Skip to content
PodSubs
ShowsPeopleGuestsJobsNewsEvents
More
BlogPodCred
Menu
ShowsPeopleGuestsJobsNewsEventsBlogPodCredNetworksSign in
Join nowSign in
  • Shows
  • People
  • Jobs
  • News
  • More
    • Guests
    • Events
    • Blog
    • PodCred
    • Networks
    Join nowSign in

ADSP: Algorithms + Data Structures = Programs

ADSP: Algorithms + Data Structures = Programs artwork

Episode 18: Special Guest Sean Parent! (Part 2)

Mar 26, 202137 minAudio

In this episode, we finish part two of our interview with Sean Parent! About the Guest: Sean Parent is a principal scientist and software architect for Adobe Photoshop. Sean has been at Adobe since 1993 when he joined as a senior engineer working on Photoshop and later managed Adobe’s Software Technology Lab. In 2009 Sean spent a year at Google working on Chrome OS before returning to Adobe. From 1988 through 1993 Sean worked at Apple, where he was part of the system software team that developed the technologies allowing Apple’s successful transition to PowerPC. Date Recorded: 2021-03-18 Date Released: 2021-03-26 FREE 1/2 Day APL Beginner Conference on March 31, 2021 Objective-C Automatic Reference Counting Objective C++ C++ std::move Trivially Relocatable versus Destructive Movable P1144 - Object relocation in terms of move plus destroy Nico Josuttis’ book C++ Move Semantics Jon Lakos’ latest book Large-Scale C++ Volume I ASL - Adobe Source Libraries Andrei Alexandrescu’s library Loki Boost.move by Howard Hinnant and Dave Abrahams STLab on Github C++ std::pair A9 Lecture that mentions Stepanov & Scheme Channel 9: E2E: Herb Sutter and Erik Meijer - Perspectives on C++ Stepanov Papers STL SGI Implementation and Docs 2018 Generic Programming Elements of Programming Intro Song Info Miss You by Sarah Jansen https://soundcloud.com/sarahjansenmusic Creative Commons - Attribution 3.0 Unported - CC BY 3.0 Free Download / Stream: http://bit.ly/l-miss-you Music promoted by Audio Library https://youtu.be/iYYxnasvfx8

Audio and video play straight from the show's own hosting. PodSubs stores and serves no media.

Transcript

From the show. Tap a time to play from there.

Sean ParentA lot of the the characters in in APL were chosen because of how they look visually. And Iota.

ConorWelcome to ADMP the Podcast, episode 18, recorded on March 18th, 2021. My name is Connor, and today with my co-host Bryce, we wrap up part two of our conversation with Sean Perry. Be sure to check out part one before listening to this one.

BryceDo you think that C should have adopted a destructive move model where the object left behind was not in a valid state but was gone?

Sean ParentIdeally, yes. Nobody could come up with a way to make it work. One of the problems with it is it is it changes the uh uh the strict ordering of destruction rules that we have in C ⁇ , which is somewhat problematic.

Sean ParentUm if anybody's used uh uh Objective C, which mostly on Apple platforms the the Objective C language gets used. A lot of people don't know, but if you just stick an an extra M on the extension for your file, so you have files that are type .mm instead of just dot M, uh then you get a language called Objective C, which is basically all of C with the Objective C constructs added in.

Sean ParentAnd Objective C in recent versions uh has a feature called ARC, which is automatic reference counting. So for a while, Objective-C had a garbage collector, and then Apple kind of backed away from the garbage collector and decided to go with automatic reference counting. Uh so Objective-C objects are uh uh uh you can think of the the the pointer that you get to them as being very similar to a shared pointer, except the language manages the shared pointer as opposed to you managing it manually.

Sean ParentBut one of the unique features of arc objects is is they have uh um uh more precise lifetime rules uh than C ⁇ objects would. So if I have a uh uh an ARC object, it is uh destructed at the end of the last sequence point where it is last used. So it is uh that allows for a destructive move of of of archetyp, um uh which saves you additional you know increment decrements on your on your F counts and it happens automatically without ever having to put in an explicit move.

Sean ParentAnd so, you know, one of the biggest problems that we have with with C ⁇ is that explicit move is is a is a cast circumventing the type system, and that makes the move operation unsafe. Um you end up with this shell of an object, and you have to take care in how you deal with that that object with a with an unspecified value.

BryceRight. Yeah, it it it's it's it's a odd quirk of the function called std colon colon move, that it does not actually move, it just casts it to a type that then the thing the f whatever it gets passed to could move from it.

Sean ParentCould move, right, right, right. And within the language, if you just have an implicit move, then the only thing the language is going to do after the move is destruct it. And so that makes the move safe. But when you circumvent the type system and you cast cast with std move, um uh uh uh now you've got got an unsafe operation. And there's only a few places in the language where std move is required, right, um uh to get a move.

Sean ParentAnd if you had the rule that that you could somehow decorate a type and say, you know, this type is d is is uh uh uh uh uh supports a destructive move, um uh then you could get rid of those those cases and you could just say this thing will automatically move on the you know on the last use, or the compiler's allowed to automatically move it on the on the last use. And that would make those cases safe because now your object would be out of scope and you couldn't reference it after that.

Sean ParentUm but it would be a significant change to the language.

BryceAnd it was something that I believe was considered in the you know original design of Move, but um uh the destructive model, they they were not able to work it out in the C11 era.

Sean ParentYeah. Yeah, there were a lot of ideas in that time frame. I think post then, I think a number of people have a lot of you know newer ideas and kind of a newer take on it. Um one of the best proposals I've seen float around is the idea of uh uh uh uh just having something be trivially movable. And it turns out most types are trivially movable. And what that means is it is if you just mem copied them and then didn't do anything with the hole they left, that would be sufficient.

BryceIs that is that Arthur O'Dwyer's paper you're thinking of?

Sean ParentUh I am horrible with names. So I believe it is, yeah.

ConorSo I think it's Arthur O'Dwyer's paper.

Sean ParentYeah, so so you know, I think if we had a notion of trivial trivially movable and the notion of allowing the compiler to then say, if my type is trivially movable, then I can can move it on last use. Um then you get this automatic move on last use, and you don't have to cast. And it's safe because if then if then you add another use at post that point, then that becomes the last use. And and you know, you don't get a any kind of compiler error that says, hey, you used this after it fell out of scope.

Sean ParentIt did fall out of scope, but now you just extended the scope, and what was a previous move safely became a copy, and and now the last use becomes a move. Um uh uh and so you know it's less for the user to type and it's easier to get correct, and it's less error prone if if we could get to that state.

ConorSo what uh what paper or what mailing are you aiming to for your paper to be coming out in?

Sean ParentUh, you know, I'm trying to rush through at least getting a uh an R0 out because um uh uh this came about because in in in John Lacos' upcoming book, he's got these sections called annoyances, and he reached out to me and said, you know, I've got I know you've got this annoyance about move. Uh could you write a couple pages on the annoyance of move? And so so we just kind of finished up a couple rounds of editing.

Sean ParentIt was I just wrote it right like right after ACCU wrapped up. And and part of that is is I want to cite a paper um with a proposal to fix this annoyance. And I had started about a year ago on actual wording for how I would go about fixing this annoyance, and I had a round of conversations at that time with uh with with Nico, who was r at the time what kind of set it all off was he was writing his book on move and he had on on R-value semantics and he had reached out to me to get some clarifications on some things.

Sean ParentSo we had a conversation about that, and that led to my complaint, and we kind of started on some wording, and I roped in Howard Hennant and and Herb Sutter and and got some you know additional feedback, and so I just kind of had this this bunch of scraps of notes um uh for something that could have been a proposal. I was hoping that uh an an employee who recently left Adobe was going to, and he had joined the Standards Committee, was going to kind of pick it up and see it through. Uh but he recently left, so and then this whole thing with John came up, so now I'm resurrecting it.

Sean ParentSo I'm gonna try to get a get a quick uh uh uh uh number for it and get a paper, you know, into whatever the next mailing is that would just basically say, you know, it's gonna outline the problem with some potential proposed wording. Um uh but uh uh the whole fact that, you know, over time the you know the the requirements have been weakened so much in in the standard makes it a little complicated.

Sean ParentBut I think that we can we can cover it with just kind of an upfront requirement, which we that that's how we handle the move semantics already, that basically says, you know, there's a precondition that for all of these functions that they have this notion of a domain of operation. And and and we can't re-associate that domain of operation with specific semantics because that would then be be a breaking change because it would be strengthening the semantics of of the existing library.

Sean ParentUm but we can, I think, introduce the notion of the domain of operation and then simplify the wording in a whole bunch of the algorithms where we don't have to say, you know, you know, the the the equal equal has to apply in all of these cases because that's just globally covered by this notion of a domain of operation. Um so for for a lot of things that would then then simplify the wording, you know, and then then we get into the the wording of how uh how we make the exceptions uh for uh uh you know but uh uh for the post conditions of a move from object, which basically says it gets uh pulled out of the domain of operation for everything but destruction and assignment two.

Sean ParentAnd then there's a little complication in there, which is uh uh uh if you swap an an uh uh if you s if you swap, uh if you do a self-swap, so you call swap a to a. In the middle of swap a to a it's going to do uh move from a move from object to itself. And the standard indirectly guarantees that swap AA uh uh uh uh is valid and should be a no-op.

Sean ParentUm uh uh and so that adds an additional wrinkle, which is how do you specify that you can uh you can move that you can you know move from an object that's been moved from into itself and get that into the spec. So I've got I've got kind of two options for proposed wordings on that. One of the proposed wordings gets gets a bit complicated and to satisfy the other conditions of of move, it it requires you to introduce a a fixed point, um uh which the standard hasn't done done anyplace else, but it kind of wraps everything up neatly, even if it's a little complex.

Sean ParentUh the other approach is to just say no, we don't support uh uh swap AA, which would be a breaking change, but most people on the standards committee um uh agree that anytime in the wild they've actually seen a self-swap, it's been a bug anyways.

ConorSo this sounds like if I if I recall correctly, at one point you told me, Sean, that uh you don't identify as a C committee uh standards member. You uh you identify as someone who sometimes drives by the committee and lobs a Molotov cocktail into it just to cause to cause discussion. Uh do you are you are you promoting yourself now to be a committee member or or are you do you still identify with the last one?

BryceI can confirm this is a Molotov cocktail. Yeah.

Sean ParentYeah. I think my actual quote is a hand grenade with Molotov cocktail works too, I guess. Hand grenade, okay. Yeah, so lobbying hand grenades uh, you know, in into the committee. Um uh uh yeah, uh uh I think this is is a little of both. Um uh you know, going through finding the the kind of you know weakening of the semantics in the standard library has kind of reminded me of of of why I get so frustrated working uh with the standards committee.

Sean ParentUm this has been my you know the the the post conditions of move since you know ASL, if you go back, it had uh uh a move library, and I think it was one of the first, it was not the first move library, but I think it was the first move library that actually worked. Um uh uh there was a library that um uh uh Andre Alexandrescu wrote that was uh I think it was called called Loki or Loki.

Sean ParentUm that introduced the notion of move. It was a fairly complex library, and in most cases it didn't actually work, meaning that it would fail to generate uh a move. And then uh Howard Hinnant and Dave Abrahams wrote another uh move library that they put out that uh kind of worked but failed in a in a bunch of cases. That was Boost Move? That was Boost Move.

ConorAnd um ASL for those that don't know, I believe is the Adobe standard library, correct? Adobe Source Library. Or source library.

Sean ParentYeah. And uh they're they're still kicking around, still available on uh if you go to uh uh github.com slash uh stlab is uh is is where they are. And we've got a website, stlab.cc, which I think is a nice domain name for uh for the libraries.

BryceAnd the idea of all these of all libraries of this class was before C11, when move was a language feature, there were a bunch of attempts to emulate move in a library uh you know, partially with macros, partially with other things, to varying degrees of success.

Sean ParentTo varying degrees of success, yeah. And so so ASL uh uh uh uh uh implemented move which relied on on RVO, which is return value optimization. And uh and it turns out at that time I validated a whole bunch of compilers. Um every compiler, if you put it in optimization mode, uh uh uh uh uh uh uh implemented uh uh named return value optimization, which was allowed by the standard at the time, but not required.

Sean ParentI don't think it's actually required till 17, C17. Um so even going way back to you know C03, everybody implemented it for release builds. And so uh uh in ASL we had a move library that looks very much like the move library that was that that's standardized in that you know it was it was ST Lab move was the way you invoked move. And and that got you know a lot of commercial use and actually worked very well to good effect.

Sean ParentIn fact, the last standards committee meeting I actually attended um uh uh the reason why I attended was to to pitch a library solution instead of all the complexity that came in with with R-value references. And and with you know, and I demonstrated that if we just required uh uh uh N R VO, which we do now, and all the compilers already implemented it, uh that we could get you know more than 90% the benefit of of move semantics.

Sean ParentThere are some cases that you that you can't do that way, um uh but the vast majority of cases w you can. Um so we could get get the the the we could handle the vast majority of the cases uh without introducing a new language construct of R value references, which are kind of weird and all the complexity that that came out of that. And and it was interesting because the the counter-argument was largely that without R-value references we didn't get uh forwarding references.

Sean ParentAnd and at that time, R-value references and forwarding references were the same thing. And later forwarding references became something else with the same syntax. So had we known that at the time, you know, maybe we could have gone with a library sub with a library solution for for moving and had a separate construct for forwarding.

BryceWould we still have needed forwarding references in a world where we didn't have our value references?

Sean ParentUh uh if you want kind of the the the the perfect forwarding, yeah, you do. Because you need to be able to know, you know, is this thing a move or a copy and carry that down through through your levels.

ConorSo so do you remember what uh committee meeting that was at, the last one you went to?

Sean ParentI don't remember the dates. It was in Bellevue, Washington. Um and basically in it was interesting because in I gave a presentation and I gave a demonstration of the ASL Move Library, and I I you know I hadn't written a paper on it. I just you know come in at the request of a committee member to kind of show it. And and there was a you know, it was a a small group that was actually worried about R values, and we had a straw vote, and the straw vote vo vote there was to yes, we should abandon R value references and go with a library solution.

Sean ParentAnd then largely because of political reasons in the um in the larger meeting, um uh that got overturned. And so so, which for me was an interesting lesson, which is, you know, convincing the people who matter isn't necessarily enough. Right, right, you know, convincing the people who are actually working with the stuff. And and it wasn't a hundred percent, uh uh, you know, you know, even in the small room, but it was it was was in favor.

Sean ParentAnd um uh yeah, yeah.

BryceThat would have been 2008, by the way. 2008.

Sean Parent2008, yeah. So that um uh so we kind of lost in the in the the in the in the larger committee, and that's when when our value references got got voted. I think the terminology is voted out of the standard, which has the reverse meaning that of what you would actually think, which means it was in the standard. Um so uh uh uh and then the other thing that came about was uh uh allocators at the d at the time, which book broke the notion of a regular type for pair.

Sean ParentAnd that was something that I was, you know, if this is why, you know, st the implementation of std pair is so incredibly long and complex and that pair supports allocators and and it's it's it's in in my opinion ridiculous.

BryceUm I think you're not wrong. It is quite an unintuitive that pair needs to support allocators and the the number of times in the committee that a good feature has been unnecessary has been hung up on, oh, do we need to add allocator support for this thing and the pains associated with that?

Sean ParentUm yeah, so the the stdd.pair gaining allocator support, which broke regularity of std pair. Uh uh that even more than you know the R-value references. I thought, you know, Howard's proposal was sound. I just thought it was too much complexity for the language, but breaking regularity of pair was the uh uh you know, for me was kind of the last straw. And at that point I walked out of the meeting, and that's the last standards committee meeting I've attended.

BryceSo I um so j so John Lako, who's uh another member of the committee, uh very well known for his talks about in advocacy for allocators, um, he has in recent years started promoting the idea that that allocators should be something that are supported at the language level. Um and I actually think that he's probably right about this because the amount of complexity that um has been added to the library to support.

BryceAllocators as a library facility is really great. That's come with a huge amount of cost. And for types that avoid that cost, you know, for things that we say don't support allocators, that then that that takes control away from users. And I I I do wonder if maybe we would have been better off if we had thought up some language-based solution for customizing allocation.

Sean ParentYeah, the the the fundamental problem with allocators is is they invert the the whole part relationship. But uh basically a uh you know an an object that gets allocated is part of the allocator. And and so you can't put the allocator into the part. That's that's that's an an inversion of of that relationship. And uh uh you know that uh that doing the inversion is uh creates what I call you know an incidental data structure.

Sean ParentIt it means now now really what I've got is an allocator that contains objects, but I've expressed it the other way around. I've got a a bunch of objects pointing to an allocator that actually contains them. Um so my opinion has always been that we don't actually need allocators. What we need are better data structures uh that that manage their parts.

Sean ParentYou know, th that manage the allocation of their parts. Um uh so so flip that that that equation around. And that's you know harder to do than just say, well, we can take a vector and we can slap an allocator into it, and now we have uh uh different behavior for a vector. It means you actually have to have to think through um uh uh uh uh uh uh you know a class that makes sense of of of uh of uh uh structures uh that require custom allocation.

Sean ParentUm uh you know, it it so it means you end up with more types and there's not quite the same uniformity, but you get rid of of these these issues.

ConorIs there a library out in the wild that uh implements this kind of idea?

Sean ParentUh you know, I don't think there's a a library out in the wild, but if you look at um uh uh uh you know some of the adapters, which are, you know, if you just just look at you know adapting a vector to be uh a priority queue by using a heap structure in there, um uh that's really just using the vector as if it were an allocator uh for a a heap structure object. And so so you know, one could imagine having something like you know a pooled list if you wanted to say, oh, I've got a a list uh where the the you know the the the uh uh uh for the allocations for the nodes, since I know they're all fixed sized, I can do that more efficiently within some pool, and so I could create a create a structure for that.

Sean ParentUm so no, you know, I don't know of a library that's that's that that's taken that approach, although you know I will say you know this is the approach that you know uh like the Photoshop teams takes. We don't say, oh, well, we've got within our tiling system uh uh which is this you know quad tree data structure. Uh uh we don't say say, and that's connected to this allocator. The allocator for the quad tree structure is part of the quad tree structure.

ConorAlright, well we're we're like fifteen minutes past our scheduled time. And I have a stack of questions, but we'll have to so Sean w uh at ACCU uh said he'd be happy to be a regular recurring guest, which is definitely something that uh we're gonna make happen. Um we'll have to defer the whole macOS X uh that um ends in the Greg Gilly uh story for a whole episode.

BryceUm because I wanted to ask about that, but I knew like that's we're gonna have to do a whole episode of Sean's stories about about Apple and Mac OS.

ConorYeah. Well, there's it's gonna we're we're gonna have to part I don't know how we're gonna partition this up. We said mismatch earlier, and now we're saying partition. Uh but also all the stories from um when you worked with Alex Stepanov, and then also too, we haven't even talked about your time before uh before Adobe. And also too, I was oh, so maybe I'll ask I'll ask this quick follow-up because it it was you know, we we talked a lot about Stepanov uh today. Um and you mentioned sort of he got into C and C while at bot uh at Bell Labs.

ConorUm and we had sort of chatted about this at ACCU, is that I recently discovered from watching one of the uh A9 lectures uh that were recorded when Stepanov worked at um Amazon uh that he actually spent a decade um teaching scheme. Uh and that like I I like to think I know quite a bit about Stepanov and that I've you know I've spent a lot of time on Stepanov Papers.com, but I had no idea that uh he he sort of made the joke that he he lost a decade to scheme. Um so I'm not sure if you want to comment because you knew Stepanov a lot better um about uh you know I don't know uh uh how much to add to that.

Sean ParentUh yeah, he he was uh uh you know he had certainly taught courses in scheme and and uh you know he'd been an associate professor, I forget, at I think at RPI, if I might get that wrong. Um so yeah, so he was an avid scheme user, and all the time I worked with him, he would frequently prototype an algorithm first in scheme and then write it in in C. So he was definitely uh uh very comfortable in Scheme.

ConorYep, so we all gotta go learn uh Lisp or your favorite language of choice.

Sean ParentUh yeah, on that, you know, i i one thing I think that uh I agree with you, Connor, uh uh, about is that you know learning other languages uh uh has a huge amount of value. I think one of the things that set off my career was I went to school at Seattle University and there were twelve of us in our graduating class for computer science. It was their first graduating class in computer science at the university.

Sean ParentAnd we had a couple of young professors who taught the course and they decided that that uh or who taught the program, but the the the professors there decided that they would teach every course in a different language and not teach the language. So you would walk in and you know, first day of of a class, and they would say, uh uh, you know, this class we're going to be looking at um uh uh databases, so you're going to be programming in Ingress, so go learn Ingress, and uh uh we're not teaching it.

Sean ParentAnd in in you know, this class we're going to be doing uh uh image processing and we're going to do all of that in Lisp. So go learn Lisp. And and so y you know, we could talk about the pros and cons. I think there were only a couple of us for which this this plan actually worked. I'm not saying it was a good teaching strategy. Um but for me it was one that did, and it just gave a huge exposure and appreciation to a bunch of different languages very early in my career.

ConorAnd uh this is this is crazy that you're saying this because literally just last night I was watching a uh a talk from Channel 9 with Herb Sutter and Eric Meyer. Uh most of our listeners probably know Herb Sutter is the chair of the ISO C committee. Eric Meyer uh worked on C sharp, and he's a big um functional programming advocate, and they had an hour-long discussion. And at the end of it, Herb Sutter says probably one of the most important things for his career um and the best university uh course that he took and he went to Waterloo in Canada was a course on comparative languages where every week they looked at a different language and they looked at Lisps and Prologue and et cetera.

ConorAnyway, so just the fact that I'm I literally listened to that last night, and now you're saying basically the exact same thing with a different flavor is crazy.

Sean ParentYep. Yep, yeah, yeah. I think there's huge value in that. Um, you know, language affects how you think, and so you know, learning other languages uh uh uh allows you to think about problems from from different perspectives.

ConorSo I think what Sean, and this is where we'll end the episode, is uh trying to say is that everybody needs to go learn APL. Um absolutely.

Sean ParentYou guys gotta put a link at the at the in your show notes for for where to get your APL keyboard and and uh just I don't even have an APL keyboard.

BryceI'm gonna I'm gonna send you one, buddy. I'm gonna send you one.

Sean ParentYeah. All right. It's funny, I can type a bunch of the APL characters just because I used to my my uh uh development environment was something called MPW, which was Apple's programmer's workshop, which was at the time AT ⁇ T owned Unix and was suing everybody who did a Unix knockoff, but MPW was kind of a Unix knockoff. But to not uh uh uh step on AT ⁇ T's toes, um uh uh they used uh different s syntax for everything.

Sean ParentAnd so the the like the wildcard character, if you were doing a search instead of just being an asterisk was, you know, a a squiggly equal sign. Um so so yeah, they were all all you know non-ASCI, unusual characters, and and I still to this day uh you know, when I'm typing in bash, will type those just by habit and have to go, oh no, wait, that's I need a need an asterisk, not a squiggly equal sign there.

ConorThis makes me think, well, now I said we'd end the episode, but this now made me thought of you taught me something, uh Sean at ACCU, which is uh you said you you remarked while listening to an earlier episode is that I didn't I I said that I I don't actually know why they chose IOTA um to represent sort of like a sequence, an arithmetic sequence that increases by one. And you said you actually know. Do you want to educate our listeners why they chose iota? Sure.

Sean ParentIf if uh so so a lot of the the characters in in APL were chosen because of how they look visually. And an iota looks like a small one and it's kind of a backwards j. So it's like a small one and then it kind of hooks out to the right. And so if you say, you know, iota 5, that kind of looks like 1 dot dot dot 5. So, and and that's what iota 5 would do, is it would generate the sequence from 1 to 5.

Sean ParentYou know, APL was 1-based instead of 0-based. Um, you know, out of that now we get, you know, std.io, which is not a glyph, it's spelled out as iota, and it's zero-based by default.

ConorSo but um, there's only one algorithm in the thrust algorithm library that uh renamed a C standard algorithm. And Bryce, do you want to What IOTA?

BryceYeah. You mean std sequence? But but there's a difference. Thrust sequence predated std iota.

ConorThat's not true. We've had this discussion before because it was in the SGI docs, and so a lot of us actually think.

Sean ParentIt's not even in the SGI, it's in the original HPSTL. There's a lot of stuff in the original HP STL that didn't make it into the standard, but yeah, Iota is that is that on Stepanov Papers.com?

ConorI believe, yeah. Yeah, yeah.

Sean ParentThe HP STL is up there.

ConorMaybe we'll have to do an episode where Bryce and I dig through and see all the algorithms that um Alex had to cut off.

Sean ParentYeah, yeah. They went through uh Alex and Bjarna went through uh an editorial process where Bjarna wanted to strip everything that wasn't necessary. So kind of uh it's interesting, you know, what's left is you have a couple of the higher level functions like uh uh you know stable sort and sort, and almost everything in the rest of the libraries is just to build stable sort and sort.

ConorYep. They're all puzzle pieces. I'll link your talk, Generic Programming from 2018 at Pacific C, which is one of my absolute favorite talks of yours, uh, because I think you talk about that in that talk. I believe.

Sean ParentI believe I do, yeah.

ConorAlright, with that, I feel like we should we should wrap it up. Thank you so much, Sean. Okay, thank you guys. Once again, thanks to Sean Parent for coming on. It was an absolute blast chatting with him. And one final announcement is that if you are listening to this before March 31st, there is a free half day conference for APL beginners or the APL curious that are looking to learn a little bit more about the language. You can go to dialogue.com, that's dyalg.com, or you can go to the show notes for a direct link to register.

ConorHope you enjoyed and have a great day.

Open the transcript file

Hosts and Guests

Guests

  • Sean Parent Guest Inferred

Credit the people on this episode

Be the first to rate this episode

More episodes

All episodes
  1. Episode 313: C++26 Reflection vs CRTP, AI, Kids & More with Zach Laine

    Nov 20, 202647 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about C++26 Reflection vs CRTP, AI, kids & more. Link to Episode 313 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP:...

    • Apple Podcasts

    Credit the people on this episode

  2. Episode 312: The C++26 Contracts Fiasco

    Nov 13, 202652 minAudio

    In this episode, Conor and Bryce chat with Zach Laine about the C++26 contracts fiasco. Link to Episode 312 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode

  3. Episode 311: Career Changes, GPUs & More with Zach Laine

    Nov 6, 202628 minAudio

    In this episode, Conor and Bryce talk with Zach Laine about his career and the career changes he has made. Link to Episode 311 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials...

    • Apple Podcasts

    Credit the people on this episode

  4. Episode 310: Leaving NVIDIA for... ?

    Oct 30, 202624 minAudio

    In this episode, Conor and Bryce talk about Bryce's next chapter and where he is leaving NVIDIA for. Link to Episode 310 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The...

    • Apple Podcasts

    Credit the people on this episode

  5. Episode 309: Default Construction & CppCon 2026 with Jason Turner

    Oct 23, 202629 minAudio

    In this episode, Conor and Ben chat with Jason Turner about default construction in C++. Link to Episode 309 on Website Discuss this episode, leave a comment, or ask a question (on GitHub) Socials ADSP: The Podcast:...

    • Apple Podcasts

    Credit the people on this episode