of the perfect place to be an engineer starting out my career and, and I still 00:21:25.760 --> 00:21:32.460 lean on a lot of the things I learned there about like being a, being a great and productive engineer, um, today, 00:21:32.470 --> 00:21:35.389 especially as I'm looking at, like, how do we define our own culture as a company? 00:21:35.659 --> 00:21:42.289 Tim: Yeah, I mean, that must be quite intimidating because it's a significantly different set of skills, 00:21:42.629 --> 00:21:48.100 essentially being asked to be the senior leader, especially in a small company. 00:21:48.800 --> 00:21:54.820 And you're founding this from the east coast of the United States, outside of Mountain View. 00:21:54.879 --> 00:21:55.489 Charlie: That's right, yep. 00:21:55.800 --> 00:21:59.370 Tim: Most people are building software companies outside of Silicon Valley. 00:22:01.429 --> 00:22:09.625 So, do you have any, you know, like, pearls of wisdom, or any bits and pieces for people that are kind of, maybe at 00:22:09.625 --> 00:22:15.075 a four to nine people, you know, like sub 10 individual company that are 00:22:15.145 --> 00:22:20.645 growing healthily to really set that foundation for like the next 10 years. 00:22:21.164 --> 00:22:24.194 Charlie: So for us, um, yeah, I'm based in New York. 00:22:24.204 --> 00:22:28.544 Our team is nine full time, um, including me. 00:22:28.704 --> 00:22:35.744 Um, and no, like no two people live in the same state. 00:22:36.225 --> 00:22:40.020 Um, and outside the US, no two people live in the same country. 00:22:40.020 --> 00:22:46.379 So like we have about half the teams in the US, half the team is, um, is, uh, international outside the US. 00:22:46.669 --> 00:22:55.395 Um, and we span time zones from, Pacific through to, um, IST, like, uh, Bangalore would be the furthest. 00:22:56.155 --> 00:22:58.155 So like, it's a very large range of time. 00:22:58.155 --> 00:22:58.305 So... 00:22:58.855 --> 00:23:01.045 Tim: I'm sure your accountant is very happy with you. 00:23:01.054 --> 00:23:01.424 Charlie: Yeah. 00:23:01.644 --> 00:23:09.275 Um, it has that, I think it's sort of like the least of the challenges, but yeah, that is, that is one of the challenges. 00:23:09.715 --> 00:23:13.065 Um, I mean, I think for us, like the main thing that the main things 00:23:13.065 --> 00:23:18.355 I focused on are one, um, we've grown this team very deliberately. 00:23:18.689 --> 00:23:29.280 So like, um, you know, I think in March of last year, uh, that was when I first hired anyone apart from myself. 00:23:29.280 --> 00:23:32.629 And we grew, I hired two people, like in March we grew from one to three. 00:23:32.939 --> 00:23:36.199 And then from, from March, so over the past, you know, a little over a year, 00:23:36.199 --> 00:23:39.610 we grew from like three to nine and, you know, it was incremental, right? 00:23:39.610 --> 00:23:44.405 It was like, One month we had four, and then two months later we had five, and then a few months later we had six. 00:23:44.625 --> 00:23:50.884 Um, and, you know, even like three to six to nine, those are all like really different companies. 00:23:50.885 --> 00:23:56.995 Um, yeah, and so everyone's like, you know, some people are rolling their eyes being like nine is not a 00:23:56.995 --> 00:24:02.675 big company, but like, if you were three, then like suddenly nine just, you know, like a lot of the processes 00:24:02.685 --> 00:24:05.105 that you put in place when you were three like don't work anymore. 00:24:05.415 --> 00:24:11.745 Um, and so, being flexible about how you change, like how you work, I think is really important. 00:24:11.835 --> 00:24:16.115 Um, yeah, I think one of the main things that we've tried to focus on, especially being a distributed 00:24:16.115 --> 00:24:21.805 company, is, um, uh, And it sounds cliche, but I'll give some examples. 00:24:21.855 --> 00:24:26.045 Like, I think in order to work effectively, like, asynchronously, 00:24:26.550 --> 00:24:30.010 You have to have a team that really like trusts each other a lot. 00:24:30.510 --> 00:24:37.810 Um, and my favorite example of trust or a very concrete example of the thing that we try to do is like, If you're 00:24:37.810 --> 00:24:43.140 reviewing someone's pull request and you think it's close, but you want some things to change, like approving it 00:24:43.140 --> 00:24:49.789 with feedback rather than like leaving comments or requesting changes is like a way more effective thing to do. 00:24:50.359 --> 00:24:55.869 Um, because it can save like a whole day in turnaround time, because imagine you get some comments, then you go to sleep, 00:24:55.870 --> 00:24:59.269 they wake up, they address the comments, you wake up, eventually you approve it. 00:24:59.269 --> 00:25:01.339 They're asleep then they wake up and merge it. 00:25:01.339 --> 00:25:01.589 Tim: Right. 00:25:01.819 --> 00:25:06.040 So like, But that requires trust, like you have to be able to trust 00:25:06.050 --> 00:25:09.810 and have good norms and establish norms around, if I give feedback. 00:25:10.534 --> 00:25:13.264 Charlie: First of all, I know that they're actually going to look at it, take it seriously. 00:25:13.294 --> 00:25:17.135 They don't have to address it, but I can trust that if they don't, they 00:25:17.135 --> 00:25:20.054 will at least consider it and have good reasons for, for not addressing it. 00:25:20.054 --> 00:25:21.185 And they will tell me why. 00:25:21.485 --> 00:25:27.254 So like trying to build a culture that's really based on trust, I think, I think if you like have a culture that's 00:25:27.254 --> 00:25:32.915 really based on distrust, um, or, uh, or not, or, or at least not based on trust, the alternative is very bad. 00:25:32.915 --> 00:25:33.075 Tim: Yeah. 00:25:33.075 --> 00:25:34.855 The alternative is unhealthy, right? 00:25:34.865 --> 00:25:36.294 Charlie: And also I just think way harder. 00:25:36.515 --> 00:25:43.445 Like, it's hard to work asynchronously effectively, um, you have to be able to trust people to like make good responsible 00:25:43.445 --> 00:25:49.905 decisions and also be respectful of, uh, you know, the feedback you give and, uh, and, and your own opinions. 00:25:50.175 --> 00:25:53.394 So, um, that's been a really important thing for us. 00:25:53.415 --> 00:26:02.155 Tim: That is a very hard thing to maintain and it's much harder when you, if you don't start in the right place. 00:26:03.495 --> 00:26:07.235 And I can speak from my experience of working at a very large company. 00:26:08.850 --> 00:26:12.860 When essentially individual teams or individual departments are almost 00:26:12.860 --> 00:26:18.110 at war with each other because of the way that the metrics work. 00:26:19.129 --> 00:26:27.299 Essentially, you lose a lot of faith that the people outside of your sphere of influence have your interests at heart. 00:26:27.340 --> 00:26:33.050 Whereas it's nice to know that Our colleagues really care about the product. 00:26:33.050 --> 00:26:35.350 We really care about our customers and our users. 00:26:35.929 --> 00:26:37.810 Uh, so that's, that's really positive. 00:26:37.810 --> 00:26:44.610 One of the other things though, approving with comments, I like that approach, but I did have one instance 00:26:44.610 --> 00:26:51.050 in my career when people had one, I suppose, yeah, I guess we call him 00:26:51.050 --> 00:26:55.740 my manager, uh, engineering manager, let's say he would always approve. 00:26:56.800 --> 00:27:01.830 But sometimes it would be like a tacit, like a very like, but this needs to change. 00:27:01.909 --> 00:27:07.899 And I was of the opinion, I was like, actually, if you need significant refactoring, I would rather that you 00:27:07.899 --> 00:27:16.520 hit the changes required button because that like sets me up with this mental model of, oh, okay, so that's fine. 00:27:16.860 --> 00:27:18.139 You're being honest with me. 00:27:18.490 --> 00:27:27.814 Um, that's another way of, um, Uh, I, I think respecting the protocol and like, not being afraid. 00:27:27.894 --> 00:27:31.194 I like rejection is a very, very, like, I don't think I've ever had a 00:27:31.194 --> 00:27:37.814 PR explicitly said someone said no to, but, um, this does not make, but 00:27:39.364 --> 00:27:45.154 Charlie: I mean, if you've contributed to open source, you probably have at some point, but yeah, no, I don't. 00:27:48.169 --> 00:27:53.090 It's much more likely that your patch is just going to sit there for a long time. 00:27:53.479 --> 00:28:01.229 Tim: Um, so do you want to talk me through the relationship with Zip VC funding and open source? 00:28:01.549 --> 00:28:04.600 Because this must have been something you've thought about personally. 00:28:04.934 --> 00:28:07.944 And, um, I'm, I'm, I'm really interested in your take. 00:28:08.034 --> 00:28:08.724 Charlie: Yeah, totally. 00:28:08.784 --> 00:28:11.724 Can I react to the thing you were saying before though, quickly before we... 00:28:11.814 --> 00:28:12.174 Tim: Oh, sure. 00:28:12.324 --> 00:28:12.634 Charlie: Yeah. 00:28:12.844 --> 00:28:19.974 So I think like, um, a few things on the, like, it's actually helpful when people request changes, like that general theme. 00:28:20.424 --> 00:28:22.674 Um, first of all, yeah, definitely. 00:28:22.694 --> 00:28:27.964 Like, I think there's a lot of, um, implicit norms that get like established on a team. 00:28:28.324 --> 00:28:34.115 Um, and I, I think if you, if someone puts up a PR and, you know, There's actually like important like discussion 00:28:34.115 --> 00:28:40.764 or dialogue to be had about some of the core decisions, then you should not approve it But part of it is like well, 00:28:40.764 --> 00:28:45.924 like you should you should signal that like hey, this is what I'm thinking about blah blah Sometimes it can be a sign of 00:28:45.924 --> 00:28:52.635 process failure I think if you like get to a PR and there's like a significant disagreement about design Sometimes it 00:28:52.635 --> 00:28:58.715 depends like what the PR is supposed to be and what you're trying to accomplish with it But but but another part here 00:28:58.715 --> 00:29:02.985 is like If you're gonna build a culture that's based on trust, like, you start 00:29:02.985 --> 00:29:06.074 with a certain level of trust, but then it needs to be, like, built up over time. 00:29:06.334 --> 00:29:10.975 So, like, you sh part of it is, like, we assume good intent and we assume trust to start. 00:29:11.570 --> 00:29:15.510 There are things you can do to, like, lose people's trust, and there are things you can do to, like, build people's 00:29:15.510 --> 00:29:22.179 trust up further, and so, you know, early on, like, we might, um, start with a certain tolerance or a certain 00:29:22.179 --> 00:29:26.139 level of, like, we're gonna prove this with feedback, or, like, we're gonna, let's have some discussion about blah, 00:29:26.139 --> 00:29:31.499 blah, and, like, use some of these as learning opportunities, um, but, but I guess for us it's, like, start from 00:29:31.499 --> 00:29:34.759 a high degree of trust and try to, like, build it up over time, and when 00:29:34.759 --> 00:29:38.860 we see people lose trust, then try to figure out ways that we can restore it. 00:29:40.120 --> 00:29:50.759 Tim: Yeah, sure, because I guess trust tends to neutral over time with like a lack of intervention. 00:29:50.969 --> 00:29:57.959 I can see that happening on both angles in relationships where there might have been a significant disagreement, well 00:29:57.959 --> 00:30:02.195 one of the You know, common patterns is that people will essentially not 00:30:02.205 --> 00:30:04.715 talk to each other for a couple of months and then be able to come back. 00:30:04.715 --> 00:30:09.884 It doesn't work so much in an organizational setting, but, um, Are 00:30:09.885 --> 00:30:14.955 you actively looking for signals that trust is being maintained between? 00:30:15.450 --> 00:30:16.070 Your stuff. 00:30:16.070 --> 00:30:22.500 I mean, I don't probably, I don't want to peer too much into the company, but, um, that, that would be kind of interesting. 00:30:22.570 --> 00:30:23.150 Charlie: I do. 00:30:23.150 --> 00:30:23.350 Ya. 00:30:23.610 --> 00:30:28.589 I mean, I look at like, I mean, the funny thing is, right, that, like, we're remote. 00:30:28.590 --> 00:30:32.100 So like most of our, like almost all of our activity and interaction, 00:30:32.749 --> 00:30:37.680 um, I mean, a very, very large portion of it is like fully public. 00:30:38.140 --> 00:30:42.890 Um, because we don't have, we have one private Git repo and it's our 00:30:42.990 --> 00:30:46.950 company website and everything else that we do is public, right? 00:30:46.990 --> 00:30:54.119 And, um, we developed UV, that was a private repo from like October to February and then we released it. 00:30:54.420 --> 00:30:58.080 But like the whole Git history and all that discussion, everything is there. 00:30:58.470 --> 00:31:04.390 Um, and then, uh, so like a lot of what we do is just like fully public. 00:31:04.780 --> 00:31:09.100 And then, um, you know, all of our internal communications, they're private 00:31:09.110 --> 00:31:11.900 to outside the company, but you know, it's all public within the company. 00:31:12.180 --> 00:31:16.530 Um, so like, uh, because we're distributed, it's just like, there's a 00:31:16.530 --> 00:31:20.789 lot of visibility into like what, how, how people are engaging and, and all that. 00:31:20.829 --> 00:31:28.900 Um, and you know, I look at, I, I, I think like being a leader, a lot of it's like trying to pick up on small signals. 00:31:29.170 --> 00:31:29.410 Right. 00:31:29.410 --> 00:31:34.090 So like looking at how, and not in a way where I'm like snooping on people, but like, okay, when people 00:31:34.090 --> 00:31:39.460 engage in a PR, like what's going on, like someone requested changes or like, was that reasonable? 00:31:39.460 --> 00:31:40.559 Like, how do I think about that? 00:31:40.559 --> 00:31:44.409 And then, um, or like, are they treating these different people on 00:31:44.410 --> 00:31:48.499 the team fairly, or like, did this person, uh, take this feedback well? 00:31:48.500 --> 00:31:53.220 And, and, and so I do like look at all that stuff and I think about it and I talk about it with people on the team. 00:31:53.260 --> 00:31:59.725 Um, and then I guess the other piece, it's a little bit different, but like, I don't know, sometimes I try and do things, 00:31:59.775 --> 00:32:06.105 maybe this is stupid, but like, sometimes I try and do things to like exhibit behaviors that I think are good, um, 00:32:08.205 --> 00:32:08.585 yeah. 00:32:08.585 --> 00:32:10.845 Tim: Yeah, look, I think modeling is important, right? 00:32:11.065 --> 00:32:15.065 As long as it isn't you attempting to project something that doesn't exist. 00:32:15.294 --> 00:32:22.310 And it strikes me that you're actually projecting, you're exhibiting the way that, you know, 00:32:22.310 --> 00:32:28.750 some idealized state, potentially, but, um, but that's, that's okay. 00:32:28.900 --> 00:32:29.759 You can create. 00:32:29.860 --> 00:32:31.920 You're not going against the grain, really. 00:32:31.929 --> 00:32:33.679 You're just kind of reinforcing. 00:32:34.099 --> 00:32:38.819 It's interesting to hear that, you know, sometimes you will go and read the comments, and I think this 00:32:38.820 --> 00:32:45.289 is really important whenever I've supervised or met with people, or even when I've gone for a promo, promotion. 00:32:46.235 --> 00:32:48.755 I've had feedback at like interview loops. 00:32:48.765 --> 00:32:54.185 It's like, oh, no, no, we've, we've looked through your interactions and they're really, really positive. 00:32:54.875 --> 00:32:59.694 So that's, uh, a really good sign for us that you'll be a good fit for the 00:32:59.695 --> 00:33:04.115 team because when there are changes that you'd like, you're very, uh, tactful. 00:33:04.144 --> 00:33:10.425 And so that's, um, so just, just a hint, I suppose, to everyone that, 00:33:10.425 --> 00:33:16.455 um, Being thoughtful and mindful of the way that our messages are, 00:33:18.535 --> 00:33:23.449 you know, that has an impact that essentially can last longer than your tenure at the company. 00:33:23.780 --> 00:33:25.250 Charlie: Um, yeah, yeah, totally. 00:33:25.250 --> 00:33:28.160 Tim: The git, the git history is gonna be there for a while. 00:33:28.850 --> 00:33:29.420 . Um... 00:33:29.840 --> 00:33:30.100 Charlie: Ya. 00:33:30.660 --> 00:33:30.880 Ya. 00:33:32.100 --> 00:33:39.870 Tim: So the machine learning company that you were at, you wrote a blog post in 2022 talking about mypy. 00:33:40.110 --> 00:33:40.680 Charlie: Oh, yeah. 00:33:40.680 --> 00:33:40.740 Yeah. 00:33:41.940 --> 00:33:44.700 Tim: And uh, and static typing. 00:33:44.965 --> 00:33:47.845 So this is obviously, so we're sort of several years down the track. 00:33:47.845 --> 00:33:56.845 So you started interning, I guess, 2015, um, by, by 2022, you had become a senior engineer with a lot of, 00:33:57.965 --> 00:34:05.075 with sort of incubated, let's say within, uh, this context of lots of very significant role models. 00:34:05.235 --> 00:34:14.920 And now you, you are building, uh, a very important product and you really want to make sure that your code is robust. 00:34:15.060 --> 00:34:18.210 And so what was your main message in that blog post? 00:34:18.250 --> 00:34:24.980 Charlie: Um, so yeah, so that originated, yeah, after Khan Academy, I worked at a company called Spring Discovery. 00:34:25.040 --> 00:34:30.010 Um, so drug discovery and diagnostics company, um, all based on computer vision. 00:34:30.010 --> 00:34:36.190 So we were like taking very high resolution pictures of cells and then trying to train, um, models on 00:34:36.190 --> 00:34:43.630 them to understand what happens when you introduce drugs or add viruses or whatnot um, and in that context I 00:34:43.630 --> 00:34:48.389 was maintaining a lot of our software infrastructure machine learning infrastructure data infrastructure. 00:34:48.389 --> 00:34:55.025 So we just had Um, a lot of Python, uh, because we were doing scientific computing and so all our infrastructure was also 00:34:55.025 --> 00:35:01.314 Python, um, and we were a pretty small, like, core engineering team, like, uh, you know, like the, let's say the company 00:35:01.315 --> 00:35:07.495 as a whole was like 25 to 30, the sort of, um, computational team, which was 00:35:07.505 --> 00:35:10.334 like software engineering, data science, machine learning, was like eight. 00:35:10.885 --> 00:35:15.755 And like two of us were basically responsible purely for infrastructure and the rest were in some, were either 00:35:15.755 --> 00:35:19.785 like machine learning researchers who were writing, who were building some of their own infrastructure, but, 00:35:20.235 --> 00:35:22.675 uh, they were either machine learning researchers or data scientists, right? 00:35:22.675 --> 00:35:27.615 Whose questions were to, uh, leverage data to make discoveries and answer questions. 00:35:27.635 --> 00:35:28.519 So we were. 00:35:28.610 --> 00:35:33.020 We were building a lot of infrastructure for them and we were building infrastructure for, um, like the wet 00:35:33.020 --> 00:35:36.320 lab scientists who would then look at all the data that we generated and, and 00:35:36.320 --> 00:35:40.689 again, try to answer questions like which drugs are most promising on X or Y axes. 00:35:41.290 --> 00:35:49.010 Um, so we were a small team trying to do a lot and like the main way that we got a big lift was, was tooling, right? 00:35:49.019 --> 00:35:51.720 It was like static typing was like a big part of it. 00:35:51.720 --> 00:35:57.569 We had, um, at least by my standards, a pretty large and extremely heavily typed code base, and that 00:35:57.599 --> 00:36:01.865 enabled us to to, um, I mean, Python typing, we talk about it a lot. 00:36:01.875 --> 00:36:03.305 There's like a lot of opinions on it. 00:36:03.345 --> 00:36:06.165 Um, at least in that context, it, it enabled us to do a lot of 00:36:06.165 --> 00:36:08.815 refactors that otherwise would have been completely impossible. 00:36:08.845 --> 00:36:14.205 Tim: So you made the claim, I don't, not so much the claim, but that you. 00:36:14.540 --> 00:36:24.809 had probably the most strict, you set MyPy to be as difficult as possible for your developers for a code base 00:36:24.810 --> 00:36:28.060 that was roughly 300, 000 lines of code, give or take, let's say. 00:36:28.420 --> 00:36:32.939 And so that's a significant amount of code maintained by a pretty small team. 00:36:33.319 --> 00:36:40.939 And you decided to run, do Python, let's say on hard mode and you would disallow any, for example, 00:36:42.029 --> 00:36:43.750 Charlie: Yeah, yeah, yeah. 00:36:43.750 --> 00:36:46.870 So like the any type, yeah, we didn't allow that. 00:36:47.000 --> 00:36:54.229 Um, uh, and it's, it's very hard to, yeah, it's, I, who knows if that claim is true. 00:36:54.229 --> 00:36:56.879 It's probably not, but like we had a very, we had basically, we 00:36:56.880 --> 00:36:59.929 had a very large Python code base that was extremely strictly typed. 00:37:00.029 --> 00:37:07.609 Um, and, uh, you know, that blog post really was about the, it's sort of a common theme with some of the Python 00:37:07.609 --> 00:37:12.910 tooling, which is that, and it's one of the reasons really, probably the primary motivation for starting to 00:37:12.910 --> 00:37:19.040 work on ruff was that, um, the tooling just really struggled to scale with the systems that we were building. 00:37:19.739 --> 00:37:27.205 Um, and in ruff, you know, I think ruff has since actually grown to ... have a lot of other value propositions that 00:37:27.205 --> 00:37:30.885 aren't just performance, some of which were intentional, some were unintentional. 00:37:31.185 --> 00:37:35.345 Um, but, uh, but performance was really like the big one that we started with. 00:37:35.704 --> 00:37:40.054 Um, the unintentional, I think you're reacting to this idea of like, some 00:37:40.055 --> 00:37:44.225 of them are unintentional, like, rough bundles a lot of tools together. 00:37:44.275 --> 00:37:51.760 So like sometimes when we see people adopt rough, they might replace like 20 or 30 dependencies, like individual tools, um, 00:37:51.810 --> 00:37:58.779 because in Python, um, the, the tooling landscape is, is very fragmented, um, I guess fragmented is kind of pejorative. 00:37:58.899 --> 00:38:05.109 Um, there are lots of different tools that people kind of compose together to get to their end user, their end experience. 00:38:05.379 --> 00:38:06.910 You know, there are different linters. 00:38:06.910 --> 00:38:08.080 The linters have lots of plugins. 00:38:08.379 --> 00:38:10.100 Import sorting is a separate tool. 00:38:10.120 --> 00:38:11.669 Code formatting is a separate tool. 00:38:11.700 --> 00:38:13.169 Type checking is a separate tool. 00:38:13.629 --> 00:38:16.600 There are separate tools for like modernizing your code. 00:38:16.779 --> 00:38:20.180 Um, there's just like a lot of different tools that people tend to chain together. 00:38:20.660 --> 00:38:21.970 And in Ruff, we ended up. 00:38:22.359 --> 00:38:23.979 Basically bundling all that stuff together. 00:38:23.979 --> 00:38:28.870 So it's like one tool that you download and install and learn and exposes all those capabilities. 00:38:29.509 --> 00:38:34.819 And that was sort of an unintentional thing because it was very hard to build a plugin system. 00:38:36.129 --> 00:38:40.359 Like early on, I was like, I don't really know like how to build a plugin system here. 00:38:40.430 --> 00:38:45.209 It's just like, there are, it's actually a pretty interesting design space of like, how can you build a plugin system? 00:38:45.570 --> 00:38:46.960 For a tool that's written in Rust. 00:38:46.980 --> 00:38:50.490 And there are actually a lot of interesting options there and we may do it eventually. 00:38:50.770 --> 00:38:52.570 But early on, I was like, I don't really know how to do this. 00:38:52.570 --> 00:38:55.640 So like, but this project wants to use my tool and they need this functionality. 00:38:55.640 --> 00:38:57.120 So I'm just going to build it into the tool. 00:38:57.440 --> 00:39:00.520 And like, we ended up, we kind of ended up bundling a lot of this stuff by accident. 00:39:00.659 --> 00:39:03.809 It became a very big part of the value proposition because especially 00:39:03.809 --> 00:39:05.980 for small projects, maybe they don't care at all about performance. 00:39:06.475 --> 00:39:10.875 But like the idea of being able to greatly simplify their toolchain is really useful. 00:39:11.395 --> 00:39:16.075 So, um, that was something that was a little bit unintentional, but has become a big part of what we're 00:39:16.075 --> 00:39:21.685 trying to do, which is like a lot of, we want to simplify a lot of the tooling by bundling things together. 00:39:22.065 --> 00:39:27.485 Um, and it's similar for packaging, like packaging again, Um, for any one thing you want to do in Python, there's like 00:39:27.505 --> 00:39:30.759 10 different tools you could use and you might need like three or four of them. 00:39:30.759 --> 00:39:33.475 And you have to figure out like which ones and you have to understand. 00:39:35.185 --> 00:39:39.005 We don't need to talk about Python packaging, but it's, and again, it's a similar thing where like we 00:39:39.005 --> 00:39:44.995 want to ship like UV, like right now I would say it can comfortably replace like three or four things. 00:39:45.145 --> 00:39:47.635 Like instead of three or four separate things, you can just use UV. 00:39:47.964 --> 00:39:54.415 And like, we want to like expand the scope of the number of things it can do so that it's more of a self contained experience. 00:39:54.734 --> 00:40:01.874 Um, but that was a funny thing to realize over time was, um, we started from this 00:40:01.884 --> 00:40:05.725 position of it's hard to build plugins, so let's just implement everything ourselves. 00:40:06.204 --> 00:40:07.734 Um, and that turned into a feature. 00:40:08.394 --> 00:40:17.804 Tim: So how do you introduce a tool to an ecosystem without coming across as trying to invade. 00:40:19.265 --> 00:40:20.904 Charlie: Yeah, that's a really good question. 00:40:20.935 --> 00:40:27.055 And actually something I've been, I've been very grateful or like, I don't know, I feel lucky, I guess. 00:40:27.095 --> 00:40:32.125 Like the Python ecosystem in general has been very welcoming to, like, our work 00:40:32.125 --> 00:40:35.345 and the stuff that we're building and it definitely didn't have to be that way. 00:40:35.415 --> 00:40:41.415 Like I could also see a world where people were, really, there was a lot more pushback and people were a lot more upset. 00:40:42.665 --> 00:40:47.705 And in particular for me because like I haven't really been like part of the Python ecosystem. 00:40:48.645 --> 00:40:52.480 Like I've been average Python consumer, right? 00:40:52.480 --> 00:40:56.770 I was like writing a lot of Python code, but like, I'm not reading the enhancement proposals. 00:40:56.830 --> 00:40:58.980 I'm not on the Python forum. 00:40:59.369 --> 00:41:03.450 Um, I'm not like a contributor even to any like Python. 00:41:03.499 --> 00:41:04.820 This is before I started on ruff. 00:41:04.840 --> 00:41:07.079 Like I wasn't like contributing to open source even, right. 00:41:07.079 --> 00:41:11.060 I was like average passive consumer of open source and of Python. 00:41:11.300 --> 00:41:12.960 That is the vast majority of people. 00:41:13.200 --> 00:41:18.720 Yes, it is the vast majority of people, um, which is something that I have to kind of remind myself of over time 00:41:18.720 --> 00:41:23.880 because now I've, now I like know a lot of people in the ecosystem and I'm like way more plugged into what's going 00:41:23.880 --> 00:41:29.559 on and I'm a lot more of an expert in the standards and, and things like that and you have to, you have to like 00:41:29.909 --> 00:41:33.740 remind yourself that like most people do not come from that, um, perspective. 00:41:33.740 --> 00:41:37.339 Like I wouldn't have known, I didn't know the difference between like a built, I didn't, I didn't know the difference 00:41:37.339 --> 00:41:42.450 between a built distribution and a source distribution in Python, like until I started working on that stuff. 00:41:42.730 --> 00:41:52.929 Um, so anyway, but, um, I, it's been interesting to come into that ecosystem as a little bit of an outsider, um, uh, 00:41:53.449 --> 00:41:57.600 uh, you know, someone who I didn't really know, like many of the people or whatnot. 00:41:58.319 --> 00:42:01.049 Um, you know, I think a few things that we've. 00:42:01.290 --> 00:42:09.150 We've also tried to do, um, One, um, We've tried to build things that are, 00:42:09.150 --> 00:42:14.670 like, relatively composable and that you can use, um, in sort of a piecemeal way. 00:42:14.730 --> 00:42:17.400 Like, they come together in a holistic, as a holistic tool. 00:42:17.589 --> 00:42:20.210 But, like, in ruff, we have, like, for example, a linter and a formatter. 00:42:20.620 --> 00:42:23.640 And, like, there's a lot, you can use just the linter or just the formatter. 00:42:23.660 --> 00:42:25.980 And there's a lot of people who use ruff as just a linter or a 00:42:25.980 --> 00:42:29.880 formatter and then they, um, they use some, they use black, right? 00:42:29.930 --> 00:42:33.100 Or they use a different, or they use pylint alongside ruff. 00:42:33.410 --> 00:42:40.940 And so part of it has been like trying to make clear and also be part of an ecosystem where like, 00:42:40.950 --> 00:42:43.509 we're not necessarily trying to like cannibalize those tools. 00:42:43.509 --> 00:42:51.370 Like we spend a lot of time interacting with the people who work on black and we talk a lot about preview style and 00:42:51.370 --> 00:42:54.650 like what style changes should we make and like sometimes we report bugs. 00:42:54.660 --> 00:42:56.310 Sometimes we fix bugs right in black. 00:42:56.340 --> 00:43:02.869 Um, Similar for pylint like a lot of people use pylint with ruff and so we have a good relationship. 00:43:02.889 --> 00:43:06.749 You know we sponsor someone who works on pylint and like we have a good relationship with the tool and like 00:43:07.039 --> 00:43:10.720 That's helpful for both tools because we can build a better experience. 00:43:11.850 --> 00:43:16.935 So, you know part of it is like Uh, and again, it's similar with UV and like with Pip, like we've talked with 00:43:16.935 --> 00:43:20.425 the Pip people a lot and like the projects have pretty different goals. 00:43:20.425 --> 00:43:24.285 And I think it's great that there could be multiple viable installers in Python 00:43:24.285 --> 00:43:27.385 and that the standards and the specs evolve enough to make that possible. 00:43:27.855 --> 00:43:34.915 Um, and so, uh, you know, I think, I think part of it is like, I don't really view it necessarily as zero sum. 00:43:35.400 --> 00:43:40.790 And, um, I, I see us more as like part of an ecosystem of 00:43:40.790 --> 00:43:43.730 tools and, you know, obviously I want us to build the best thing. 00:43:43.740 --> 00:43:48.029 Like, otherwise, I don't know why work on something if you don't want it to be great. 00:43:48.180 --> 00:43:52.089 But like, but, um, you know, I also recognize that like all those 00:43:52.089 --> 00:43:55.180 tools are used by like, also used by like millions of projects. 00:43:55.240 --> 00:43:57.990 And, um, a lot of people who use our stuff also use that stuff. 00:43:57.990 --> 00:43:58.230 Right. 00:43:58.230 --> 00:44:00.440 And like, I want to make Python as a whole better. 00:44:00.810 --> 00:44:04.784 So I don't know, that's some of how I think about this, but, um, yeah. 00:44:06.154 --> 00:44:13.234 You know, I think I think part of it too is just like Be like just genuinely just like being a nice person. 00:44:13.395 --> 00:44:13.855 I guess. 00:44:13.875 --> 00:44:17.245 I don't know like I Think like I just I'd 00:44:17.274 --> 00:44:18.144 Tim: Right, right. 00:44:18.165 --> 00:44:21.914 Come in there and say, "Hey, you need to use this. 00:44:22.215 --> 00:44:22.945 You're all wrong." 00:44:23.165 --> 00:44:28.745 Charlie: Yeah, no, I I just try and like, you know, I hope we build great things and I try to just let them like 00:44:28.745 --> 00:44:36.835 speak for themselves and obviously I add try and advertise our work in various places, but I Try pretty hard not to 00:44:36.835 --> 00:44:41.984 make it about Highlighting flaws and other tools and try to make it a lot more about highlighting the things that 00:44:41.984 --> 00:44:47.319 we do And, um, overall I've been, I've been really happy with that and I'm happy 00:44:47.319 --> 00:44:50.020 with the collaboration that we're able to have with people in the ecosystem. 00:44:50.190 --> 00:44:50.860 Tim: Fantastic. 00:44:51.170 --> 00:44:57.299 Do you want to, I noticed that, oh, sorry, uh, uh, I allowed us to slip away from one of the questions. 00:44:57.609 --> 00:44:58.910 Charlie: Can we go back to the VC? 00:44:58.940 --> 00:45:00.640 Can we go back to talking about VC Venture? 00:45:00.709 --> 00:45:00.999 Yeah. 00:45:01.009 --> 00:45:01.470 Yeah. 00:45:01.600 --> 00:45:01.750 Okay. 00:45:01.750 --> 00:45:02.210 Tim: Yeah. 00:45:02.319 --> 00:45:02.799 Yeah. 00:45:03.130 --> 00:45:03.700 That's right. 00:45:03.720 --> 00:45:05.700 And it's like, Yeah. 00:45:05.700 --> 00:45:10.230 I'm, I don't want to push on this, but it's something that I'm kind of curious about. 00:45:10.370 --> 00:45:10.980 Charlie: I was going to bring it. 00:45:11.029 --> 00:45:11.990 I was going to bring it back up. 00:45:12.029 --> 00:45:12.190 Yeah. 00:45:12.190 --> 00:45:12.999 But I wasn't sure when. 00:45:14.339 --> 00:45:14.639 Tim: Yeah. 00:45:14.680 --> 00:45:14.950 Yeah. 00:45:15.040 --> 00:45:15.629 No, no, no. 00:45:15.669 --> 00:45:17.680 Now let's, um, let's, let's, let's go there. 00:45:17.879 --> 00:45:18.269 Charlie: Yeah. 00:45:18.290 --> 00:45:25.629 So I think like part of it, you know, for me early on when I decided to like start a company and raise 00:45:25.629 --> 00:45:30.584 money around it, um, I just knew, I guess there were a few, like, Yeah. 00:45:32.605 --> 00:45:35.605 You know, I didn't necessarily have like a huge master plan written out in my head. 00:45:35.655 --> 00:45:42.205 Um, the things that I knew were like, If we raise money, I would know how to use it effectively. 00:45:42.955 --> 00:45:47.124 Um, as in, there was a lot, there was just like a huge amount of stuff for us to do. 00:45:47.425 --> 00:45:52.085 And the early adoption of the tooling had demonstrated to me that people really wanted it. 00:45:52.415 --> 00:45:57.115 Like, we were like this, You know, we had projects like SciPy, right? 00:45:57.115 --> 00:46:02.475 And like, like really like long tenured projects were adopting a tool that was very much like alpha. 00:46:02.715 --> 00:46:06.514 And so I was like, there is a, there is a strong desire for this kind of tooling. 00:46:06.514 --> 00:46:12.265 Like I know that we can, um, build great, hopefully great stuff that people will really want to use. 00:46:12.265 --> 00:46:19.425 Um, and you know, I knew that if I was going to keep working on it full time, or sorry, if I was going to keep working 00:46:19.425 --> 00:46:22.095 on this tool and achieve the goals of the tool, it had to be full time. 00:46:22.095 --> 00:46:25.104 Um, Like, it was just too much work for it not to be. 00:46:25.105 --> 00:46:35.144 Um, and, uh, you know, and further, I had this idea in my head for, like, things we could do if we built this tooling, right? 00:46:35.144 --> 00:46:39.765 And this is where it comes back to, like, building a business, is like, if we build this, you know, this 00:46:39.765 --> 00:46:43.584 great tool chain, like, and what kind of positioning does that give us to 00:46:43.595 --> 00:46:46.584 build and sell commercial services that integrate with it really well? 00:46:47.065 --> 00:46:53.365 Um, and, you know, when you start a, a company, especially a venture, again, you don't have to necessarily have 00:46:53.365 --> 00:46:57.555 everything figured out, but you do have to have a vision for what you're trying to, what you're trying to achieve. 00:46:57.555 --> 00:47:01.505 And for us, it was like, uh, and is, we want to build, um, 00:47:01.555 --> 00:47:04.865 you know, this high performance unified Python tool chain in Rust. 00:47:05.210 --> 00:47:09.760 And then we want to use that, um, uh, that to sell you kind of like 00:47:09.760 --> 00:47:11.970 the natural next thing you need when you're working with Python. 00:47:11.970 --> 00:47:13.919 So it could be all sorts of different things, right? 00:47:13.920 --> 00:47:18.630 Could be like package hosting and registries could be CI CD integrations could be deployment and hosting. 00:47:18.630 --> 00:47:23.469 Like I just saw all these opportunities where if we built this really amazing tooling layer, then for people who 00:47:23.469 --> 00:47:27.760 are already using our tooling, there's a lot of problems we could solve by integrating with that tooling and 00:47:27.760 --> 00:47:33.029 integrating across them, like, okay, we're your linter and your package manager and all these different things. 00:47:33.639 --> 00:47:39.040 Um, You know, I think for me, like the way I view it is, um, we, 00:47:41.710 --> 00:47:44.990 like we want to build all this great tooling and that requires investment. 00:47:45.010 --> 00:47:47.280 And so we need to build a sustainable business around it. 00:47:48.000 --> 00:47:52.250 I think a couple, one interesting thing about this company, which is a little bit different than a lot of other open 00:47:52.250 --> 00:47:58.940 source companies for better or for worse is like the traditional business model for venture-backed companies in open source 00:47:58.980 --> 00:48:05.779 is you often build, uh, you have some sort of thing that needs to be hosted and then you charge people, you then you host 00:48:05.780 --> 00:48:11.060 it and people pay you money for that and then you sell other enterprise services on other enterprise features on top of it. 00:48:11.480 --> 00:48:13.190 We actually like don't fit that model. 00:48:13.540 --> 00:48:20.620 Um, uh, it's not like, I mean, I think you could imagine like what a hosted ruff is like, I could probably like, 00:48:21.030 --> 00:48:26.650 you know, I could sketch something out, but it doesn't, it's not really like what people are doing with the tooling. 00:48:26.929 --> 00:48:29.359 Um, you know, it's not like we have a piece of database software 00:48:29.359 --> 00:48:31.720 and then we're going to host the database for you and sell the hosting. 00:48:32.180 --> 00:48:36.230 So, um, I actually think, I mean, it's good and it's bad. 00:48:36.460 --> 00:48:39.250 It's bad because there are fewer companies that I could point to. 00:48:39.280 --> 00:48:41.710 I can point to some, but there are fewer companies I can point to that follow. 00:48:42.320 --> 00:48:43.550 What we were trying to do. 00:48:44.210 --> 00:48:49.120 But I think it's, I think it's good because it means that at least from my perspective, I think we're really 00:48:49.130 --> 00:48:52.430 incentivized to like build out the open source in a very holistic way. 00:48:52.669 --> 00:49:00.240 Um, and, uh, you know, we're not like, we don't suffer from a lot of the same licensing issues, I guess, 00:49:00.250 --> 00:49:04.130 that people follow in that model where you're a little bit disincentivized. 00:49:04.750 --> 00:49:08.170 I guess I don't want to like talk bad about any of those companies, but it's like, sometimes you're kind of distant. 00:49:08.430 --> 00:49:09.930 The incentive alignment is not perfect. 00:49:10.310 --> 00:49:12.400 I think it's the alignment of incentives and that's why they're very imperfect. 00:49:13.720 --> 00:49:17.770 And for us, I want to have the incentive structure such that we build the open source. 00:49:17.810 --> 00:49:19.750 Our goal is for it to grow as much as possible. 00:49:19.970 --> 00:49:24.380 Um, you know, ideally, you know, if I, put on my CEO hat, ideally it 00:49:24.380 --> 00:49:28.320 becomes, um, you know, a funnel for you, for paying customers, right? 00:49:28.320 --> 00:49:29.810 So we know all these people are using our stuff. 00:49:30.120 --> 00:49:32.540 We build and sell services that integrate with it really well. 00:49:32.590 --> 00:49:35.190 And then our incentive structure is continue to build out the open source. 00:49:35.199 --> 00:49:36.590 Ideally make it really popular. 00:49:36.779 --> 00:49:38.690 Ideally make it something that companies can use. 00:49:38.899 --> 00:49:44.650 Um, and then use that as a mechanism to attract customers and people who will pay for our stuff. 00:49:45.010 --> 00:49:47.670 Um, so that's, that's the incentive. 00:49:47.710 --> 00:49:51.969 I care a lot about like the incentive structure and like that is the incentive structure I want to have. 00:49:52.865 --> 00:49:59.875 I do think, um, I think there's like some, you know, some like challenges around, 00:49:59.985 --> 00:50:04.515 um, they're not really that different from other open source projects, I guess. 00:50:04.515 --> 00:50:08.525 But like, you know, we have a team of nine people who are working full time on this stuff. 00:50:08.655 --> 00:50:16.950 Um, and we also want it to be a project that has, um, Ideally, I would 00:50:16.950 --> 00:50:19.590 like to get our products to a point such that they outlive the company. 00:50:19.750 --> 00:50:22.210 Like that is not true today, for sure. 00:50:22.600 --> 00:50:26.040 Um, like we don't have, we have no one on who... 00:50:26.070 --> 00:50:28.539 We have no one with commit access to the project that doesn't work 00:50:28.540 --> 00:50:31.970 at the company, for example, and like a healthy, sustainable... 00:50:32.469 --> 00:50:34.420 Tim: May I raise a hypothetical? 00:50:34.420 --> 00:50:42.860 Which is, let's say, the Python Software Foundation says, look, ruff is now so 00:50:43.100 --> 00:50:48.850 important that we're going to, to ... We'd like you to assign copyright to us. 00:50:49.430 --> 00:50:53.380 Uh, and essentially the funding model is all, is kind of the same, but 00:50:53.380 --> 00:51:00.020 you are now the primary, ... We're the owners of the software. 00:51:00.710 --> 00:51:07.999 We have in some sense control, but you, your team now has the most knowledge of it. 00:51:08.030 --> 00:51:09.139 Does that make any sense? 00:51:09.660 --> 00:51:16.400 Charlie: Um, like do I understand the situation or is this, or do I think, do I agree that that would be a good outcome? 00:51:17.360 --> 00:51:19.619 Tim: No, no, no, no, no, no, no. 00:51:19.619 --> 00:51:23.660 Like, no, I don't even know if there is a question there, but it'd be 00:51:23.680 --> 00:51:29.440 interesting to hear how the software project could outlive the company. 00:51:29.760 --> 00:51:37.145 Presumably that would mean giving it to some other shepherd to kind of, you know, Or something else to kind of maintain. 00:51:37.485 --> 00:51:37.905 Charlie: Yeah. 00:51:37.945 --> 00:51:38.395 Yeah. 00:51:38.535 --> 00:51:42.755 I think there are like some other projects that have done a really good job of this that I look to as role models. 00:51:42.755 --> 00:51:51.704 Like one is, um, confusingly similarly named, but Astro A S T R O in the web and the JavaSc[ript], in the web 00:51:51.704 --> 00:51:58.045 ecosystem, let's say, um, like they just have a really good governance structure in my opinion, at least. 00:51:58.115 --> 00:52:05.405 Um, and you know, for example, Uh, you know, there, there are roles, there's like a sort of a hierarchy of roles, 00:52:05.545 --> 00:52:11.255 and when you join the Astro Technology Company, you're just like slotted in somewhere in that hierarchy, and like 00:52:11.275 --> 00:52:16.114 people who don't work at the company could be above you, or you know, or below you, based on the contributions that they've 00:52:16.114 --> 00:52:21.135 made in open source, um, and there are like sort of clear guidelines for how you move up or down in that hierarchy, 00:52:21.495 --> 00:52:25.585 um, and, and, you know, and they have like, they have an OpenCollective. 00:52:25.755 --> 00:52:30.345 So like they have like, they have corporate sponsors and other people can donate to the project. 00:52:30.565 --> 00:52:35.744 And then those funds are used to support, oh, they, those funds cannot go to employees of the Astro company. 00:52:35.745 --> 00:52:41.664 They're used to support open source contributors, um, uh, you know, events for the project, et cetera, et cetera. 00:52:41.665 --> 00:52:44.715 So they just, in my opinion, at least like they've thought about 00:52:44.715 --> 00:52:47.815 it a lot and put a lot, it's very hard to do that effectively. 00:52:47.825 --> 00:52:50.655 It takes like a ton of work, especially when you're doing like a million other things. 00:52:50.970 --> 00:52:55.540 Um, but, so I have a lot of respect for it and like, that's kind of what I want to build to over time. 00:52:55.970 --> 00:52:59.480 Um, but, but we're certainly, we're definitely not there yet. 00:53:00.340 --> 00:53:08.290 Tim: Now I want to respect your time, but also I want to see if I can suck out some of, uh, your knowledge. 00:53:08.340 --> 00:53:19.350 And there's one sort of final kind of question that I've got, which is, Was there a part of developing Ruff that you 00:53:20.325 --> 00:53:25.515 really relished in from like a technology or data structure kind of problem. 00:53:26.055 --> 00:53:30.725 Um, was it, you know, if you were to like think, well, actually that was, you know, when you started out, 00:53:30.735 --> 00:53:34.224 it was really fun and interesting kind of intellectually engaging. 00:53:35.294 --> 00:53:38.125 Uh, what was that like? 00:53:38.204 --> 00:53:40.095 And, and, and could you explain that? 00:53:40.635 --> 00:53:41.395 Charlie: Yeah, totally. 00:53:41.395 --> 00:53:46.365 I mean, I think like for me, I'd never really built anything like a linter before. 00:53:46.495 --> 00:53:49.705 Um, and so. 00:53:51.130 --> 00:53:57.260 Like a lot of the initial infrastructure around like semantic analysis was really interesting. 00:53:57.270 --> 00:54:02.500 So, um, you know, when, when I first shipped and the first ruff release 00:54:02.730 --> 00:54:06.399 supported a pretty small number of rules, but I wanted to make sure... 00:54:06.410 --> 00:54:07.900 It was really a proof of concept. 00:54:08.230 --> 00:54:11.240 But I wanted to make sure that it could support some fairly sophisticated rules. 00:54:11.250 --> 00:54:13.680 So, like, unused import detection, for example. 00:54:14.030 --> 00:54:21.620 That requires like, um, a fair amount of semantic analysis, like, you need to be able to identify, basically, you need 00:54:21.620 --> 00:54:25.890 to be able to identify which imports are used and which are not, so you need to be able to resolve references, you need 00:54:25.900 --> 00:54:29.460 to be able to deal with, like, shadowing, like someone imports something and then 00:54:29.890 --> 00:54:32.839 assigns to a variable with the same name, and then that variable is used, right? 00:54:32.840 --> 00:54:37.849 I mean, it's like, basic stuff when, from the perspective of the programmer, but from a perspective of program 00:54:37.850 --> 00:54:44.300 analysis, as in the Python, the end Python programmer understands a lot of those rules and understands whether an 00:54:44.300 --> 00:54:47.580 import's used or not, but like, from the perspective of program analysis, 00:54:47.590 --> 00:54:50.170 it just required a lot of things I had never really seen or done before. 00:54:50.480 --> 00:54:57.240 Um, uh, a lot of that stuff I learned by reading other codebases. 00:54:57.860 --> 00:55:04.850 Like, open source is pretty amazing because you can just go look at how people solve certain problems. 00:55:05.079 --> 00:55:11.610 Um, and, and so like the PyFlakes codebase, which was the, uh, um, uh, 00:55:11.620 --> 00:55:15.180 where we, a lot of the rules that they have are rules that we then implemented. 00:55:15.400 --> 00:55:18.990 Like, I read that really thoroughly in trying to understand, like, how do they deal with bindings? 00:55:19.020 --> 00:55:20.929 What are the different representations that they used? 00:55:20.960 --> 00:55:26.000 And, like, that model has evolved substantially over time, especially as I've brought people into the team who 00:55:26.000 --> 00:55:31.050 are, like, have a lot more experience than me at those kinds of problems, and, like, we've evolved that semantic model. 00:55:31.370 --> 00:55:36.190 Um, but that was a really cool thing, like, being able to do program analysis in that way. 00:55:36.560 --> 00:55:42.205 Um, I also, maybe another thing that was really rewarding for me in a slightly different way. 00:55:42.775 --> 00:55:43.305 Um. 00:55:44.075 --> 00:55:44.745 We... 00:55:45.475 --> 00:55:54.124 when we started working on the formatter, um, it's a very interesting set of problems, um, around code formatting 00:55:54.214 --> 00:56:01.840 and, um, uh, one of the first two people that I hired, uh, his name is Mika, And he worked on, um, he'd worked on 00:56:01.840 --> 00:56:07.739 a Rust-based formatter for JavaScript before and he brought in a lot of knowledge around, kind of like, the right 00:56:07.739 --> 00:56:12.040 way to do that, um, and adapting it to Python and some of the quirks of Python. 00:56:12.300 --> 00:56:19.160 And so, like, working on that and seeing, um, some of the things he did for performance, some of the things that were 00:56:19.160 --> 00:56:23.895 just, like, hard problems to solve, like comment association, is like so hard. 00:56:24.135 --> 00:56:26.935 Um, like how do you, like, yeah. 00:56:26.935 --> 00:56:32.655 I would say like 75 percent of the challenge of the formatter is comments. 00:56:33.425 --> 00:56:37.455 Like the basic code formatting is not, I mean, it is hard. 00:56:38.444 --> 00:56:43.155 Um, but like comments are like so much of the complexity. 00:56:43.730 --> 00:56:46.260 Because, anyway, it's not necessarily interesting, but 00:56:46.260 --> 00:56:46.930 Tim: Where should you put them? 00:56:46.970 --> 00:56:47.710 Charlie: Yeah, where do you put them? 00:56:47.910 --> 00:56:49.540 Um, and they can go basically anywhere. 00:56:49.740 --> 00:56:53.780 Um, and so the chain formatting, you can put comments between basically any two tokens. 00:56:53.849 --> 00:56:56.309 So like, not quite, but like almost. 00:56:56.340 --> 00:57:00.299 Um, so it just creates like a real explosion of cases that you need to handle. 00:57:00.320 --> 00:57:00.794 Um, yeah. 00:57:01.255 --> 00:57:03.165 But the basic rules of like, how do you figure out? 00:57:03.205 --> 00:57:04.145 Is it a leading comment? 00:57:04.155 --> 00:57:05.255 Is it a trailing comment? 00:57:05.265 --> 00:57:07.035 What node should it be associated with? 00:57:07.035 --> 00:57:12.224 Is it part of the, you know, if you have a comment on the first argument of a function definition, is it 00:57:12.224 --> 00:57:15.595 part of like the function definition or is it part of the argument? 00:57:15.634 --> 00:57:20.615 Um, you know, and there's just a lot of, like, understanding the rules around that was very, it was really interesting. 00:57:21.875 --> 00:57:23.995 It became painful over time, but it was very interesting. 00:57:24.005 --> 00:57:24.415 Tim: Extremely. 00:57:24.875 --> 00:57:28.105 You snuck in something else in there, which I thought was really, really nice, which. 00:57:28.630 --> 00:57:34.350 Um, you know, essentially once you decided to build a compiler company, 00:57:34.810 --> 00:57:43.310 you brought in experts and then you were able to benefit from others' expertise. 00:57:43.340 --> 00:57:50.910 And I think one of the hard, or like the big challenge when you move from solo to a small team, and then as your 00:57:50.910 --> 00:57:56.185 team grows, is that you need, you start hiring people that are better than you. 00:57:56.965 --> 00:57:57.105 Charlie: Yeah. 00:57:57.105 --> 00:57:57.715 That's good though. 00:57:59.575 --> 00:57:59.895 Yeah. 00:58:00.395 --> 00:58:03.215 Tim: And, uh, 100 percent it's good. 00:58:03.275 --> 00:58:08.614 And it just kind of reinforces that idea of trust, which, um, which is extreme. 00:58:08.614 --> 00:58:08.984 Charlie: Yeah. 00:58:08.985 --> 00:58:18.655 I mean, well, maybe just last thing I'll say really quick is like the biggest, like starting a company, like it's, um, 00:58:20.185 --> 00:58:23.825 Sorry, the bigger evolution, the bigger change in my life was not like starting 00:58:23.825 --> 00:58:27.255 a company, it was actually like hiring, hiring employees and people to work there. 00:58:27.275 --> 00:58:32.725 Because like, when I started a company, like, yes, I'm putting some things on the line, but it's just me. 00:58:33.375 --> 00:58:37.690 And then when other people join, like, Yeah, it just becomes, um, you 00:58:37.690 --> 00:58:41.000 know, suddenly the responsibility like goes far beyond yourself. 00:58:41.380 --> 00:58:46.480 Um, and so that to me was the biggest change was like bringing on other 00:58:46.480 --> 00:58:50.650 people on the team, and, you know now having a responsibility to them as well. 00:58:50.989 --> 00:58:56.440 Um, so anyway, yeah, uh, it's it's it is an amazing vote of confidence to, like... 00:58:57.860 --> 00:58:59.960 Have people come join you on what you're doing. 00:59:00.230 --> 00:59:06.790 Um, and I think we've built, um, a team that's like so much better than I ever could have imagined. 00:59:06.800 --> 00:59:12.080 Like we've just tracked, we've just been able to, um, hire some really incredible people, but it's also, 00:59:12.310 --> 00:59:16.860 yeah, for anyone out there who's considering starting a company, it is also a lot of responsibility, right? 00:59:16.860 --> 00:59:21.989 It's other people's entire lives, um, or sorry, or their livelihoods, I guess would be the right word. 00:59:22.330 --> 00:59:29.249 Um, but yeah, anyway, so maybe, uh, I don't know, a somber note to end on or an energizing note, right? 00:59:29.419 --> 00:59:29.869 I don't know. 00:59:30.459 --> 00:59:41.350 Tim: Yeah, no, I think, like, it's, it's, it's, it is an energizing, um, note and it It is important to me that 00:59:41.530 --> 00:59:51.130 senior leaders at even small companies are aware of the responsibility that they have for the employees. 00:59:51.170 --> 00:59:55.900 I think that, um, Yeah, it is people's livelihoods. 00:59:56.020 --> 01:00:01.360 And as, uh, as someone also at a very similar, oh, you know, I'm 01:00:01.360 --> 01:00:04.750 actually at a much earlier stage in my, my own business, really. 01:00:04.950 --> 01:00:10.969 Um, I can definitely empathize with how, for me, it's sort of 01:00:11.020 --> 01:00:14.589 tight, slightly terrifying, this idea of do we bring people on? 01:00:15.269 --> 01:00:25.060 Um, because You know, I've got X months of, uh, let's say four to four to six months of runway personally 01:00:25.060 --> 01:00:28.440 that I've been able to, you know, and it's like, but what if that runs out? 01:00:28.560 --> 01:00:29.210 What happens then? 01:00:30.050 --> 01:00:30.610 Um, 01:00:30.680 --> 01:00:31.230 Charlie: Yeah. 01:00:31.330 --> 01:00:40.919 Tim: Anyway, I, um, I'm just, uh, so thrilled that you're able to share an hour or so with me today, Charlie. 01:00:41.299 --> 01:00:43.860 I am curious, how should people connect with you? 01:00:43.910 --> 01:00:46.160 Um, follow Ruff, follow the company. 01:00:46.210 --> 01:00:46.410 Charlie: Yeah. 01:00:46.410 --> 01:00:49.060 The best of the best is on Twitter/X. 01:00:49.350 --> 01:00:51.060 Yeah, it's just @charliermarsh. 01:00:51.400 --> 01:00:54.440 Um, and that's where I'm, that's where I'm most active. 01:00:54.590 --> 01:00:59.930 Um, and, uh, but otherwise on GitHub, I mean, to the degree, it was not quite, I don't know if I would 01:00:59.930 --> 01:01:02.869 describe GitHub as a social network, but you can follow me on GitHub. 01:01:04.040 --> 01:01:07.254 Um, uh, but yeah, the best, the best is on X. 01:01:08.055 --> 01:01:08.555 Tim: Superb. 01:01:08.705 --> 01:01:09.475 Okay, fantastic. 01:01:09.665 --> 01:01:10.375 Hey, thanks again. 01:01:10.465 --> 01:01:15.045 And, uh, I, uh, I'm going to be watching you very intently online. 01:01:15.675 --> 01:01:19.525 Charlie: Yeah, thank you so much for, um, uh, for having me on, for reaching out, for setting this up. 01:01:19.525 --> 01:01:20.784 It was super fun for me too.