WEBVTT NOTE This file was generated by Descript 00:00:13.212 --> 00:00:15.952 This is Self -Directed Research, where our hosts, 00:00:16.012 --> 00:00:19.192 James and Amos, give each other 20 minute presentations that usually 00:00:19.192 --> 00:00:22.362 last longer than that about whatever they've been obsessing over lately. 00:00:22.782 --> 00:00:27.722 For more episodes, show notes, and transcripts, visit sdr-podcast.com. 00:00:28.122 --> 00:00:30.022 New episodes drop every Wednesday. 00:00:30.622 --> 00:00:32.862 Today's episode is sponsored by fasterthanlime. 00:00:33.032 --> 00:00:35.082 More info at the end of the episode. 00:00:35.612 --> 00:00:39.502 This week, Amos presents, "How Usable is Crane Lift Today?" 00:00:44.585 --> 00:00:45.235 Okay, let's go. 00:00:45.335 --> 00:00:49.795 So, this week I want to talk about: I want everyone to congratulate 00:00:49.795 --> 00:00:51.175 me on being so generous. 00:00:51.515 --> 00:00:52.135 Not really. 00:00:52.425 --> 00:00:52.725 Okay. 00:00:52.725 --> 00:00:53.615 So here's what happened. 00:00:53.675 --> 00:00:59.036 The Rust compiler: it's a machine that turns your code into a metric ton 00:00:59.526 --> 00:01:01.906 of LLVM intermediate representation. 00:01:02.026 --> 00:01:06.646 And then LLVM turns that into native code, and then your linker just sleeps in a 00:01:06.646 --> 00:01:08.676 loop, probably, because it's that slow. 00:01:08.775 --> 00:01:09.945 Puts a big bow around it. 00:01:09.945 --> 00:01:13.795 Magically, yeah, you have an executable file appear on your disk 00:01:13.875 --> 00:01:18.185 and it's between, let's say two and 700 megabytes, depending on where you work and 00:01:18.185 --> 00:01:19.845 how many debug information you pull in. 00:01:20.185 --> 00:01:21.715 That's basically what the Rust compiler does. 00:01:21.745 --> 00:01:22.045 Actually. 00:01:22.055 --> 00:01:25.015 Well, no, no- that's the compiler side gig. 00:01:25.445 --> 00:01:27.265 I think was the talk Esteban gave. 00:01:27.485 --> 00:01:30.111 So the main thing the Rust compiler does is parse and emit diagnostics. 00:01:30.111 --> 00:01:32.601 And then as a side gig, it generates network. 00:01:33.024 --> 00:01:39.673 But, problems with LLVM: it's a big pile of C++, it's a research project 00:01:39.683 --> 00:01:42.733 that has been productionized and then remained a research project. 00:01:42.743 --> 00:01:45.085 So there's a lot of things happening. 00:01:45.415 --> 00:01:48.645 Like, I think they turned their people optimization register 00:01:48.825 --> 00:01:52.575 allocator from code to like a model. 00:01:52.615 --> 00:01:53.815 Like let's just train a model. 00:01:53.825 --> 00:01:56.155 Cause it's going to be better than whatever heuristics we can come up with. 00:01:56.165 --> 00:01:57.285 So there's things like that. 00:01:57.285 --> 00:02:00.700 There's like really strange vectorization things happening. 00:02:01.240 --> 00:02:02.810 LLVM is still research ground and- 00:02:02.853 --> 00:02:04.163 So these are all optimizations. 00:02:04.173 --> 00:02:07.963 Like we have some code that comes in and LLVM goes, "I see what you're doing. 00:02:07.963 --> 00:02:09.163 There's a better way to do that." 00:02:09.183 --> 00:02:12.263 And without actually knowing the source code form or with limited 00:02:12.263 --> 00:02:14.763 access to it just goes, "No, no, no. 00:02:14.763 --> 00:02:15.623 It's better like this." 00:02:15.653 --> 00:02:16.188 Yeah, yeah, yeah. 00:02:16.188 --> 00:02:19.248 And the way it actually works is like it transforms the input into 00:02:19.248 --> 00:02:21.188 different intermediate representations. 00:02:21.368 --> 00:02:25.118 And then using those forms, it's easier to analyze like what you can optimize 00:02:25.348 --> 00:02:28.644 because you, you realize: Oh, those two things come from the same computation. 00:02:28.831 --> 00:02:29.991 So we can just merge them or whatever. 00:02:30.001 --> 00:02:31.166 There's a lot of that going around. 00:02:31.186 --> 00:02:35.196 So LLVM is a blessing and a curse because the point that they made the 00:02:35.196 --> 00:02:37.926 choice to use it- let's say I'm not familiar with the entire history- 00:02:38.161 --> 00:02:39.861 it was the good obvious choice. 00:02:40.181 --> 00:02:43.949 And since then, another project has happened, for WebAssembly reasons 00:02:44.279 --> 00:02:49.179 called Cranelift, but it does happen to generate AMD64 code or x86-64 or 00:02:49.179 --> 00:02:52.109 whatever you want to call it, and a bunch of other architectures now as well. 00:02:52.449 --> 00:02:54.509 I don't exactly know who had the idea at first, but they were like, 00:02:54.509 --> 00:02:59.127 "Well, Cranelift, the WebAssembly based projects, the backend is Rust. 00:02:59.337 --> 00:03:00.427 The Rust compiler is Rust. 00:03:00.727 --> 00:03:01.827 Let's arrange something." 00:03:01.910 --> 00:03:05.900 I don't have any of the dates or the timeline off the top of my head, but 00:03:06.274 --> 00:03:12.094 it went as far as like a few months ago on the official Rust blog, they talked 00:03:12.094 --> 00:03:17.564 about using that backend, the Cranelift CodeGen backend for debug builds. 00:03:17.944 --> 00:03:22.324 And it is now distributed on a rustup for at least 64 bit Linux. 00:03:22.808 --> 00:03:26.878 And I was very interested in it, but as it happens, I switched entirely 00:03:26.878 --> 00:03:29.238 over to Macs for a very simple reason. 00:03:29.308 --> 00:03:33.636 It's that macOS is the only operating system that works on desktop in 2024. 00:03:33.636 --> 00:03:36.568 Because you have Linux, which is, you know, the year Linux 00:03:36.568 --> 00:03:38.377 on the desktop is any year now. 00:03:38.667 --> 00:03:40.507 And then you have Windows, which is being actively made 00:03:40.517 --> 00:03:42.267 worse with every passing month. 00:03:42.597 --> 00:03:44.327 I defended it, until I couldn't. 00:03:44.627 --> 00:03:44.957 I don't know. 00:03:44.957 --> 00:03:46.047 I just have so many workflows. 00:03:46.047 --> 00:03:48.067 I was like, "Well, this is great, but it only works on Mac." 00:03:48.077 --> 00:03:50.067 And I was like, "What if I only had Macs? 00:03:50.427 --> 00:03:51.287 Then everything is great." 00:03:51.337 --> 00:03:53.247 Essentially, except when AirDrop doesn't work. 00:03:53.774 --> 00:03:57.904 For me, it was just having a reasonable ARM architecture processor, 00:03:57.904 --> 00:04:01.954 because like the M-series Macs versus any other laptop- like on a desktop, 00:04:01.954 --> 00:04:06.264 it's more debatable, I feel like, but at least in terms of like efficiency 00:04:06.264 --> 00:04:10.259 and performance for a mobile device, like a laptop, which I use very often. 00:04:10.549 --> 00:04:14.889 I was looking at like Framework or M-series Macs and it was like: okay, 00:04:14.909 --> 00:04:17.009 it pretty much has to be M-series Mac. 00:04:17.146 --> 00:04:19.436 That is definitely what pulled me in on the laptop level. 00:04:19.476 --> 00:04:22.996 But then I had like the nice laptop and then I had to go back to my Windows 00:04:23.036 --> 00:04:25.836 desktop and just interacting with Windows- just even transferring files. 00:04:25.836 --> 00:04:27.669 I'm like- I'm not going to email files for myself. 00:04:27.669 --> 00:04:29.339 I'm not going to use USB sticks. 00:04:29.339 --> 00:04:30.569 Like I don't even have USB ports. 00:04:30.569 --> 00:04:31.289 I have USB C. 00:04:31.289 --> 00:04:31.869 What do I do? 00:04:32.369 --> 00:04:32.779 I don't know. 00:04:33.319 --> 00:04:36.156 And, I was always interested in the Cranelift codegen backend. 00:04:36.391 --> 00:04:41.099 But it didn't work on MacOS- aarch64, whatever the actual 00:04:41.099 --> 00:04:42.369 name of the architecture is. 00:04:43.819 --> 00:04:47.098 And recently it started working and I was like - it's time to check it out. 00:04:47.278 --> 00:04:49.258 And I don't know if you know, but I write articles. 00:04:49.298 --> 00:04:54.578 I'm a bit of an author myself, you may say, and my website is powered by a 00:04:54.578 --> 00:04:57.148 gigantic bunch of Rust code as well. 00:04:57.498 --> 00:05:00.838 I've sat down and built a static site generator where you go, "Aha, 00:05:00.848 --> 00:05:02.578 I have written code that generates HTML." 00:05:02.588 --> 00:05:06.696 You went like the one step further of like, "I have a fully dynamic running 00:05:06.696 --> 00:05:10.696 web server backend that powers my entire site that I've written while 00:05:10.696 --> 00:05:16.116 writing articles about writing a website dynamic backend in Rust." 00:05:16.131 --> 00:05:18.891 Yes, and I want you to know that every single 00:05:18.891 --> 00:05:20.901 decision and isolation made sense. 00:05:21.191 --> 00:05:24.571 I still don't know how I ended up here with like 20,000 lines of code 00:05:24.601 --> 00:05:27.210 to maintain and operate on Kubernetes. 00:05:27.240 --> 00:05:28.540 But somehow we're here. 00:05:28.940 --> 00:05:31.440 And the reason I was talking about built performance and whatnot in the 00:05:31.440 --> 00:05:35.450 previous episode was because I'm trying to get it back into a shape where I 00:05:35.450 --> 00:05:38.890 can actually iterate and add features rather than just make sure it builds 00:05:38.910 --> 00:05:40.221 with the latest version of crates. 00:05:40.593 --> 00:05:45.336 So: that mountain of code ends up having about 700 transitive 00:05:45.346 --> 00:05:48.136 dependencies, which is not all my fault. 00:05:48.196 --> 00:05:48.626 I don't know. 00:05:48.626 --> 00:05:49.886 I pulled the S3 crates. 00:05:49.886 --> 00:05:51.366 That's a hundred dependencies right there. 00:05:51.366 --> 00:05:54.416 Probably the aws-sdk-s3 crate. 00:05:54.799 --> 00:05:55.709 I'm doing what I can, okay? 00:05:55.709 --> 00:05:59.719 I'm doing video encoding, I'm doing image encoding, I'm doing a bunch of templating. 00:05:59.729 --> 00:06:00.759 It's complicated! 00:06:00.759 --> 00:06:01.639 It's not all my fault. 00:06:01.639 --> 00:06:05.499 I'm doing compression, broadly, gzip, whatever, Patreon API, all the APIs. 00:06:05.549 --> 00:06:06.989 There's so much functionality in there. 00:06:07.021 --> 00:06:08.021 And 00:06:08.140 --> 00:06:10.860 I tried the Cranelift codegen backend, fully expecting 00:06:10.860 --> 00:06:15.066 it to vomit an error and then just immediately stop after a couple of crates. 00:06:15.596 --> 00:06:20.066 But to my surprise, only like a couple crates didn't build out of the 700. 00:06:20.346 --> 00:06:26.170 So there's ring because it has asserts that are supposed to run at compile time. 00:06:26.180 --> 00:06:28.520 And I'm not even exactly sure how that works. 00:06:28.860 --> 00:06:30.515 But it was very, very unhappy about that. 00:06:30.515 --> 00:06:34.095 But the nice thing about asserts is that like, let's just make sure this actually 00:06:34.095 --> 00:06:39.185 looks like 64-bit ARM on MacOS and you can just comment them out because yes, it is. 00:06:39.195 --> 00:06:41.225 It's regular MacOS, trust me, it's going to be fine. 00:06:41.705 --> 00:06:42.765 So that took care of ring. 00:06:43.705 --> 00:06:49.515 And then the other thing was in the half crate, which you may know, 00:06:49.515 --> 00:06:52.415 actually from your embedded work, but it's like it defines float16. 00:06:52.753 --> 00:06:57.042 And so it has assembly, and so we actually moved away from LLVM assembly 00:06:57.042 --> 00:06:59.002 to like Rust assembly syntax... 00:06:59.092 --> 00:07:00.302 I want to say one edition ago? 00:07:00.302 --> 00:07:02.642 I don't know exactly how that change happened, but I know I had to update 00:07:02.642 --> 00:07:04.272 my series that had inline assembly. 00:07:04.822 --> 00:07:07.662 Yeah, so we had ring, we had half, and half wouldn't compile. 00:07:07.662 --> 00:07:11.699 So I just- it turns out that it was pulled in by a crate I didn't actually use. 00:07:12.069 --> 00:07:14.889 So I just dropped that dependency and uh, everything built! 00:07:15.459 --> 00:07:16.149 But didn't link! 00:07:16.469 --> 00:07:20.489 It didn't link because- and we talked about that recently, so it all tied 00:07:20.489 --> 00:07:23.919 in, this is why I was excited to talk about this- it doesn't link 00:07:24.309 --> 00:07:29.081 on Xcode 15, which is the release where they optimized the linker. 00:07:29.081 --> 00:07:31.351 I'm assuming they rewrote a large part of it. 00:07:31.661 --> 00:07:34.369 So you can downgrade to Xcode 14 and then it works. 00:07:34.649 --> 00:07:39.819 Or you can use a patch that someone wrote and it's the most hilarious patch 00:07:39.819 --> 00:07:43.748 because it's like really domain specific linker knowledge, like it's the wrong 00:07:43.748 --> 00:07:45.403 kind of relocation being generated. 00:07:45.543 --> 00:07:46.393 Wait, a patch to what? 00:07:46.403 --> 00:07:47.102 The compiler? 00:07:47.163 --> 00:07:48.998 To the Cranelift codegen backend. 00:07:49.098 --> 00:07:49.268 Yeah. 00:07:49.498 --> 00:07:51.508 It's not a patch, it's not a pull request. 00:07:51.718 --> 00:07:55.348 It's like someone did 'git diff' and then posted the result in a git 00:07:55.368 --> 00:07:56.778 command saying, "Don't use this. 00:07:56.808 --> 00:07:57.568 I don't know what I'm doing. 00:07:57.568 --> 00:07:59.788 Someone who actually knows what they're doing should take care of this." 00:08:00.128 --> 00:08:02.608 But if they don't know what they're doing, who does? 00:08:02.638 --> 00:08:04.478 Cuz I don't know what's in the patch I don't understand it... 00:08:04.782 --> 00:08:06.952 There's this whole fight on Twitter, and I know you 00:08:06.952 --> 00:08:10.952 found this thread too, of like: Linkers are the like intersection 00:08:10.952 --> 00:08:13.082 point of so many integration schemes. 00:08:13.402 --> 00:08:20.042 It's like the integration point- it is the cursed knowledge, stare-into-the-void item 00:08:20.332 --> 00:08:25.592 that is at the intersection of like five other cursed stare-into-the-void items. 00:08:25.702 --> 00:08:28.462 I was not happy with all the responses on that thread because 00:08:28.462 --> 00:08:30.582 some of it are just like, "Oh, it's cursed, you'll never fix it." 00:08:30.782 --> 00:08:34.444 So I think the issue is generally that people like to avoid- 00:08:34.450 --> 00:08:35.960 like, linkers are unpleasant. 00:08:35.980 --> 00:08:39.106 Like, I feel like this is the old way of handling abstraction is you 00:08:39.106 --> 00:08:40.806 pretend that things don't exist. 00:08:41.026 --> 00:08:43.756 Then they're allowed to grow as big and uncomfortable and 00:08:43.756 --> 00:08:45.466 terrible to deal with as they want. 00:08:45.686 --> 00:08:47.806 Because you just go, "Just don't, don't look at it. 00:08:47.806 --> 00:08:49.446 Don't make eye contact, and you'll fine." 00:08:49.446 --> 00:08:53.163 Linkers are like the ideal of that: you just hope that you never 00:08:53.173 --> 00:08:54.737 have to hear the word 'linker.' 00:08:54.767 --> 00:08:57.202 And this ties into two other things I already talked about 00:08:57.222 --> 00:09:00.692 on this podcast, which are: I was wrong about splitting your code base 00:09:00.692 --> 00:09:03.572 into many different crates because it creates interfaces and interfaces 00:09:03.572 --> 00:09:04.892 by definition are boundaries. 00:09:05.122 --> 00:09:06.512 The exact same thing happens. 00:09:06.522 --> 00:09:10.542 You start to ignore what's on the other side of the boundary and you're 00:09:10.552 --> 00:09:13.662 missing out on optimizations or simplifications that could have been done. 00:09:13.932 --> 00:09:16.731 And, let me tell you: ever since I merged everything back into one big 00:09:16.731 --> 00:09:19.577 crate, I'm going, "Oh, this thing is duplicated all over the place. 00:09:19.597 --> 00:09:20.827 I can simplify the whole thing. 00:09:20.837 --> 00:09:23.768 I can build this thing once instead of doing it on demand and 00:09:23.788 --> 00:09:25.348 like reallocating over and over." 00:09:25.548 --> 00:09:28.068 Like, the whole structure of the thing becomes so much clearer 00:09:28.068 --> 00:09:29.318 now that it's one big blob. 00:09:29.648 --> 00:09:32.638 And I'm looking forward to finding the interfaces that do make sense 00:09:32.658 --> 00:09:33.648 and actually split it up again. 00:09:33.948 --> 00:09:35.608 But specifically for linkers, I think you're right. 00:09:35.648 --> 00:09:36.086 I think 00:09:36.086 --> 00:09:36.886 we 00:09:36.894 --> 00:09:40.084 collectively, like people around compilers tend to think 00:09:40.099 --> 00:09:41.239 that it's not our problem, right? 00:09:41.239 --> 00:09:42.559 We just emit the object files. 00:09:42.559 --> 00:09:44.879 Whatever happens after that is none of our concern. 00:09:45.229 --> 00:09:50.009 So back to the Cranelift code generation backend: you can either downgrade to Xcode 00:09:50.009 --> 00:09:52.639 14, which is not a future proof solution. 00:09:52.649 --> 00:09:56.739 It's not what's favored by the maintainers of cg_clif is the short name for the 00:09:57.409 --> 00:09:58.809 Cranelift code generation backend. 00:09:58.859 --> 00:10:01.952 And the patch that is in the comments as an inline diff 00:10:01.975 --> 00:10:04.345 file that I did apply works... 00:10:04.385 --> 00:10:05.405 until it doesn't? 00:10:05.405 --> 00:10:07.769 So it did compile all the crates individually which 00:10:07.989 --> 00:10:09.329 already involves some linking. 00:10:09.622 --> 00:10:09.922 Yeah. 00:10:09.922 --> 00:10:13.922 Cause typically when things get linked, you create one archive or 00:10:13.922 --> 00:10:17.792 library that you link together as an archive or like a unified thing. 00:10:17.802 --> 00:10:21.221 And then, essentially at the end the linker goes, "Merge all 00:10:21.221 --> 00:10:25.231 of those intermediate archives into one big final executable." 00:10:25.961 --> 00:10:28.801 And so there's many different opportunities for linker errors to happen. 00:10:28.876 --> 00:10:32.127 You don't actually have to downgrade to Xcode 14, you can just use a 00:10:32.127 --> 00:10:34.887 linker flag called ld_classic. 00:10:35.457 --> 00:10:39.657 And of course there's like 26 different ways to spell it wrong? 00:10:40.027 --> 00:10:42.307 If you just get it wrong, it's just ignored. 00:10:42.727 --> 00:10:44.997 I love that, I love when programs do that. 00:10:44.997 --> 00:10:47.237 Linker UX is real good. 00:10:47.427 --> 00:10:49.527 Just pass a flag it doesn't know about it's like, "Okay" 00:10:50.204 --> 00:10:51.334 That's compatibility! 00:10:51.384 --> 00:10:55.351 Yeah, it is for sure, which is why when I derive deserialize, 00:10:55.368 --> 00:10:59.111 I always have like 'error on unknown fields.' I forget the exact attribute. 00:10:59.135 --> 00:11:00.844 Oh yeah, serde ignore? 00:11:00.891 --> 00:11:01.561 Not ignore. 00:11:01.723 --> 00:11:02.103 You're right. 00:11:02.103 --> 00:11:03.900 Yeah, yeah, I, know exactly what you're talking about. 00:11:04.010 --> 00:11:06.190 I want to look at the schema change and I want to know about it. 00:11:06.680 --> 00:11:10.090 Well, I could version my configs, but I'm- I'm a, I'm a one man shop. 00:11:10.140 --> 00:11:10.880 I don't need to do that. 00:11:11.520 --> 00:11:12.650 I'd rather have production trash. 00:11:12.815 --> 00:11:16.095 Schema evolution is a conversation for another day and we should 00:11:16.095 --> 00:11:18.055 talk about schema evolution at some point. 00:11:18.080 --> 00:11:19.320 It doesn't fit in 20 minutes. 00:11:19.320 --> 00:11:20.560 We can never talk about it here. 00:11:20.705 --> 00:11:21.135 No 00:11:21.375 --> 00:11:24.310 So, there is a linker flag called ld_classic. 00:11:24.445 --> 00:11:28.815 I was trying to use it, but as you can imagine, building an alternative codegen 00:11:28.835 --> 00:11:32.435 backend for rustc is a bit involved. 00:11:32.445 --> 00:11:33.835 So there's an actual build system. 00:11:33.835 --> 00:11:36.415 So if you're building the Rust compiler, rustc, if you check it out, which I 00:11:36.415 --> 00:11:38.975 encourage everyone to do, because they have a really good contributors guide. 00:11:39.325 --> 00:11:40.425 Everyone can just clone it. 00:11:40.425 --> 00:11:41.575 It's not that big. 00:11:41.576 --> 00:11:44.641 It can download pre built versions of LLVM. 00:11:44.851 --> 00:11:46.161 It doesn't actually take that long. 00:11:46.161 --> 00:11:48.451 I thought it was going to take like a whole day to build for the first time. 00:11:48.451 --> 00:11:49.631 I thought it was like Chromium or whatever. 00:11:49.631 --> 00:11:49.861 No. 00:11:50.171 --> 00:11:52.281 Contributing to rustc is actually not that scary. 00:11:52.661 --> 00:11:57.791 It has this thing called x.py, which is a Python script that ends up 00:11:57.801 --> 00:12:02.021 building a bunch of Rust, which is then the actual build system for rustc. 00:12:02.331 --> 00:12:04.813 So the rustc build system is mostly written in Rust, but it has this 00:12:04.833 --> 00:12:06.653 Python wrapper, so it's cross platform. 00:12:06.883 --> 00:12:09.643 And the Cranelift codegen backend has a similar thing. 00:12:10.053 --> 00:12:14.483 It has y.sh and y.ps1 and whatever for other platforms. 00:12:14.733 --> 00:12:18.373 And it also builds some Rust to also end up invoking the Rust compiler and 00:12:18.373 --> 00:12:21.673 it applies patches to the standard library so that it gets to compile 00:12:21.692 --> 00:12:22.932 with that backend and whatever. 00:12:23.049 --> 00:12:26.884 For some reason throughout this whole process, the linker flag got lost. 00:12:27.014 --> 00:12:31.914 So not only will the linker not complain if you give it a nonsense flag, or Rust 00:12:31.934 --> 00:12:35.384 flags will be ignored if you forget -Wl. 00:12:35.384 --> 00:12:37.954 So it's not a linker flag, it's a normal flag. 00:12:38.344 --> 00:12:42.154 Just a bunch of drivers and frontends and programs just ignore flags if they 00:12:42.154 --> 00:12:43.714 don't know about them, which is lovely. 00:12:44.144 --> 00:12:46.954 Does the phrase linker driver mean anything to you? 00:12:47.114 --> 00:12:48.784 To me, yes, but you should still explain. 00:12:49.509 --> 00:12:51.229 Okay- I was going to say, I'm probably gonna get this wrong, cause 00:12:51.229 --> 00:12:54.249 this is again, cursed knowledge, but like- a lot of the times you don't necessarily 00:12:54.249 --> 00:12:59.602 even invoke the linker as its own binary program, you call the compiler with 00:12:59.602 --> 00:13:04.450 certain arguments because hopefully the compiler knows what it's supposed to be 00:13:04.450 --> 00:13:09.069 doing and then can pass on more of that information when it invokes the linker. 00:13:09.120 --> 00:13:13.630 A lot of those flags that you set are rustc flags, which means you 00:13:13.630 --> 00:13:18.670 basically are giving them to rustc with forwarding instructions to the linker. 00:13:18.860 --> 00:13:21.190 So you're not even talking to the linker itself. 00:13:21.260 --> 00:13:21.620 I don't know. 00:13:22.175 --> 00:13:26.625 I've built some build system type stuff where you do independently do it, but 00:13:26.625 --> 00:13:29.975 it means that you then need to remember everything that you gave to the compiler, 00:13:30.185 --> 00:13:34.995 and when Rust has, like, cargo followed by rustc, the frontend, followed by whatever 00:13:35.235 --> 00:13:38.765 codegen backend it has, followed by whatever linker you're using, there's just 00:13:38.765 --> 00:13:43.975 a lot of parts that really, really need to agree on what configuration means to them. 00:13:44.230 --> 00:13:44.900 That is correct. 00:13:44.930 --> 00:13:49.570 The actual flag is ld_classic, but that's the flag for LLD, which is invoked by 00:13:49.580 --> 00:13:57.738 CC, and the flag for CC is so what we were saying: " -Wl, -ld_classic," but 00:13:57.748 --> 00:13:59.478 then we're actually passing a Rust flag. 00:13:59.508 --> 00:14:09.188 So the full chain is '-Clink-arg=-Wl, ld_classic' which: don't ask 00:14:09.188 --> 00:14:10.678 me why I know that by heart! 00:14:10.678 --> 00:14:12.018 This is just the trauma- 00:14:12.062 --> 00:14:13.953 Gonna say trauma is the reason that you have to know that. 00:14:15.828 --> 00:14:16.538 Yes it is. 00:14:18.108 --> 00:14:19.358 Wow, this was a short story. 00:14:19.358 --> 00:14:20.573 I didn't think it was going to last that long. 00:14:20.932 --> 00:14:23.682 The problem was: I think it's safe to say I haven't checked the GitHub, 00:14:23.712 --> 00:14:27.740 but the person that has been behind most of the effort of like, hooking 00:14:27.740 --> 00:14:30.800 up rustc to Cranelift- he hasn't developed all of Cranelift by himself. 00:14:30.820 --> 00:14:31.510 It's a different team. 00:14:31.540 --> 00:14:33.990 It's like Wasmtime, the Wasmtime team? 00:14:33.990 --> 00:14:37.690 I don't know exactly who works on what, but like the Bytecode Alliance, Wasmtime, 00:14:37.710 --> 00:14:40.380 Cranelift, whatever-related-word-cloud. 00:14:41.050 --> 00:14:42.560 And then rustc is a different set of people. 00:14:42.680 --> 00:14:45.440 And then between rustc and Cranelift: Bjorn! 00:14:46.130 --> 00:14:49.920 Bjorn the third, no bjorn3, on GitHub, whom I met at EuroRust. 00:14:50.035 --> 00:14:51.569 I asked, " Are you getting paid for this? 00:14:51.569 --> 00:14:52.689 Do you get any kind of support?" 00:14:52.959 --> 00:14:55.639 And he was like, "Well, the Rust Foundation is giving some things 00:14:55.639 --> 00:14:57.309 including like a VM and whatnot." 00:14:57.639 --> 00:15:00.849 And as I was messing around with the Cranelift codegen backend, 00:15:00.869 --> 00:15:02.369 I realized: it's so close. 00:15:02.399 --> 00:15:03.339 It seems so close. 00:15:03.339 --> 00:15:05.944 I don't know if it actually runs because the last linker command failed. 00:15:06.254 --> 00:15:08.631 I asked him, "Hey, do do you have access to a Mac machine? 00:15:08.631 --> 00:15:10.316 Like an M1 or an M2 or anything? 00:15:10.356 --> 00:15:10.946 Have they given you a VM?" 00:15:11.726 --> 00:15:14.176 And the answer is of course, no, because it's so easy to give 00:15:14.176 --> 00:15:16.606 someone a 64-bit Linux VM, right? 00:15:16.864 --> 00:15:19.909 It costs like pennies or whatever on whatever cloud platform, 00:15:20.209 --> 00:15:22.049 but the M2s and whatnot... 00:15:22.834 --> 00:15:25.674 kind of hard to get by, kind of more awkward to start a 00:15:25.674 --> 00:15:26.714 VM yourself and everything. 00:15:26.714 --> 00:15:27.724 I had never done it before. 00:15:27.724 --> 00:15:31.964 I just heard that like recently it became possible to log into 00:15:31.964 --> 00:15:39.294 your iCloud account from a MacOS ARM VM ran from an ARM machine? 00:15:39.314 --> 00:15:41.556 I didn't even know that was "not a thing" before. 00:15:41.606 --> 00:15:43.376 I didn't even know that was a limitation that existed. 00:15:43.376 --> 00:15:46.936 Yeah, Mac has weird licensing rules on virtualization, which means 00:15:47.106 --> 00:15:51.796 essentially every virtualized, like, hosting provider of Mac is sort of in 00:15:51.796 --> 00:15:56.223 a gray area, and also they're required to have physical machines on hand. 00:15:56.463 --> 00:16:02.818 Like, you can only rent out the VM- not on demand, it has to be, like, in 00:16:02.828 --> 00:16:05.538 increments of 24 hours or something like that, which means they have 00:16:05.538 --> 00:16:08.878 to have physical units on the shelf that they only rent out once a day. 00:16:08.918 --> 00:16:09.348 I don't know. 00:16:09.558 --> 00:16:11.108 Maybe it's changed in recent years. 00:16:11.288 --> 00:16:18.048 Apple made the licensing of essentially OSX ridiculous, which makes virtualization 00:16:18.058 --> 00:16:21.018 of it very hard, which means in turns it becomes very expensive. 00:16:21.077 --> 00:16:25.287 So long story short, I created a VM on my Mac Studio, which 00:16:25.287 --> 00:16:28.827 I bought for video editing reasons, and it is way overpowered for what 00:16:28.827 --> 00:16:29.997 I need to do on a daily basis. 00:16:30.027 --> 00:16:31.347 Like I'm not always compiling Rust. 00:16:31.357 --> 00:16:33.641 Sometimes I think in between compiles, believe it or not. 00:16:33.871 --> 00:16:37.367 And so I made a VM, gave it to Bjorn and now he, he is actually able to 00:16:37.377 --> 00:16:41.297 build stuff and like test it in real time, not wait for CI, which it's as 00:16:41.297 --> 00:16:43.187 you know, it's a long iteration loop. 00:16:43.327 --> 00:16:43.707 Whatever. 00:16:44.097 --> 00:16:44.697 Can't find the words. 00:16:45.117 --> 00:16:49.527 It just makes you think, I don't know if he ever was asked that question before. 00:16:49.527 --> 00:16:53.117 Cause there's so many issues about 64-bit MacOS ARM. 00:16:53.507 --> 00:16:56.137 But he never had the machine, so he was never able to test. 00:16:56.167 --> 00:16:59.167 And it cost me next to nothing because I already had the machine here. 00:16:59.177 --> 00:17:01.547 I just created the VM, created a VPN, and then there you go. 00:17:01.587 --> 00:17:03.127 Remote desktop SSH. 00:17:03.127 --> 00:17:03.811 Go wild, do what you want. 00:17:04.276 --> 00:17:07.956 We're so surprisingly close to so many things that people think are 00:17:07.956 --> 00:17:14.166 very far away, just of getting the triple point of: the person who knows about 00:17:14.166 --> 00:17:18.641 this, to have the time and money to do it, and to have the resources to do it, 00:17:18.641 --> 00:17:20.184 like the access to the things they need. 00:17:20.231 --> 00:17:23.031 And you see this a bunch in Rust, I feel like: Someone makes some very 00:17:23.031 --> 00:17:27.045 tiny token donation, and they go, "Okay, well, yeah, then this person can 00:17:27.055 --> 00:17:28.515 buy the VM or the laptop they need." 00:17:28.675 --> 00:17:31.545 And then, two weeks later, something you thought was years away is done. 00:17:31.705 --> 00:17:32.415 And ready. 00:17:32.515 --> 00:17:37.075 There's so much of that inefficiency of resource distribution in reality. 00:17:37.360 --> 00:17:40.250 And this is why I'm happy to announce my new charity, which 00:17:40.270 --> 00:17:42.050 is called Mac minis for Everyone. 00:17:42.200 --> 00:17:44.200 You can donate now, go to macminisforeveryone. 00:17:44.230 --> 00:17:46.730 com, and I will send everyone a Mac mini. 00:17:46.950 --> 00:17:50.020 And then finally, we won't have to wait for things to compile on 00:17:50.020 --> 00:17:51.610 your fucking Acer laptop again. 00:17:52.250 --> 00:17:52.640 There you go. 00:17:58.867 --> 00:18:00.917 This episode is sponsored by fasterthanlime. 00:18:01.263 --> 00:18:04.553 fasterthanlime is a 34 year old person named Amos who makes articles and 00:18:04.553 --> 00:18:06.093 videos about computers for a living. 00:18:06.281 --> 00:18:08.621 They tend to go on long explorations to find out the answers to 00:18:08.621 --> 00:18:10.781 questions that have been bothering them since they were a kid. 00:18:11.422 --> 00:18:15.222 Rust and Linux are often involved, but also asynchronous IO, dynamic 00:18:15.222 --> 00:18:17.982 linking, thread locals, and other topics that only a few people get the 00:18:17.982 --> 00:18:19.472 chance to mess with on their own terms. 00:18:20.097 --> 00:18:23.637 You can support Amos's work by reading articles on their website, watching their 00:18:23.637 --> 00:18:27.247 videos on YouTube, and by sponsoring them on GitHub sponsors or Patreon. 00:18:27.693 --> 00:18:30.013 Thank you to Faster Than Lime for sponsoring this episode.