1 00:00:03,106 --> 00:00:06,726 Welcome to Compose, a software development podcast featuring 2 00:00:06,726 --> 00:00:10,056 the Rust programming language and how it interacts with the world. 3 00:00:11,616 --> 00:00:18,486 This episode features interoperability with C++ and how that might progress 4 00:00:18,516 --> 00:00:20,511 over the years and even decades ahead. 5 00:00:21,786 --> 00:00:24,486 To talk us through it all is David Sankel from Adobe. 6 00:00:25,266 --> 00:00:28,816 David is an invited speaker at this year's Rust Forge Conference. 7 00:00:29,536 --> 00:00:32,476 Rust Forge itself is happening in Wellington, New Zealand, and 8 00:00:32,476 --> 00:00:33,826 you're still very welcome to come. 9 00:00:35,446 --> 00:00:38,566 For those of you who would even like a discount, try visiting 10 00:00:38,876 --> 00:00:42,836 rustforgeconf.com/compose. 11 00:00:46,301 --> 00:00:48,781 I guess I'm at heart a a computer nerd. 12 00:00:48,901 --> 00:00:53,996 So I'm sitting here and in front of me is like these little microprocessors, right? 13 00:00:53,996 --> 00:00:55,286 That I'm just programming for fun. 14 00:00:55,796 --> 00:00:59,076 But I guess who I am probably of interest to this community, I am a 15 00:00:59,076 --> 00:01:03,996 principal scientist at Adobe and I run Adobe Software Technology Lab. 16 00:01:04,416 --> 00:01:09,386 And that is a team which is focused on what is going to be happening in the 17 00:01:09,386 --> 00:01:11,936 next 10 years with software development. 18 00:01:12,506 --> 00:01:15,896 And we're very much, our heads are in that space and we have to figure out how we're 19 00:01:15,896 --> 00:01:18,116 gonna have Adobe align with that future. 20 00:01:18,146 --> 00:01:22,126 And also how can we help steer the industry so that it can 21 00:01:22,126 --> 00:01:24,466 better support what's going to be happening going forward. 22 00:01:24,466 --> 00:01:26,026 So that's my day job. 23 00:01:26,096 --> 00:01:30,476 And I'm also very involved in C++ Standardization committee, and now more 24 00:01:30,476 --> 00:01:34,936 recently involved in Rust and specifically Rust Interop and doing that at Adobe. 25 00:01:34,936 --> 00:01:35,566 So that's 26 00:01:36,016 --> 00:01:38,631 I feel like for many people that kind of sounds like a bit of 27 00:01:38,631 --> 00:01:40,561 a dream job, look, basically. 28 00:01:41,701 --> 00:01:42,871 It is a dream job. 29 00:01:43,031 --> 00:01:45,021 Let me tell you a story about that because it's funny. 30 00:01:45,021 --> 00:01:48,691 So Sean parent who is known in the C++ community was running this team. 31 00:01:49,091 --> 00:01:51,611 And he invited me to, to join it. 32 00:01:51,701 --> 00:01:55,181 And he said that this is, she sold it pretty much as a dream job. 33 00:01:55,271 --> 00:01:58,391 And it was like, this is my dream job. 34 00:01:59,081 --> 00:02:02,231 And I got there and he convinced me to swap positions with him. 35 00:02:02,441 --> 00:02:05,261 So I would run the team and he would have my dream job. 36 00:02:05,441 --> 00:02:07,121 And he managed that. 37 00:02:07,271 --> 00:02:09,251 And I was like, I see what you did there. 38 00:02:09,311 --> 00:02:13,841 You created the dream job and then you took it for yourself and made 39 00:02:13,841 --> 00:02:16,181 me the manager of the team, which is, slightly less than the dream 40 00:02:16,181 --> 00:02:17,411 job of actually being on the team. 41 00:02:17,451 --> 00:02:17,811 But yeah 42 00:02:17,861 --> 00:02:24,201 It is, it takes a certain amount of humility to humility is possibly the 43 00:02:24,201 --> 00:02:32,541 wrong word, but to be really, and supportive of others and actually sharing 44 00:02:32,541 --> 00:02:35,481 in the facilitation of others' success. 45 00:02:35,481 --> 00:02:41,391 And so the being a manager is, I actually think that's a better paid, let's say, but 46 00:02:41,391 --> 00:02:44,571 also a slightly underrated career track. 47 00:02:45,001 --> 00:02:48,061 People look down on management in some ways as no longer 48 00:02:48,061 --> 00:02:49,171 being technical or something. 49 00:02:49,171 --> 00:02:53,961 But I think that it it is in some sense, gracious that you are able to give 50 00:02:53,961 --> 00:02:59,331 that up and just to enable, let's say, multiply your own thoughts by X. So 51 00:02:59,341 --> 00:03:00,781 that's a good thing too, in my opinion. 52 00:03:01,781 --> 00:03:02,771 Oh yeah, definitely. 53 00:03:02,861 --> 00:03:06,891 And I say this mostly in jest and as lead of software technology 54 00:03:06,891 --> 00:03:10,761 lab, I'm still very much involved in doing the actual technology. 55 00:03:10,831 --> 00:03:11,701 'Cause it's a very small team. 56 00:03:11,701 --> 00:03:13,561 There's only four people in the team. 57 00:03:13,991 --> 00:03:18,051 So yeah, I get to stay hands on and do all the crap work that Sean was doing before. 58 00:03:18,141 --> 00:03:21,471 So I'm happy to have him spend his time not doing that crap work and take that 59 00:03:21,516 --> 00:03:21,756 Yeah, 60 00:03:21,756 --> 00:03:24,336 look, some people have to be professional, meaning goers, now 61 00:03:25,336 --> 00:03:25,556 Yes. 62 00:03:25,666 --> 00:03:32,086 anyway, we're gonna need the like the actual chip number or like the 63 00:03:32,331 --> 00:03:35,241 part number of the microcontroller that you're playing with at some 64 00:03:35,241 --> 00:03:37,021 point because I'm sure that the, 65 00:03:37,466 --> 00:03:41,726 This is the STM32C011J4M6. 66 00:03:41,966 --> 00:03:42,266 There you go. 67 00:03:42,321 --> 00:03:46,911 For anyone that is coming to Rust Forge in, I guess it's six weeks time from 68 00:03:46,911 --> 00:03:52,961 this recording, Espressif is actually donating several dozen microcontrollers. 69 00:03:52,991 --> 00:03:55,871 We haven't actually announced this, so this is the big reveal, 70 00:03:56,201 --> 00:03:57,821 but it should be the case. 71 00:03:58,211 --> 00:04:00,306 I hope we're getting at least one for everyone that it 72 00:04:00,306 --> 00:04:01,481 turn, that, that turns up. 73 00:04:01,841 --> 00:04:05,441 Yeah, no, so there, there's a segue in there as well, but that's really great 74 00:04:05,981 --> 00:04:06,431 and. 75 00:04:07,356 --> 00:04:08,796 I am gonna be first in line for that thing. 76 00:04:08,831 --> 00:04:09,311 Awesome. 77 00:04:09,731 --> 00:04:10,151 Awesome. 78 00:04:10,571 --> 00:04:17,501 The the next thing is, so it, you are looking at the next 10 years of 79 00:04:17,501 --> 00:04:22,451 programming, let's say, in a very broad way... rust is part of that story. 80 00:04:23,451 --> 00:04:23,841 Yes. 81 00:04:24,201 --> 00:04:29,231 When do you think large companies, such as Adobe or there are many 82 00:04:29,501 --> 00:04:34,391 billion dollar companies, let's say, when do you think large enterprises 83 00:04:34,901 --> 00:04:37,526 would've really started to notice rust? 84 00:04:38,526 --> 00:04:40,326 I think it's already started. 85 00:04:40,536 --> 00:04:41,636 And I can tell you 86 00:04:41,636 --> 00:04:42,481 Please do. 87 00:04:42,601 --> 00:04:42,841 Yeah. 88 00:04:42,896 --> 00:04:43,616 from the beginning. 89 00:04:43,626 --> 00:04:47,936 Back in 2024 they created this memory Safety tiger team. 90 00:04:48,326 --> 00:04:48,596 Okay. 91 00:04:48,596 --> 00:04:52,506 And it included reps from security, the software technology lab developer 92 00:04:52,536 --> 00:04:56,346 experience, and legal reps. And the reason that it was created is because 93 00:04:56,346 --> 00:05:01,056 there was all this discussion of legislation related to memory safety. 94 00:05:01,656 --> 00:05:05,526 And they wanted to figure out like, what is this? 95 00:05:05,526 --> 00:05:06,936 What are we gonna do about it? 96 00:05:07,036 --> 00:05:11,826 And you had the NSA you had, cisa, you had the White House 97 00:05:11,826 --> 00:05:13,266 even chiming in on this stuff. 98 00:05:13,326 --> 00:05:16,626 And at one point there was even a bill, I think it was like a defense spending bill, 99 00:05:16,626 --> 00:05:20,441 like a typical bill that, that when it was in the House of Representatives, the 100 00:05:20,441 --> 00:05:26,171 Congress they had like creating a memory safety plan for the Department of Defense. 101 00:05:26,291 --> 00:05:30,101 And this little line got removed when it went in the Senate. 102 00:05:30,501 --> 00:05:33,561 So it didn't actually end up in the final thing that got approved, but the 103 00:05:33,561 --> 00:05:36,441 fact that it was there was like huge. 104 00:05:36,861 --> 00:05:39,921 And so this team got together and we're like, okay, what 105 00:05:39,921 --> 00:05:41,481 do we need to do about this? 106 00:05:41,481 --> 00:05:45,611 And we investigated and we looked at what was going on with like the 107 00:05:45,611 --> 00:05:47,051 published data that people were having. 108 00:05:47,051 --> 00:05:50,051 So you look at the numbers, it's okay, the number of exploits in software 109 00:05:50,441 --> 00:05:55,511 that are due to memory safety versus all the other exploits was like 70%. 110 00:05:55,841 --> 00:05:58,891 And you look at Microsoft's numbers, it's 70%. 111 00:05:58,891 --> 00:06:01,211 You look at Google's numbers at 70%. 112 00:06:01,271 --> 00:06:03,811 And then we looked internally and it's yeah, it's also 113 00:06:03,811 --> 00:06:04,951 right around that number too. 114 00:06:04,951 --> 00:06:09,961 So there, there's like something to this and the idea that you can switch your 115 00:06:09,961 --> 00:06:15,931 technology and get rid of 70% of your memory safety vulnerabilities like that. 116 00:06:16,591 --> 00:06:17,086 That was a big 117 00:06:17,176 --> 00:06:17,396 And 118 00:06:18,081 --> 00:06:22,551 my recollection of those papers as well, it's not just 70% of bugs, let's say, but 119 00:06:22,581 --> 00:06:25,851 70% of severe and critical, severe use. 120 00:06:25,881 --> 00:06:27,096 So it's like these are the. 121 00:06:27,841 --> 00:06:31,261 The majority of the worst cases of security problems. 122 00:06:31,651 --> 00:06:37,231 And yeah and we, and so this was a, so I guess it sounds like there were 123 00:06:37,231 --> 00:06:41,661 probably engineers thinking, ah, for the last multiple decades, and I'd 124 00:06:41,661 --> 00:06:47,261 be interested in C++ has had this in front of mind for a long time. 125 00:06:47,801 --> 00:06:52,841 How do we deal with buffer overflows and how do we deal with memory safety in a 126 00:06:52,841 --> 00:06:57,941 way that is strongly backwards compatible, which I suppose is the C++ mantra really. 127 00:06:58,581 --> 00:07:05,861 And then we have this new programming language, which somehow is able to 128 00:07:05,951 --> 00:07:12,671 offer memory safety without imposing garbage collection or some sort of 129 00:07:12,671 --> 00:07:17,921 other form of runtime checking to be able to allow things to work. 130 00:07:19,001 --> 00:07:24,641 But it was really the, it sounds like it was really the fact that it had started 131 00:07:24,641 --> 00:07:30,671 to introduce legal risk that really started to motivate internal people to 132 00:07:30,671 --> 00:07:34,181 say, okay, we're gonna fund something, at least some sort of discussion to make 133 00:07:34,181 --> 00:07:35,771 a decision on how we're gonna proceed. 134 00:07:36,771 --> 00:07:43,901 It, for us, like the legal risk was why upper level management got involved and 135 00:07:43,901 --> 00:07:46,301 they said, can you investigate this? 136 00:07:46,931 --> 00:07:51,101 When we investigated it, it wasn't like, oh yes, there is significant legal risk. 137 00:07:51,101 --> 00:07:53,011 We're looking at these, laws that are coming out. 138 00:07:53,431 --> 00:07:54,391 It was more. 139 00:07:55,361 --> 00:07:56,081 What about 140 00:07:56,166 --> 00:07:56,606 Yeah. 141 00:07:56,606 --> 00:07:56,966 actually. 142 00:07:57,101 --> 00:08:01,001 want our customers to use safe software. 143 00:08:01,451 --> 00:08:02,771 That's totally in our interest. 144 00:08:02,811 --> 00:08:06,471 And that's actually our driving motivation in the, in this group, 145 00:08:06,471 --> 00:08:10,251 is we want to produce the most safe software that we can for our consumers. 146 00:08:10,251 --> 00:08:15,991 You imagine, if somebody comes out with a PDF that exploits your computer, 147 00:08:16,171 --> 00:08:17,791 how many people are using Acrobat? 148 00:08:17,841 --> 00:08:21,171 We, look, there's more people using, have Acrobat Reader on their computer 149 00:08:21,381 --> 00:08:22,851 than people that use Firefox, 150 00:08:23,031 --> 00:08:24,081 like in total. 151 00:08:24,291 --> 00:08:28,561 So it's a huge swath and if you take state actors or whatever yeah, 152 00:08:28,561 --> 00:08:30,951 if you wanna, get into people's computers, you're gonna try to get 153 00:08:30,951 --> 00:08:32,751 in the software that most people use. 154 00:08:32,781 --> 00:08:34,431 And we're definitely in that space. 155 00:08:34,461 --> 00:08:34,731 Yeah. 156 00:08:34,731 --> 00:08:39,891 and PDF is not a simple format to protect against really. 157 00:08:40,051 --> 00:08:43,291 I don't mean this, I don't mean to cast any shade, but we've got multiple 158 00:08:43,291 --> 00:08:44,581 programming language interpreters. 159 00:08:44,581 --> 00:08:49,231 There's at least a JavaScript runtime plus there's two other languages, right? 160 00:08:49,231 --> 00:08:53,941 It's own little macro language as well as Postscript in there too, I think. 161 00:08:54,181 --> 00:08:54,541 but. 162 00:08:55,541 --> 00:08:57,281 It is extremely complicated. 163 00:08:57,281 --> 00:08:57,761 Yes. 164 00:08:57,841 --> 00:09:01,321 And the fact that it can embed a JPEG or something like that and 165 00:09:01,411 --> 00:09:01,711 Oh yeah. 166 00:09:01,741 --> 00:09:06,151 'cause now every single codec that you support also has decompression. 167 00:09:06,241 --> 00:09:10,821 And you've also got inter essentially, like you have these baby interpreters 168 00:09:10,821 --> 00:09:13,911 for file formats, which are also exploitable as browser vendors 169 00:09:13,911 --> 00:09:15,591 have discovered multiple times. 170 00:09:16,591 --> 00:09:16,831 Yeah. 171 00:09:16,831 --> 00:09:17,791 And, okay. 172 00:09:17,881 --> 00:09:24,331 So that would be good to be able to defend against these kind of poisoned 173 00:09:24,361 --> 00:09:29,341 PDFs, which of course this is, that that's a horrifically big vector of 174 00:09:29,701 --> 00:09:33,331 problems because of course you're going to double click the PDF and open it. 175 00:09:33,491 --> 00:09:34,991 When you get that via email. 176 00:09:35,991 --> 00:09:36,441 Yeah. 177 00:09:37,141 --> 00:09:40,771 But like on the flip side, so from a software technology lab perspective 178 00:09:41,371 --> 00:09:45,751 before memory safety became something that everybody knew about and was 179 00:09:45,751 --> 00:09:49,261 talking about, we were already trying to figure out what's going to be 180 00:09:49,771 --> 00:09:51,421 the future of programming languages. 181 00:09:51,731 --> 00:09:52,611 Is it going to be C++? 182 00:09:53,241 --> 00:09:54,921 Is it gonna be one of these other languages? 183 00:09:55,251 --> 00:09:59,631 And from our perspec perspective on that team, the memory safety wasn't 184 00:09:59,631 --> 00:10:03,461 actually the most interesting thing about rust, as a language to move to. 185 00:10:04,031 --> 00:10:09,401 It's just, flatly better tech for most applications, even if you didn't 186 00:10:09,401 --> 00:10:14,231 have the memory safety aspect, if you look at like enums pattern matching 187 00:10:14,231 --> 00:10:19,341 traits, like all the procedurally macro driven libraries that are in there. 188 00:10:19,731 --> 00:10:22,881 And there's a lot of stuff like that just made it interesting. 189 00:10:23,121 --> 00:10:26,241 Hey, we want people to develop in these languages because they're gonna be more 190 00:10:26,241 --> 00:10:29,871 productive and the code they're writing is gonna be more maintainable and it's 191 00:10:29,871 --> 00:10:33,171 gonna be easier to hire developers because it's easier to get them to learn 192 00:10:33,171 --> 00:10:34,491 the language and be able to work on it. 193 00:10:34,491 --> 00:10:38,811 This, these are like other big drivers that I think stand on their own 194 00:10:38,811 --> 00:10:40,791 without even looking at the memory 195 00:10:40,986 --> 00:10:46,656 Yeah, I have been in very similar discussions where people look and say, 196 00:10:46,686 --> 00:10:48,366 oh gosh, we've got this huge problem. 197 00:10:48,516 --> 00:10:49,956 And that's the motivating factor. 198 00:10:50,106 --> 00:10:53,316 But it turns out that rust is also a very good programming language 199 00:10:53,616 --> 00:10:55,116 and it has lots of nice things. 200 00:10:55,116 --> 00:10:59,606 We have Iterators, for example, that compose and just chain together quite 201 00:10:59,606 --> 00:11:01,976 nicely and also get optimized quite well. 202 00:11:02,426 --> 00:11:09,426 and then, from the point of view of comparing rust to, let's say a Java C 203 00:11:09,426 --> 00:11:14,736 Sharp or even a Python or something, and my background is primarily in web 204 00:11:14,736 --> 00:11:18,276 services and so forth, but we would also say that we could probably reduce 205 00:11:18,276 --> 00:11:23,416 memory by, I don't know, 90% and the cost of running an application. 206 00:11:23,516 --> 00:11:28,566 Suddenly we only require 80 megabytes of RAM at steady state. 207 00:11:28,566 --> 00:11:31,866 So we don't need giant instances with lots and lots of heap. 208 00:11:32,016 --> 00:11:36,636 So actually we are able to produce applications that run faster, serve 209 00:11:36,636 --> 00:11:38,616 users quicker, and or also more secure. 210 00:11:38,616 --> 00:11:42,346 And yeah, and by the way, yeah, also rust is quite a nice programming language. 211 00:11:43,346 --> 00:11:48,326 And now we've got this other problem that, you are advocating for a 212 00:11:48,326 --> 00:11:53,906 programming language or at least saying to the rest of the company, by the 213 00:11:53,906 --> 00:11:56,901 way the grass is greener over there. 214 00:11:57,901 --> 00:12:02,591 How, but at some point, the immune system of the organization 215 00:12:02,801 --> 00:12:10,251 responds to you and says, ah, we have a lot of other languages. 216 00:12:10,251 --> 00:12:12,561 We don't want to add another one to the mix. 217 00:12:13,561 --> 00:12:14,131 Yep. 218 00:12:14,501 --> 00:12:18,851 And we're definitely see those really healthy immune system responses 219 00:12:18,941 --> 00:12:20,111 just like any, anywhere else. 220 00:12:20,621 --> 00:12:24,016 And I think that a lot of what's been useful. 221 00:12:25,166 --> 00:12:29,246 To dispel some of the concerns that people have has been the great 222 00:12:29,246 --> 00:12:31,106 data that's been put out there. 223 00:12:31,446 --> 00:12:36,366 So one of the concerns someone might say is oh, but Rust is really complex, right? 224 00:12:36,416 --> 00:12:37,916 It's a hard language to learn. 225 00:12:38,366 --> 00:12:41,456 We can't wait for people to spend three years learning this language 226 00:12:41,516 --> 00:12:43,136 to be able to be productive in it. 227 00:12:44,026 --> 00:12:48,086 And then for that point out to the Google study that came out where it 228 00:12:48,086 --> 00:12:52,056 was like, I think it was after three months of development and rust from 229 00:12:52,056 --> 00:12:55,656 someone that has no experience at the beginning, two thirds outta people 230 00:12:55,916 --> 00:12:59,906 felt like they were very confident and productive in this language, right? 231 00:12:59,906 --> 00:13:02,216 So that you just point to the data. 232 00:13:02,216 --> 00:13:06,126 It's like you can go with your anecdotal, feeling about it, but 233 00:13:06,126 --> 00:13:08,166 why do that when we can look at the data when there's been like, good 234 00:13:08,166 --> 00:13:09,336 studies that have been put out there. 235 00:13:09,766 --> 00:13:13,066 And so a lot of things like this it's just like getting the right information 236 00:13:13,066 --> 00:13:14,356 in front of people so that they can. 237 00:13:14,971 --> 00:13:16,651 They don't have to go based on their gut. 238 00:13:16,651 --> 00:13:19,021 They can go based on something more solid and tangible. 239 00:13:19,151 --> 00:13:20,626 I think that's been extremely helpful. 240 00:13:20,781 --> 00:13:23,151 I've never worked at Google, but I've got lots of friends there. 241 00:13:23,781 --> 00:13:29,841 And one of the things that makes me believe that research is that Google 242 00:13:29,841 --> 00:13:35,541 has full control over its build system and the developer experience as well. 243 00:13:35,966 --> 00:13:36,296 Mm-hmm. 244 00:13:36,371 --> 00:13:39,101 of the things that I would really like to pick up on was 245 00:13:39,221 --> 00:13:40,841 another claim around hiring. 246 00:13:41,141 --> 00:13:43,781 'cause I've heard that hiring rust developers is really hard. 247 00:13:44,781 --> 00:13:50,391 So one thing I can say is that hiring C++ developers is incredibly hard. 248 00:13:50,991 --> 00:13:54,441 And for us I can't share with you the data, but I can say that 249 00:13:54,441 --> 00:13:58,521 it is astronomically harder than hiring for any other language 250 00:13:58,521 --> 00:14:01,031 that we have especially in the us. 251 00:14:01,371 --> 00:14:04,771 And that is another one of the motivations as to why we're looking at. 252 00:14:05,521 --> 00:14:08,491 Other languages that we can bring on developers because if you've 253 00:14:08,491 --> 00:14:13,291 got a new project and you can't staff it then, but you can change 254 00:14:13,291 --> 00:14:14,341 the language and staff it better. 255 00:14:14,341 --> 00:14:16,621 You may just change the language based on that alone. 256 00:14:17,251 --> 00:14:22,481 And so we looked a lot at the adoption of Rust in the wider community and we looked 257 00:14:22,481 --> 00:14:24,521 at again, we're like, let's see the data. 258 00:14:25,061 --> 00:14:29,091 And so there was a Stack Overflow survey that that came out. 259 00:14:29,091 --> 00:14:34,081 I think if you look at the 2024 Stack Overflow survey we crunched the data and 260 00:14:34,081 --> 00:14:38,951 we looked at the number of developers that are C++, and we looked at what 261 00:14:38,951 --> 00:14:41,921 portion of C++ developers right now. 262 00:14:42,481 --> 00:14:47,761 Are also rust developers and it was 20%, kinda blew our mind. 263 00:14:48,031 --> 00:14:51,061 20% of all C++ developers are also developing rust. 264 00:14:51,061 --> 00:14:53,881 And this is like the question was asked like, have you programmed this 265 00:14:53,881 --> 00:14:55,471 seriously, like in the past year? 266 00:14:56,131 --> 00:14:59,581 And that's huge and, but maybe that's Stack Overflow could be weird. 267 00:14:59,611 --> 00:15:01,051 Only weird people flow out that survey. 268 00:15:01,441 --> 00:15:06,901 But we also looked at the data from GitHub and one for every five C++ 269 00:15:06,901 --> 00:15:10,111 PRs on GitHub, there's one Rust PR. 270 00:15:10,921 --> 00:15:12,931 So again, like similar kind of number. 271 00:15:12,991 --> 00:15:17,521 There's just a lot of people programming Rust it turns out. 272 00:15:17,851 --> 00:15:21,101 And I've asked this question at conferences that I give talks at what, how 273 00:15:21,101 --> 00:15:22,511 many of you are programming Rust as well? 274 00:15:22,511 --> 00:15:23,861 It's like similar number everywhere. 275 00:15:23,861 --> 00:15:25,721 It seems to be right around 20%. 276 00:15:26,141 --> 00:15:29,761 And that's changed from 16% like five years ago. 277 00:15:30,121 --> 00:15:33,421 It's just been like a drastic increase in the number of Rust developers. 278 00:15:33,421 --> 00:15:36,856 So I, I. Do not have concerns of hiring Rust developers. 279 00:15:36,856 --> 00:15:38,956 If you say you got a rust position, you got like a bunch 280 00:15:38,956 --> 00:15:43,036 of people going wild because they would love to do rust full time. 281 00:15:43,096 --> 00:15:47,416 Just because it's, you just feel better programming in that language. 282 00:15:47,416 --> 00:15:50,086 Like going from Rust back to C++ kind of hurts a little bit. 283 00:15:50,086 --> 00:15:54,266 You die a little bit inside when you come back to that C++ project, and so there's 284 00:15:54,286 --> 00:15:59,626 just enough enthusiasm for it that's, that isn't a concern, although people raise it. 285 00:15:59,686 --> 00:16:01,246 But look at the data again. 286 00:16:01,296 --> 00:16:03,576 It, it really isn't in actuality a concern. 287 00:16:03,606 --> 00:16:07,986 Yeah, there is a lot of I forget the term, but it, there's a particular 288 00:16:08,106 --> 00:16:12,816 psychological bias where we are afraid of change because of its loss aversion, I 289 00:16:12,816 --> 00:16:21,086 think where it's will, people are hesitant to take on change or risk even when the. 290 00:16:21,971 --> 00:16:24,281 The outcome is likely to be positive. 291 00:16:24,701 --> 00:16:28,691 The other thing that I have found people resisting is this idea 292 00:16:28,691 --> 00:16:30,521 that Rust is as good as it sounds. 293 00:16:31,331 --> 00:16:34,661 People make this claim around, oh, it's a really lovely programming language to use. 294 00:16:34,661 --> 00:16:37,241 We finally have a genuine type system. 295 00:16:37,421 --> 00:16:41,561 We have no, we now, pointers don't really exist in the language. 296 00:16:42,561 --> 00:16:45,531 And it's like, where's the catch? 297 00:16:46,531 --> 00:16:55,396 and I guess the next thing to chat about would be when so Photoshop sounds 298 00:16:55,396 --> 00:16:57,556 like there are changes happening. 299 00:16:59,206 --> 00:16:59,926 What's happening? 300 00:17:00,926 --> 00:17:01,106 Yeah. 301 00:17:01,206 --> 00:17:07,226 We are we are at the stage to where we've done a couple of pilot projects of where 302 00:17:07,226 --> 00:17:11,466 we're doing rust inside of Photoshop, and we're seeing how that's going. 303 00:17:11,466 --> 00:17:12,556 Like what exactly happened? 304 00:17:12,556 --> 00:17:15,286 So we took a team, we said, okay, we have a project here. 305 00:17:15,596 --> 00:17:18,836 We want you to do this in rust for Photoshop. 306 00:17:19,166 --> 00:17:20,846 What is your experience of that? 307 00:17:21,266 --> 00:17:26,441 And it came out what you would expect, the people who use the language loved it. 308 00:17:26,491 --> 00:17:29,561 They're reticent to go back to like a C++ kind of thing. 309 00:17:30,061 --> 00:17:34,141 Um, after that, um, they saw the productivity gains, uh, 310 00:17:34,231 --> 00:17:35,821 especially being able to use cargo. 311 00:17:36,091 --> 00:17:38,221 I mean, one of the most powerful experiences you can 312 00:17:38,221 --> 00:17:39,541 have is when you're like, Hmm. 313 00:17:40,351 --> 00:17:43,591 I think it'd be nice if we had a, a, a JavaScript interpreter in 314 00:17:43,591 --> 00:17:47,791 this program, and you just add a JavaScript interpreter with like 315 00:17:47,791 --> 00:17:50,871 one line, uh, in your cargo thing. 316 00:17:51,301 --> 00:17:56,521 But coming from a C++ background that is black magic, that should not be that 317 00:17:56,821 --> 00:18:02,341 I look who doesn't love tinkering with sea make or whichever build systems 318 00:18:02,431 --> 00:18:04,531 that you feel like creating today. 319 00:18:04,871 --> 00:18:07,211 Why even have one build system when you could have four in 320 00:18:07,211 --> 00:18:08,141 your project or something? 321 00:18:08,141 --> 00:18:12,791 I don't mean to be too rude, but Rust has definitely benefited from being 322 00:18:12,791 --> 00:18:19,196 late essentially, or learning from the experiences of the past and oh, 323 00:18:19,196 --> 00:18:20,996 that's, so that's really validating. 324 00:18:21,116 --> 00:18:25,106 So you've got this public research, we've got this Google study, 325 00:18:25,106 --> 00:18:28,286 we've had other kind of language other wider bits and pieces. 326 00:18:28,496 --> 00:18:35,336 And now internally you're validating that yourself and you're not proven, you are 327 00:18:35,486 --> 00:18:42,686 that then able to demonstrate that not just the language is okay, but it would 328 00:18:42,686 --> 00:18:48,786 work well inside the build system or inside the context of of the rest of the 329 00:18:48,786 --> 00:18:50,526 environment within, inside the company. 330 00:18:51,526 --> 00:18:51,976 Yeah. 331 00:18:52,226 --> 00:18:56,721 There was one thing that the experience pointed to that was. 332 00:18:57,791 --> 00:18:59,201 Bad for rust. 333 00:18:59,206 --> 00:18:59,486 Okay. 334 00:18:59,491 --> 00:19:01,031 'cause they give outta how many stars? 335 00:19:01,031 --> 00:19:02,711 Three, four outta five stars. 336 00:19:02,711 --> 00:19:03,401 Everything's great. 337 00:19:03,401 --> 00:19:04,091 Hunky dory. 338 00:19:04,571 --> 00:19:09,271 But when you get to the C++ interop, that was where it was like that's a problem. 339 00:19:09,271 --> 00:19:15,621 And so what we used was c bin gen. So essentially you're writing these, 340 00:19:16,071 --> 00:19:18,651 the C functions in rust, right? 341 00:19:18,651 --> 00:19:21,241 But that you can call from C and that'll generate a header, which 342 00:19:21,241 --> 00:19:22,411 we'll call those C functions. 343 00:19:22,781 --> 00:19:25,751 And then the implementation of those functions will call 344 00:19:25,801 --> 00:19:26,881 your underlying rust code. 345 00:19:27,481 --> 00:19:35,361 And this involves a lot of unsafe, it also involves a lot of cism stuff that, that 346 00:19:35,361 --> 00:19:38,431 C++ people aren't familiar with, right? 347 00:19:38,431 --> 00:19:39,931 If you go back to the. 348 00:19:40,771 --> 00:19:44,881 Back in the day when you were writing C callbacks, you have a callback 349 00:19:45,391 --> 00:19:47,941 pointer function that you function, pointer that you pass in, but you 350 00:19:47,941 --> 00:19:51,301 also have to pass and avoid star for the context of that callback. 351 00:19:51,301 --> 00:19:54,921 And that's just like this idiom that you do in C to do callbacks. 352 00:19:55,401 --> 00:19:59,361 But if you want to have a interop layer in C you have to do this 353 00:19:59,361 --> 00:20:00,441 and you have to understand it. 354 00:20:00,861 --> 00:20:05,901 And so what we found is that when you take somebody who's like C++, you 355 00:20:05,901 --> 00:20:09,651 know, expert or very familiar with the language, and the first thing they 356 00:20:09,651 --> 00:20:15,681 need to do is go to the most gnarly, awful part of rust, which is unsafe. 357 00:20:15,711 --> 00:20:20,631 Weird stuff with pointers, like that experience is not 358 00:20:20,826 --> 00:20:22,576 so let me paint the picture. 359 00:20:22,696 --> 00:20:24,311 So what we would like to do. 360 00:20:25,976 --> 00:20:31,226 Is basically pass objects around and their interfaces in some sort of ideal, 361 00:20:31,506 --> 00:20:39,476 essentially extend C++ or like bolt, either graft rust onto C++ in some way. 362 00:20:39,736 --> 00:20:44,406 Or graft C++ into rust or whatever. 363 00:20:44,426 --> 00:20:50,246 So that a rust programmer could just interact with methods from C++, uh, as 364 00:20:50,376 --> 00:20:52,956 though they were essentially native. 365 00:20:53,046 --> 00:20:58,746 But we don't have anything like that because the two compilers don't know 366 00:20:58,746 --> 00:21:00,726 any really, we don't know enough. 367 00:21:01,206 --> 00:21:02,211 They don't know enough about each other. 368 00:21:02,241 --> 00:21:07,156 They it's a, you would need to teach Rust, all of C++ in order 369 00:21:07,156 --> 00:21:09,716 for that to work in that way. 370 00:21:10,366 --> 00:21:15,096 maybe I should also chat a little bit about this concept of passing function 371 00:21:15,096 --> 00:21:20,436 pointers around for anyone that's although maybe you can correct me here, 372 00:21:20,436 --> 00:21:24,866 David, because I'm sure that this isn't technically precise, but essentially the 373 00:21:24,866 --> 00:21:29,256 C++ and Rust worlds, they need to talk to each other through symbols in the binary. 374 00:21:29,256 --> 00:21:31,836 So they, when they lo the linker creates things and all they really 375 00:21:31,836 --> 00:21:33,426 have is text strings to match on. 376 00:21:33,486 --> 00:21:34,696 It's or memory addresses. 377 00:21:35,266 --> 00:21:38,416 That void star is a memory address. 378 00:21:38,596 --> 00:21:42,106 And then C C doesn't really have a strong type system. 379 00:21:42,106 --> 00:21:46,606 So you can basically say that any chunk of memory is any given type. 380 00:21:46,606 --> 00:21:51,196 So what you'll typically do is if you're calling some external function or 381 00:21:51,286 --> 00:21:56,416 trying to call one of its functions is get the symbol, find, de resolve this. 382 00:21:56,416 --> 00:21:58,126 The, you ask the linker to resolve the. 383 00:21:58,831 --> 00:22:04,231 Memory address, what have you, and then call the function with a buffer that 384 00:22:04,231 --> 00:22:09,851 you have allocated and then expect, just hope that the function does what it 385 00:22:09,851 --> 00:22:17,951 says, and then you need to reinterpret your buffer as the result type. 386 00:22:18,041 --> 00:22:21,171 So essentially this void star just means any type at all. 387 00:22:21,261 --> 00:22:22,791 How correct was that? 388 00:22:22,911 --> 00:22:24,231 It was that ish. 389 00:22:25,231 --> 00:22:25,681 Yeah. 390 00:22:25,771 --> 00:22:26,011 Yeah. 391 00:22:26,011 --> 00:22:26,491 That's. 392 00:22:26,776 --> 00:22:28,276 That's generally correct. 393 00:22:28,276 --> 00:22:32,706 So when interfacing between C++ and Rust, if you want to do something 394 00:22:33,186 --> 00:22:38,016 higher level than C right, you're usually have a C intermediate 395 00:22:38,016 --> 00:22:41,346 layer and the C intermediate layer. 396 00:22:41,396 --> 00:22:45,806 Let's say you want to pass a C++ object over to Rust 397 00:22:45,876 --> 00:22:48,476 Rust doesn't have a idea of what C++ objects are. 398 00:22:48,476 --> 00:22:53,586 So you get a void star or you get a pointer, just a, an address and memory. 399 00:22:54,366 --> 00:22:58,726 And so when you wanna call back into C++, you just pass that 400 00:22:58,746 --> 00:23:01,596 pointer back and hopefully it does the right thing and then do that. 401 00:23:01,596 --> 00:23:04,656 But it's not just for the object, it's for any argument, for any return type. 402 00:23:05,076 --> 00:23:08,646 All this is going through some like void stars under the hood. 403 00:23:09,006 --> 00:23:13,656 But I ideally you want to be better than what we did with that project, 404 00:23:13,656 --> 00:23:16,506 which is using like cbindgen, you want to have something which is richer. 405 00:23:17,031 --> 00:23:21,011 And get this kind of interop working between C++ and 406 00:23:21,011 --> 00:23:22,281 Rust in a more seamless way. 407 00:23:22,761 --> 00:23:26,541 And so a lot of my work has been how do we make this happen? 408 00:23:26,701 --> 00:23:27,841 And how do we get this to work? 409 00:23:27,841 --> 00:23:31,116 And I can go very deep into that subject if you're interested, if... 410 00:23:31,816 --> 00:23:36,346 Hit me because the this is something that lots of people are thinking 411 00:23:36,386 --> 00:23:39,921 a lot about, and if you wanna share, they would love to listen. 412 00:23:40,921 --> 00:23:41,311 Sure. 413 00:23:41,411 --> 00:23:42,521 So to start out, 414 00:23:42,621 --> 00:23:46,821 the model of objects is very different between C++ and Rust. 415 00:23:46,871 --> 00:23:53,741 If you have a Rust object at some place in memory, you can always just 416 00:23:53,891 --> 00:23:56,861 copy the bites from that place in memory to another place in memory. 417 00:23:57,161 --> 00:24:01,391 And that effectively moves the object and the object is valid in this other spot. 418 00:24:01,841 --> 00:24:05,501 So if you do assignment in rust, like A equals B, um. 419 00:24:05,801 --> 00:24:10,241 You know, it, the, the way to think about it is, is basically taking the memory that 420 00:24:10,391 --> 00:24:14,561 or b was pointing to and, and make that the memory, the A is pointing to, right? 421 00:24:14,561 --> 00:24:16,091 It's just moving the memory around. 422 00:24:16,821 --> 00:24:25,361 Um, so in, in C++ that's not how things work in general, if you have a, an object, 423 00:24:26,201 --> 00:24:30,461 like let's say you have a, a, a stu list, like a link list or something along these 424 00:24:30,461 --> 00:24:34,181 lines, and the stu list is implemented this way, uh, in, in many compilers. 425 00:24:34,251 --> 00:24:35,091 Oh, there is another, 426 00:24:35,541 --> 00:24:36,261 lists. 427 00:24:36,641 --> 00:24:40,141 because obviously in C++ we've got, we don't have one compiler. 428 00:24:40,321 --> 00:24:44,341 We have a standard which is implemented by many vendors 429 00:24:46,251 --> 00:24:49,401 yeah, and in this case the standard library, the implementation of 430 00:24:49,401 --> 00:24:50,961 standard lists is different. 431 00:24:50,961 --> 00:24:54,341 Now, one, one implementation that Microsoft has I dunno if they 432 00:24:54,341 --> 00:24:58,001 still have it, but they had for years, is that the end node of 433 00:24:58,001 --> 00:25:02,951 the list is in the class itself. 434 00:25:03,821 --> 00:25:07,251 So you create a stud list the end nodes in the class itself. 435 00:25:07,251 --> 00:25:09,621 It doesn't have to allocate when you first create a stud list because it 436 00:25:09,621 --> 00:25:14,541 just has the start node, pointer point to an object in the class itself. 437 00:25:15,201 --> 00:25:19,911 Now the problem is if you take the memory of a stud list and you just put 438 00:25:19,911 --> 00:25:24,451 it to a different location and say, okay, now the object's over there, that 439 00:25:24,451 --> 00:25:29,671 end Noe is now pointing back to where the original stood list was, right? 440 00:25:30,241 --> 00:25:33,656 So you don't actually have a well-defined object if you just 441 00:25:33,656 --> 00:25:35,306 copy the memory to somewhere else. 442 00:25:35,636 --> 00:25:39,946 The way that you move things in C++ is through a like a move constructor 443 00:25:39,946 --> 00:25:44,806 or a move assignment operator, where you by hand say how to create the 444 00:25:44,806 --> 00:25:47,596 object in the new location, which is equivalent to the old location. 445 00:25:47,926 --> 00:25:50,806 And that works totally fine with the stood list example, because you just. 446 00:25:51,151 --> 00:25:52,861 Update that end node to the right spot. 447 00:25:53,431 --> 00:25:57,031 But this mis mismatch turns out to be a really big deal when 448 00:25:57,031 --> 00:26:01,131 you want to interact with Rust, because everything in Rust is mem- 449 00:26:01,231 --> 00:26:05,021 Just copy the memory from one place to another, and if you put a C++ object in 450 00:26:05,021 --> 00:26:08,461 Rust and you just assign it to something else, now all of a sudden you cata 451 00:26:10,051 --> 00:26:11,821 catastrophically broke your objects. 452 00:26:12,751 --> 00:26:17,461 And so that, that's like a major problem. 453 00:26:17,461 --> 00:26:21,901 If you want to be able to talk about C++ objects when you're operating within Rust. 454 00:26:22,906 --> 00:26:23,386 Wow. 455 00:26:23,446 --> 00:26:25,876 Yeah, no, that makes a lot of sense. 456 00:26:25,876 --> 00:26:30,886 And it's a pretty gnarly problem because you've got essentially two 457 00:26:30,886 --> 00:26:36,446 worldviews and that, and I was thinking, oh, couldn't you just overload the 458 00:26:36,446 --> 00:26:42,506 copy trait for particular bits and pieces, but then not really no. 459 00:26:42,556 --> 00:26:43,366 That sounds tricky. 460 00:26:44,366 --> 00:26:47,516 It's very tricky because Rust gives you this guarantee. 461 00:26:47,906 --> 00:26:53,276 A bunch of generic code is out there, which will directly mem copy 462 00:26:53,276 --> 00:26:56,186 things around, and that's totally 463 00:26:56,186 --> 00:26:57,716 valid for every type that you give it. 464 00:26:58,766 --> 00:27:01,616 And so what do you do? 465 00:27:01,856 --> 00:27:04,096 So like the ways that people have been working around this 466 00:27:04,096 --> 00:27:06,066 issue one is through pinning. 467 00:27:06,696 --> 00:27:10,836 So if you pin and type in Rust or an object in Rust, you're 468 00:27:10,836 --> 00:27:15,096 basically saying this object cannot move from this piece of memory. 469 00:27:15,936 --> 00:27:17,946 So assignment and all this kind of stuff just doesn't work 470 00:27:17,946 --> 00:27:19,246 with that as you would expect. 471 00:27:19,306 --> 00:27:20,386 'cause it's just pinned in memory. 472 00:27:20,746 --> 00:27:25,456 And if you use a pinned interface, the drawback is it's 473 00:27:25,456 --> 00:27:27,526 very not ergonomic and Rust. 474 00:27:27,616 --> 00:27:31,096 It's like you work with a pinned object, you're just like, oh, great, now I'm 475 00:27:31,096 --> 00:27:34,966 gonna have a bad day because this is just not friendly, good rust code anymore. 476 00:27:35,456 --> 00:27:40,376 And so the CXX crate, which is one, one way to do C++ Rust 477 00:27:40,376 --> 00:27:42,306 in a wrap follows this method. 478 00:27:42,396 --> 00:27:44,466 And then you have to deal with that as a drawback. 479 00:27:45,006 --> 00:27:50,526 Another option, and this is what Zengar does, so Zengar, Z-N-G-U-R, 480 00:27:50,736 --> 00:27:52,446 it's not very well known yet. 481 00:27:52,446 --> 00:27:55,416 I'm trying to get more awareness to this c plus interop solution 'cause 482 00:27:55,416 --> 00:27:56,956 it's it's very good in many ways. 483 00:27:57,196 --> 00:28:01,126 But anyway, Zengar, what they do is they say, okay, we will. 484 00:28:01,456 --> 00:28:04,526 Just allocate every time we create a C++ object. 485 00:28:04,546 --> 00:28:06,856 So it always creates a C++ object on the heap. 486 00:28:07,756 --> 00:28:10,546 And that way when things are assigned, you're just copying 487 00:28:10,546 --> 00:28:12,326 that pointer, back and forth. 488 00:28:12,326 --> 00:28:14,216 But then you're paying a heap allocation, right? 489 00:28:14,216 --> 00:28:16,406 And that we lose our high performance there. 490 00:28:16,856 --> 00:28:17,366 At least. 491 00:28:17,366 --> 00:28:17,786 At least. 492 00:28:17,876 --> 00:28:18,056 Yeah. 493 00:28:18,086 --> 00:28:18,306 So 494 00:28:18,311 --> 00:28:22,801 just, and this may, this is probably obvious to most, but I'll just 495 00:28:22,801 --> 00:28:27,091 reiterate it for my own ignorance perhaps, but mem copy is a ridiculously 496 00:28:27,091 --> 00:28:31,291 trick, which I like sub, like single digit cycles time probably. 497 00:28:31,651 --> 00:28:37,171 Whereas, and so this is, a millionth of a second a nanosecond scale. 498 00:28:37,501 --> 00:28:44,451 And then a heap allocation, is anywhere, 10 to a hundred and 150, let's say. 499 00:28:44,501 --> 00:28:48,451 It's it's a lot of time if you're creating a lot of objects or if 500 00:28:48,451 --> 00:28:51,331 you're just doing assignment a lot like sort of updates and you have 501 00:28:51,331 --> 00:28:52,711 to pay that cost every single time. 502 00:28:53,131 --> 00:28:54,871 That would be annoying. 503 00:28:55,226 --> 00:28:57,451 Y Yeah. 504 00:28:57,511 --> 00:28:58,921 If I recall correctly, it's a thousand 505 00:28:58,921 --> 00:29:00,271 times slower. 506 00:29:00,511 --> 00:29:03,241 So it's a substantial thing, especially if you get a vector of these things. 507 00:29:03,721 --> 00:29:05,991 Now you pay that for, big ON, 508 00:29:06,291 --> 00:29:10,826 Yeah, although references I guess which are just pointers really can, obviously 509 00:29:10,826 --> 00:29:15,566 they're just the, in the rust world or we call this like the U size or the size 510 00:29:15,566 --> 00:29:18,446 of the bits of the SCPU architecture. 511 00:29:18,926 --> 00:29:21,146 They're instantaneous to copy around anyway. 512 00:29:21,146 --> 00:29:25,076 So this is a, one of the advantages of like plunking things on the heap. 513 00:29:25,436 --> 00:29:29,366 You pay, you amortize the cost of that allocation once because 514 00:29:29,366 --> 00:29:32,756 you can now mem copy as many times as you require later on. 515 00:29:33,756 --> 00:29:33,976 Yes. 516 00:29:34,461 --> 00:29:34,731 Yep. 517 00:29:36,591 --> 00:29:43,161 Um, so there was a new feature that got added to C++ 26. 518 00:29:43,231 --> 00:29:46,781 We're in 25, so it hasn't come out yet, but it's in the latest C++ draft, 519 00:29:47,611 --> 00:29:51,411 um, called Trivial Relocateability. 520 00:29:52,731 --> 00:29:56,811 And what this does is it identifies. 521 00:29:57,361 --> 00:30:03,981 The types in C++ to a rough approximation, that can be me copied around because 522 00:30:03,981 --> 00:30:08,661 it turns out that stood list example, like that's, that's an outlier. 523 00:30:08,781 --> 00:30:13,221 Most things like a Stu Vector, most classes that you would write, you 524 00:30:13,221 --> 00:30:16,761 could actually just move the memory to new location and it would work. 525 00:30:16,761 --> 00:30:17,031 Okay. 526 00:30:17,031 --> 00:30:20,701 It's technically undefined behavior, but it works because that's just how it is. 527 00:30:21,241 --> 00:30:27,211 So the idea behind Trivial Relocateability was to see if there's a way to make 528 00:30:27,211 --> 00:30:31,711 this carve out of types that support this and make that defined behavior. 529 00:30:32,881 --> 00:30:37,251 And so the thought was, oh, okay then maybe any type, which is trivially 530 00:30:37,251 --> 00:30:43,741 relocatable in C++ 26, we can put this on the rust stack and do stuff with it. 531 00:30:44,671 --> 00:30:45,091 If 532 00:30:45,266 --> 00:30:47,186 did note that this was in the past tense. 533 00:30:48,186 --> 00:30:48,576 Yeah. 534 00:30:49,536 --> 00:30:54,876 The problem was there's a new architecture that Apple's coming out with is called 535 00:30:54,876 --> 00:31:03,556 ARM 64 E, and it's a security a BI change primarily created for security purposes. 536 00:31:04,246 --> 00:31:09,806 And what they did is you, if you have a polymorphic type in C++, so something 537 00:31:09,806 --> 00:31:14,026 that inherits from something with a virtual function, um, that needs to have 538 00:31:14,026 --> 00:31:17,776 a V table in C++ to, to, in case you cast it to a base class, you need to 539 00:31:17,776 --> 00:31:19,486 point to the right function and so on. 540 00:31:20,056 --> 00:31:25,946 And so there's a pointer in every polymorphic object to the V table, right? 541 00:31:26,336 --> 00:31:28,286 And this pointer is transparent to users. 542 00:31:28,286 --> 00:31:30,266 You don't actually see that it's there or whatever. 543 00:31:30,686 --> 00:31:34,886 Me copying objects with a V table is totally fine up until. 544 00:31:35,616 --> 00:31:41,146 Arm 64 E. And what they did for security purposes is they said, look, 545 00:31:42,406 --> 00:31:47,656 the V table pointer is a great way for hackers to, to own your process 546 00:31:47,656 --> 00:31:51,836 because they can just take that pointer and point it to their own code. 547 00:31:51,866 --> 00:31:55,496 And next time you call a virtual function, you're going into the attacker's code. 548 00:31:56,096 --> 00:32:00,776 So they said what we're gonna do is we're gonna take the V table pointer and use 549 00:32:00,776 --> 00:32:03,986 some of the extra bits in it because nobody actually has two to the 64 memory. 550 00:32:03,986 --> 00:32:09,836 So they're gonna use these upper level bits and add some data in there that ties 551 00:32:09,986 --> 00:32:18,296 that V table pointer to the object, which contains the V table pointer, right? 552 00:32:18,836 --> 00:32:22,976 So that way if an attacker changes the VT pointer to something else, if it 553 00:32:22,976 --> 00:32:27,086 tries to execute that V table pointer, it'll be like, Nope, can't do that. 554 00:32:27,326 --> 00:32:28,646 You lost the signature, which 555 00:32:28,841 --> 00:32:34,851 So the CPU will essentially fault because the yeah we've, it, I'm 556 00:32:34,851 --> 00:32:36,411 just going to use naive link. 557 00:32:36,411 --> 00:32:40,731 Maybe this isn't, but it sounds a lot like a capabilities model where the original 558 00:32:40,881 --> 00:32:44,971 or some sort of providence back to the original memory address, let's say. 559 00:32:45,511 --> 00:32:50,401 And if you Yeah, change the destination, those parity bits won't 560 00:32:50,911 --> 00:32:55,621 work and therefore, or let some sort of checks sum, and then, oh. 561 00:32:57,321 --> 00:33:01,881 Your program will crash, which is probably better than having some un 562 00:33:01,911 --> 00:33:06,171 foreign code execute until you want to do 563 00:33:06,171 --> 00:33:09,051 something fun like interrupt with rust. 564 00:33:10,051 --> 00:33:10,381 Yep. 565 00:33:10,451 --> 00:33:11,041 Exactly. 566 00:33:11,621 --> 00:33:16,051 And so in the C++ world, they solve this problem by saying, okay instead 567 00:33:16,051 --> 00:33:19,051 of saying, you can mem copy all these types to move them around, 568 00:33:19,531 --> 00:33:24,331 you can call this trivially relocate function and that will do it for you. 569 00:33:24,781 --> 00:33:27,511 You trivially relocate, it does the mem copy and it does all the V 570 00:33:27,511 --> 00:33:32,201 table pointer fix ups, and that's all fine and dandy in the C++ world. 571 00:33:32,221 --> 00:33:36,691 But now you go back to the Rust interop, Rust isn't gonna do any 572 00:33:36,691 --> 00:33:38,311 vtable pointer fix ups for you, right? 573 00:33:38,311 --> 00:33:40,201 It's literally copying the memory. 574 00:33:40,841 --> 00:33:44,441 And so that kind of broke that idea. 575 00:33:44,951 --> 00:33:51,441 And so what we're trying to do now is add a new feature to C++, and 576 00:33:51,611 --> 00:33:53,821 this is primarily for Rust interop. 577 00:33:54,481 --> 00:34:00,691 And what this will do is instead of having the primitive copy thing from location 578 00:34:00,691 --> 00:34:04,711 A to location B, and then do the V two table pointer, fix up, it's going to 579 00:34:05,461 --> 00:34:07,471 get a pointer to where it used to live. 580 00:34:08,581 --> 00:34:12,631 And a pointer to the current location and do the fix up that way. 581 00:34:12,631 --> 00:34:16,411 So, essentially separating the movement of the memory from the point, the v 582 00:34:16,411 --> 00:34:19,741 table pointer, fix ups, if you have that. 583 00:34:20,821 --> 00:34:25,271 Then on the rust side, if you get an object, and this is 584 00:34:25,271 --> 00:34:26,621 hard to explain it, a podcast 585 00:34:26,856 --> 00:34:27,236 yeah. 586 00:34:27,401 --> 00:34:28,031 I'll do my best. 587 00:34:28,376 --> 00:34:29,186 It's not the best, 588 00:34:29,546 --> 00:34:29,816 not 589 00:34:29,876 --> 00:34:31,556 it's not the universal medium, let's say. 590 00:34:33,491 --> 00:34:34,871 Yes. 591 00:34:34,941 --> 00:34:40,581 You create your object on the, your c plus object on the rust stack. 592 00:34:40,701 --> 00:34:44,931 But you add a little bit at the end, which is enough for a pointer. 593 00:34:46,191 --> 00:34:49,611 And so when you initially create the object, you have all your data for 594 00:34:49,611 --> 00:34:53,481 the object and you have a pointer, which is just happens to be set 595 00:34:53,481 --> 00:34:55,431 to where the object's location is. 596 00:34:56,541 --> 00:34:59,451 Now on the rest side, you can assign, you can move things around. 597 00:34:59,451 --> 00:35:01,881 Now the object is in some other location somewhere else. 598 00:35:02,346 --> 00:35:07,566 And that pointer is still set to the original place that object lived. 599 00:35:08,166 --> 00:35:11,346 The data for that object is the data that we're what it used to be, but 600 00:35:11,346 --> 00:35:13,716 it's just in a new location now. 601 00:35:13,716 --> 00:35:18,156 When you call a c plus function, you need to make sure this is all fixed up. 602 00:35:19,296 --> 00:35:23,106 So when you call your intermediate C layer, which eventually calls your C 603 00:35:23,106 --> 00:35:27,576 function, that intermediate C layer will call that special new function that got 604 00:35:27,576 --> 00:35:31,236 added to the standard, which says, Hey, here's the pointer to the old location. 605 00:35:31,236 --> 00:35:32,766 Here's the data at the new location. 606 00:35:33,156 --> 00:35:36,216 It'll fix up all the pointers for you, and now it's valid. 607 00:35:36,246 --> 00:35:39,456 And define behavior to call that c plus function that you were calling. 608 00:35:40,521 --> 00:35:42,091 That sounds innovative. 609 00:35:42,301 --> 00:35:42,811 Well done. 610 00:35:43,231 --> 00:35:45,251 It's have you unlocked the puzzle? 611 00:35:45,461 --> 00:35:46,781 Have you solved the puzzle, do you think? 612 00:35:46,841 --> 00:35:47,831 Or at least 613 00:35:48,831 --> 00:35:49,551 I, think. 614 00:35:50,436 --> 00:35:50,826 okay. 615 00:35:50,841 --> 00:35:53,941 we may have, there are still some open questions on how to deal 616 00:35:53,941 --> 00:35:58,931 with this on the rust side because that works fine and dandy for, uh, 617 00:35:58,961 --> 00:36:01,011 C++ objects you create in Rust. 618 00:36:01,321 --> 00:36:05,171 But what if C++ gives you an object or a reference to one of those objects? 619 00:36:05,171 --> 00:36:06,731 Now that one doesn't have the little 620 00:36:06,791 --> 00:36:10,451 space at the end with the pointer, so that might be a different type. 621 00:36:10,496 --> 00:36:12,836 You might want to convert it to one with the pointer thing at the end. 622 00:36:12,836 --> 00:36:13,286 I don't know. 623 00:36:13,286 --> 00:36:14,699 There, there's some, some thought 624 00:36:14,699 --> 00:36:22,398 So if I was a, although but yeah, no that, that seems like a decent enough 625 00:36:22,398 --> 00:36:28,768 chunk of the problem to be a good step forward because I presume that if I was 626 00:36:28,768 --> 00:36:34,708 importing a C library, sorry, c plus library from rust, most of the time I 627 00:36:34,708 --> 00:36:40,708 would be initializing the objects rather than essentially receiving an object 628 00:36:40,708 --> 00:36:43,618 that was created in the library itself. 629 00:36:43,768 --> 00:36:45,718 So no, that's really interesting. 630 00:36:46,168 --> 00:36:48,238 How does it change like that happen? 631 00:36:49,048 --> 00:36:49,918 It must be. 632 00:36:51,173 --> 00:36:53,543 This is a little bit about how the committee works, but there 633 00:36:53,543 --> 00:36:59,353 must be years of deliberation for things like this to make progress. 634 00:37:00,353 --> 00:37:07,343 With the committee there's a lot of formalisms and structure and all these 635 00:37:07,343 --> 00:37:11,083 kinds of things, people frequently get defeated by it because it's just like 636 00:37:11,113 --> 00:37:12,463 too overwhelming, too complicated. 637 00:37:12,463 --> 00:37:15,763 But if you understand it, there are tricks that you can do. 638 00:37:16,303 --> 00:37:18,863 So here's the trick that we're gonna do or we're gonna attempt, 639 00:37:18,863 --> 00:37:20,123 we'll see if it actually works. 640 00:37:20,153 --> 00:37:23,763 So right now, no new features are allowed for C++ 26, 641 00:37:24,173 --> 00:37:24,443 right? 642 00:37:24,443 --> 00:37:25,733 It's already been stamped. 643 00:37:25,943 --> 00:37:27,143 It's going out to the national 644 00:37:27,143 --> 00:37:29,393 bodies for ballot at this point. 645 00:37:29,793 --> 00:37:32,613 And now we deal with what's called National body comments. 646 00:37:33,543 --> 00:37:36,753 So every national body can say, okay, we released this, we got 647 00:37:36,753 --> 00:37:39,813 this spec, and we have a comment about this, a comment about this. 648 00:37:39,843 --> 00:37:42,603 And these are usually comments about bug fixes. 649 00:37:42,603 --> 00:37:45,093 You can never say add a new feature with a comment. 650 00:37:45,903 --> 00:37:49,793 And and this is a new feature in c plus 26 is trivially relocatable thing. 651 00:37:50,363 --> 00:37:53,608 And so what I'm planning on doing is making a national body 652 00:37:53,608 --> 00:37:58,708 comment for the us, which says that we got our primitive wrong. 653 00:37:59,248 --> 00:38:04,048 The trivially relocate function is not actually the simplest way to express 654 00:38:04,048 --> 00:38:08,348 this primitive like basis operation for trivially relocatable types. 655 00:38:08,918 --> 00:38:11,558 It's this other thing, which is called start lifetime. 656 00:38:11,558 --> 00:38:18,848 At and through this bug fixing process, I'm hopeful we'll be able 657 00:38:18,848 --> 00:38:20,918 to say, oh yeah, that was a mistake. 658 00:38:21,188 --> 00:38:25,688 And so before se LS plus 26 ships, we're going to have this new basis 659 00:38:25,688 --> 00:38:30,158 operation and that will unlock this rust interop thing, but also solves 660 00:38:30,158 --> 00:38:32,678 another problem that people had with the feature, which was that 661 00:38:32,978 --> 00:38:36,248 it didn't work with re alec where you basically have an existing buffer 662 00:38:36,248 --> 00:38:39,038 that you're expanding and sometimes it can expand at the same place. 663 00:38:39,398 --> 00:38:44,178 So I'm hopeful that we can get this to happen, but yeah we're using this national 664 00:38:44,178 --> 00:38:46,068 body comment process to, to slip it in. 665 00:38:46,308 --> 00:38:48,533 Otherwise we're gonna have to wait for C++ 29. 666 00:38:50,013 --> 00:38:50,883 I am curious. 667 00:38:50,883 --> 00:38:51,483 I'll watch. 668 00:38:51,513 --> 00:38:54,723 I think it'll be fascinating because especially it's almost, there's always 669 00:38:54,723 --> 00:39:00,843 a meta discussion as well, but the committee processes are challenging. 670 00:39:00,873 --> 00:39:05,763 A lot of people don't really have, the stamina or it, 'cause everyone 671 00:39:05,763 --> 00:39:09,063 is so emotionally involved because they wouldn't be there if they 672 00:39:09,063 --> 00:39:13,483 didn't think that things could move forward or improve. 673 00:39:14,173 --> 00:39:16,843 And they've got this great idea that they've been working on for years. 674 00:39:16,843 --> 00:39:21,133 They have a working implementation before they even start proposing it. 675 00:39:22,123 --> 00:39:23,743 And now you're telling me you gotta, I gotta wait. 676 00:39:23,743 --> 00:39:26,203 Like I gotta get, it's, yeah. 677 00:39:26,203 --> 00:39:28,208 I can just imagine that if I was. 678 00:39:28,733 --> 00:39:32,243 Doing this out of love or something that there's no way I would 679 00:39:32,243 --> 00:39:36,593 be able to last a five to 10 years to get new things through. 680 00:39:36,623 --> 00:39:41,453 I was looking at one time, or I think it was concepts, how there were 681 00:39:41,453 --> 00:39:47,063 multiple implementations and different ideas about the way that could work. 682 00:39:47,063 --> 00:39:53,483 And but then again for something as impactful as making a change to one 683 00:39:53,483 --> 00:39:56,663 of the most fundamental programming languages in the world maybe 684 00:39:56,663 --> 00:39:59,093 waiting five years isn't very long. 685 00:39:59,098 --> 00:40:03,898 Like, I mean, uh, C++ is, um, is everywhere. 686 00:40:04,133 --> 00:40:07,703 And so it's extremely fundamental and it's extremely important. 687 00:40:07,703 --> 00:40:12,653 The stakes are very, very high, and so it makes sense to take things slow. 688 00:40:15,398 --> 00:40:16,253 I, I think so. 689 00:40:16,253 --> 00:40:20,603 And I think that's one of its particular charisms is that. 690 00:40:21,038 --> 00:40:24,698 It is generally a slow moving language, not as slow as C 691 00:40:24,818 --> 00:40:26,228 although C is speeding up. 692 00:40:26,468 --> 00:40:28,928 But you only can take on so much new stuff at a time. 693 00:40:28,928 --> 00:40:34,298 And I think actually in my opinion, c plus has grown too fast for its own good. 694 00:40:34,898 --> 00:40:37,628 We've hit some kind of tipping point to where the complexity is more of 695 00:40:37,628 --> 00:40:39,488 a problem than it was a solution. 696 00:40:40,008 --> 00:40:42,018 And I don't know how to reverse that 697 00:40:42,128 --> 00:40:45,688 I think may maybe, human bias or something, but it's much easier to add 698 00:40:45,688 --> 00:40:50,628 an extra feature, like an extra box to the checklist or like the feature list. 699 00:40:50,838 --> 00:40:56,103 But, taking things away is really is much harder. 700 00:40:57,043 --> 00:40:59,133 Just imagine saying, actually we'll cut that out. 701 00:40:59,433 --> 00:41:00,693 Is essentially impossible. 702 00:41:00,793 --> 00:41:05,823 More generally than C++, but it's like any sort of product, for example, software 703 00:41:05,823 --> 00:41:11,373 things, software products in general, I think have this tendency to grow until 704 00:41:11,373 --> 00:41:12,468 they collapse under their own weight. 705 00:41:13,468 --> 00:41:14,518 Yeah, I think so. 706 00:41:14,578 --> 00:41:18,128 Also if you look at it from like a company perspective let's say 707 00:41:18,128 --> 00:41:20,078 you've got a huge code base. 708 00:41:20,078 --> 00:41:22,913 I think the last quote I got from Photoshop was like 79 709 00:41:22,913 --> 00:41:24,008 million lines of code, right? 710 00:41:24,008 --> 00:41:28,238 So we're talking a lot of code and somebody from a standardization body 711 00:41:28,238 --> 00:41:33,718 says, Hey, you need to change, a million lines of your code because we 712 00:41:33,718 --> 00:41:34,978 found a better way to do something. 713 00:41:36,298 --> 00:41:40,228 And the response is does that add any value to my product? 714 00:41:40,473 --> 00:41:41,533 Do the customers care? 715 00:41:42,003 --> 00:41:42,663 no, not really. 716 00:41:43,943 --> 00:41:44,233 Yeah. 717 00:41:45,578 --> 00:41:48,878 Okay, so somebody up in the ivory tower has decided there's a better way to 718 00:41:48,878 --> 00:41:52,868 do something and now I have to pay a buttload of money for, that's not gonna 719 00:41:52,868 --> 00:41:54,368 actually have any tangible benefit. 720 00:41:54,938 --> 00:41:55,388 Nah. 721 00:41:55,988 --> 00:41:57,698 Rather keep the existing code. 722 00:41:57,698 --> 00:42:00,058 You wanna do something new with the new codewe right? 723 00:42:00,248 --> 00:42:00,788 Great. 724 00:42:00,788 --> 00:42:05,768 But like that old code, if it's working and it's doing its job like please don't 725 00:42:05,768 --> 00:42:06,968 make me have to do anything with it. 726 00:42:06,968 --> 00:42:07,178 'cause the 727 00:42:07,178 --> 00:42:08,198 scale is too big. 728 00:42:08,508 --> 00:42:09,828 It is just way too much. 729 00:42:11,718 --> 00:42:20,168 Are there any lessons for individuals that you think if, who are navigating, while we 730 00:42:20,168 --> 00:42:25,768 come we've talked about Rust language, C++ language, you know, companies, but what 731 00:42:25,768 --> 00:42:29,428 about, uh, people and their own careers? 732 00:42:29,958 --> 00:42:31,818 Yeah, I think that's a really good question. 733 00:42:31,818 --> 00:42:36,908 I think I can speak in particular for C++ developers. 734 00:42:37,458 --> 00:42:40,508 For someone who's a C++ developer, they're an expert in it. 735 00:42:40,568 --> 00:42:42,398 You know, they feel very powerful in that language. 736 00:42:42,398 --> 00:42:45,638 And the idea of going to any other language, you'd feel very weakened, right? 737 00:42:45,638 --> 00:42:47,198 Because you're just so used to this language. 738 00:42:47,598 --> 00:42:51,408 If you're a c plus developer you have a choice, right? 739 00:42:51,678 --> 00:42:56,118 You will definitely, no matter what stage of your career, be able to program 740 00:42:56,118 --> 00:42:57,408 c plus for the rest of your career. 741 00:42:57,658 --> 00:42:59,188 This is not going away. 742 00:42:59,548 --> 00:42:59,698 There. 743 00:42:59,818 --> 00:43:00,928 Jobs are gonna be there. 744 00:43:01,408 --> 00:43:06,688 But it will be with a legacy language, especially towards the end of your career, 745 00:43:06,688 --> 00:43:10,618 it's gonna be like, there are some FORTRAN people today and they do their thing. 746 00:43:10,618 --> 00:43:11,698 There are some COBOL people. 747 00:43:11,818 --> 00:43:16,318 I don't think it'll be that bad in 20 years, but it'll be in that space, right? 748 00:43:16,918 --> 00:43:19,738 But just be okay with that if that's what you want to do. 749 00:43:20,458 --> 00:43:22,918 If you're looking, if you wanna be. 750 00:43:24,073 --> 00:43:28,603 Where the innovation is happening, then you're gonna have to take a 751 00:43:28,603 --> 00:43:31,783 different approach and you're gonna have to look at things like Rust 752 00:43:31,843 --> 00:43:34,603 and Rust in particular right now, I think is really important to look at. 753 00:43:35,083 --> 00:43:36,253 And don't be threatened by it. 754 00:43:36,703 --> 00:43:42,463 See, I, something that's not talked about a lot is that rust is an outgrowth of C++. 755 00:43:42,933 --> 00:43:45,133 C++ is almost its granddaddy. 756 00:43:45,133 --> 00:43:47,053 People hate it when I say that, but it's true. 757 00:43:47,503 --> 00:43:52,153 So many lessons that were learned from C++ were incorporated in a rust, 758 00:43:52,153 --> 00:43:55,153 whether it's acknowledged or not, it's just obvious in the design. 759 00:43:55,663 --> 00:44:01,033 And if you look at it like that, it's, you can be proud of it, right? 760 00:44:01,113 --> 00:44:01,583 You're a C++. 761 00:44:02,023 --> 00:44:03,373 Look at all the stuff that they learned from C++. 762 00:44:03,783 --> 00:44:04,513 That's great. 763 00:44:04,513 --> 00:44:07,943 All the stuff that I did to help grow C++ or, or build C++ 764 00:44:08,323 --> 00:44:10,873 knowledge in the community of people that I'm, I'm, I'm with. 765 00:44:11,503 --> 00:44:13,483 This is a consequence of that. 766 00:44:13,483 --> 00:44:14,504 It's a benefit of it. 767 00:44:15,223 --> 00:44:17,053 And if you look at it like that, it's not as threatening. 768 00:44:17,053 --> 00:44:18,373 It's not like an us versus them. 769 00:44:18,373 --> 00:44:19,153 It's more like. 770 00:44:19,528 --> 00:44:20,548 You know, good for them. 771 00:44:20,668 --> 00:44:24,018 You know, like, let me look into this because this is the next step in C++'s 772 00:44:24,508 --> 00:44:28,948 growth is actually Rust it's the, you wouldn't say next step in C++'s growth, 773 00:44:28,948 --> 00:44:33,508 but next step in growth for a systems engineer is using the latest tech. 774 00:44:33,888 --> 00:44:35,448 And the same is true for AI, right? 775 00:44:35,448 --> 00:44:38,538 It's like another one of these things is it's gonna be here, right? 776 00:44:38,538 --> 00:44:41,298 You're not gonna change it and make it go away. 777 00:44:41,748 --> 00:44:45,928 But you can be part of the future and what it's going to bring. 778 00:44:45,928 --> 00:44:50,098 And so I would recommend to people to be open-minded about it. 779 00:44:50,198 --> 00:44:52,178 Check it out, try it in a real project. 780 00:44:52,398 --> 00:44:55,548 You want to seek that feeling that so many people report that. 781 00:44:55,548 --> 00:44:59,388 Oh, I went and I used rust and I really don't wanna use c plus anymore after that. 782 00:44:59,778 --> 00:45:01,458 You wanna seek that experience, right? 783 00:45:01,458 --> 00:45:03,798 Because then you can get it and you can see where they're coming from 784 00:45:03,858 --> 00:45:05,438 and then, see where that takes you. 785 00:45:05,513 --> 00:45:07,463 But you gotta be open-minded, right? 786 00:45:07,463 --> 00:45:10,693 It's just gonna, you're gonna have to try out some things and get past some 787 00:45:10,693 --> 00:45:13,093 biases and just say like, all right, I'm gonna, I'm gonna give it a whirl. 788 00:45:13,458 --> 00:45:16,193 I think of Rust as a bit of a spiritual successor to C++. 789 00:45:17,463 --> 00:45:25,628 The one funny thing actually is and Rust has learned so much from systems 790 00:45:25,628 --> 00:45:32,838 programming experience over over several decades, and it's as though we have 791 00:45:32,838 --> 00:45:37,688 taken the lessons around initialization and destruction and just encoded that 792 00:45:37,688 --> 00:45:40,298 in the type system or into the compiler. 793 00:45:40,688 --> 00:45:44,558 There is this liberating feeling, knowing that you are never going 794 00:45:44,558 --> 00:45:50,858 to need to worry about, I don't know, deleting an object twice. 795 00:45:52,178 --> 00:45:57,638 It's just, oh, that's not a, that's not a concern that I need to have perhaps. 796 00:45:57,638 --> 00:46:00,228 But another thing is maybe they front load the concern because you know 797 00:46:00,228 --> 00:46:04,548 your design needs to be correct, but also, isn't it so much better that 798 00:46:04,548 --> 00:46:09,858 you get told that there's a problem early before you get told by your users 799 00:46:09,858 --> 00:46:13,338 that there's a problem or worse they don't tell you 'cause they don't know. 800 00:46:13,338 --> 00:46:17,958 And they've got introduced some sort of horrendous bug by the way that a lot of 801 00:46:17,958 --> 00:46:20,958 that stuff around C++ was accidental. 802 00:46:22,053 --> 00:46:25,543 I don't, I am not gonna be able to find the reference, but grad whore 803 00:46:25,543 --> 00:46:31,543 is was talking I think it might have been on Twitter that or saying that 804 00:46:31,543 --> 00:46:33,343 was not the original intent with Rust. 805 00:46:33,973 --> 00:46:38,293 The original intent was something much more closer to go where you 806 00:46:38,353 --> 00:46:44,383 or I guess much more with a much greater focus on concurrency. 807 00:46:45,463 --> 00:46:51,353 But then when it was adopted by Mozilla, the Firefox group really 808 00:46:51,353 --> 00:46:57,723 wanted to essentially be zero cost abstractions all the way down and had 809 00:46:57,723 --> 00:47:04,743 this very heavy, practical focus on being as lean at runtime as possible. 810 00:47:05,373 --> 00:47:07,323 Because if Rust was to ever compete. 811 00:47:08,268 --> 00:47:11,228 It needed to be able it needed to shed. 812 00:47:11,228 --> 00:47:16,288 Its its runtime, which it had at one, very pre 1.0 like it's now over a decade ago. 813 00:47:16,348 --> 00:47:21,988 But it's a slight accident of history, but also really positive that the 814 00:47:21,988 --> 00:47:24,718 language was incubated in production. 815 00:47:26,548 --> 00:47:30,298 The idea that we could, we were creating a programming language to 816 00:47:30,298 --> 00:47:38,068 be able to work in the context of a c plus program to enable multi-core. 817 00:47:38,428 --> 00:47:41,968 And that was a very, like a multimillion lines of code 818 00:47:41,968 --> 00:47:45,698 base with built by many teams. 819 00:47:45,698 --> 00:47:52,283 And so all of those features were like part of, it's, part of it 820 00:47:52,288 --> 00:47:56,778 how, part of how it grew up out of a hobby project originally. 821 00:47:57,858 --> 00:48:03,318 I think it would be like, this has just been such a really nice arc all the way. 822 00:48:03,438 --> 00:48:05,868 And thank you so much for sharing your thoughts. 823 00:48:05,868 --> 00:48:08,928 Has there been anything else that's popped into your mind as we've 824 00:48:08,928 --> 00:48:13,398 been chatting that you're like, oh, I should tell people about that? 825 00:48:14,398 --> 00:48:18,913 I guess probably the biggest thing is check out Zengar for doing c plus interop. 826 00:48:19,303 --> 00:48:25,603 The, it's way less known than it deserves as a c plus interop layer. 827 00:48:25,603 --> 00:48:26,683 Like very interesting. 828 00:48:27,103 --> 00:48:28,123 That's probably the biggest thing that 829 00:48:28,233 --> 00:48:28,523 Okay. 830 00:48:28,728 --> 00:48:29,208 Fantastic. 831 00:48:29,238 --> 00:48:32,598 I am so looking forward to your talk in New Zealand next month. 832 00:48:32,598 --> 00:48:36,958 I am just, I feel so incredibly honored that you have decided that you accepted 833 00:48:36,958 --> 00:48:39,718 our offer to, to come down here. 834 00:48:39,928 --> 00:48:41,488 And I hope that you have a very. 835 00:48:42,133 --> 00:48:44,473 Enjoyable experience in the country. 836 00:48:44,803 --> 00:48:50,173 Until then, I hope that you also actually enjoy the process of preparing for a talk. 837 00:48:50,243 --> 00:48:54,713 I know that you'll have a very receptive audience, so thank you both for sharing 838 00:48:54,713 --> 00:48:56,943 your time and sharing your expertise. 839 00:48:57,033 --> 00:48:57,663 So thanks, David. 840 00:48:58,663 --> 00:48:59,533 Oh, thank you so much. 841 00:48:59,533 --> 00:49:04,903 I am so psyched about the conference, so I can't wait to be there.