WEBVTT 1 00:00:02.943 --> 00:00:03.804 [MUSIC] 2 00:00:03.804 --> 00:00:06.790 Hello, and welcome to the Haskell weekly podcast. 3 00:00:06.790 --> 00:00:08.870 I'm your host, Taylor Fausak. 4 00:00:08.870 --> 00:00:11.750 I'm the lead engineer at ITProTV. 5 00:00:11.750 --> 00:00:15.360 With me today is Cameron Gera, one of the engineers on my team. 6 00:00:15.360 --> 00:00:16.533 Thanks for joining me today, Cam. 7 00:00:16.533 --> 00:00:17.790 >> Thanks for having me, Taylor. 8 00:00:17.790 --> 00:00:19.600 >> It's good to have you on the show. 9 00:00:19.600 --> 00:00:22.890 And on this show, we talk about Haskell, no surprise there. 10 00:00:22.890 --> 00:00:25.350 It's a purely functional programming language. 11 00:00:25.350 --> 00:00:28.052 But what specifically are we going to be talking about today, Cam? 12 00:00:28.052 --> 00:00:29.885 >> We're gonna be talking about why Haskell? 13 00:00:29.885 --> 00:00:30.730 >> Why Haskell? 14 00:00:30.730 --> 00:00:31.620 What do you mean by that? 15 00:00:31.620 --> 00:00:37.950 >> Well, there's a blog post by the semantic team at GitHub, called Haskell. 16 00:00:37.950 --> 00:00:42.340 And why they chose Haskell for that project, kind of led by Timothy Clem, so. 17 00:00:43.390 --> 00:00:46.410 >> Okay, so GitHub is using Haskell in production, that's awesome. 18 00:00:46.410 --> 00:00:48.010 >> Yeah, it's a big secret. 19 00:00:48.010 --> 00:00:50.680 >> We're not the only company out there with Haskell code. 20 00:00:50.680 --> 00:00:52.240 >> No, there's some big time. 21 00:00:52.240 --> 00:00:55.138 But before we kinda get into their article, 22 00:00:55.138 --> 00:00:58.361 why don't we kinda start with what is Semantic? 23 00:00:58.361 --> 00:01:02.380 And why maybe this Haskell good for them? 24 00:01:02.380 --> 00:01:08.140 >> Yeah, so Semantic is I guess a package or program by the GitHub team, 25 00:01:08.140 --> 00:01:13.608 and it's what they use to analyze and compare source code on their servers. 26 00:01:13.608 --> 00:01:16.855 So when you submit a portal request to GitHub, 27 00:01:16.855 --> 00:01:21.698 you may notice that in addition to being able to break it down by file. 28 00:01:21.698 --> 00:01:25.724 Jump to this file in the diff, you can also jump to specific functions and 29 00:01:25.724 --> 00:01:28.496 say, show me the changes on this function alone, 30 00:01:28.496 --> 00:01:32.200 and Symantic is a tool on their backend that allows that to happen. 31 00:01:32.200 --> 00:01:35.420 So they can parse the source code of JavaScript, or Python, or whatever, and 32 00:01:35.420 --> 00:01:38.340 figure out, this is a function, and something within that function changed. 33 00:01:39.390 --> 00:01:40.020 >> That's awesome. 34 00:01:41.480 --> 00:01:44.780 Why is this announcement big news? 35 00:01:44.780 --> 00:01:48.126 >> This is a big deal because GitHub, obviously, 36 00:01:48.126 --> 00:01:51.480 is a giant force in the development community. 37 00:01:51.480 --> 00:01:54.099 A lot of people use GitHub to manage their source code, and 38 00:01:54.099 --> 00:01:58.370 they're just a huge company, so they've got lots of people working for them. 39 00:01:58.370 --> 00:02:02.184 And this is a big deal because Haskell is a pretty small community, 40 00:02:02.184 --> 00:02:04.568 especially in relation to all of GitHub, 41 00:02:04.568 --> 00:02:07.991 there aren't too many companies out there using Haskell. 42 00:02:07.991 --> 00:02:12.232 A couple of the big ones that I can think of off the top of my head are Facebook and 43 00:02:12.232 --> 00:02:13.613 Nike, and us of course. 44 00:02:13.613 --> 00:02:15.600 But ITProTV is not as big as them yet. 45 00:02:15.600 --> 00:02:17.492 >> We're coming for you, Facebook. 46 00:02:17.492 --> 00:02:20.550 >> [LAUGH] >> ITProTV, the next Facebook in IT world? 47 00:02:22.260 --> 00:02:25.170 >> But yeah, this is just a big deal, cuz GitHub is a big company, and 48 00:02:25.170 --> 00:02:28.700 Haskell is a small language, and it is really interesting to a lot of people. 49 00:02:28.700 --> 00:02:31.560 >> Yeah, it's good to see the community growing, 50 00:02:31.560 --> 00:02:32.262 and- >> Yeah. 51 00:02:32.262 --> 00:02:36.697 >> The neighboring organizations picking up this language allows for 52 00:02:36.697 --> 00:02:38.960 the community to obviously grow. 53 00:02:38.960 --> 00:02:41.565 I mean, there's so many employees at GitHub, and 54 00:02:41.565 --> 00:02:44.730 obviously just the Semantic team right now is using Haskell. 55 00:02:44.730 --> 00:02:47.140 But I'm sure it's spreading through their organization. 56 00:02:47.140 --> 00:02:51.747 And it's gonna allow our community to continue to grow, and 57 00:02:51.747 --> 00:02:54.198 have more experts in the field. 58 00:02:54.198 --> 00:02:54.881 >> Absolutely. 59 00:02:54.881 --> 00:02:55.779 >> It's really cool. 60 00:02:55.779 --> 00:02:58.698 >> And even if you're at a smaller company, and you're interested in using 61 00:02:58.698 --> 00:03:01.520 Haskell, this can be really useful, because you can point to get up and 62 00:03:01.520 --> 00:03:03.960 say look, they're using it, and it's working for them. 63 00:03:03.960 --> 00:03:06.060 It can work for us too, right? 64 00:03:06.060 --> 00:03:10.680 >> And a lot of the articles out there about why people choose Haskell 65 00:03:10.680 --> 00:03:13.520 are pretty normal. 66 00:03:13.520 --> 00:03:17.520 They're pretty standard like this this and this, but 67 00:03:17.520 --> 00:03:22.440 why do you think you know GitHub kinda chose this different path? 68 00:03:24.350 --> 00:03:26.180 >> This different path to using Haskell? 69 00:03:26.180 --> 00:03:27.170 >> Right. 70 00:03:27.170 --> 00:03:27.920 >> Yeah. >> So for 71 00:03:27.920 --> 00:03:31.545 a lot of the blog posts that I've read in the past about why companies 72 00:03:31.545 --> 00:03:36.015 chose Haskell, you sort of see the same reasons trotted out. 73 00:03:36.015 --> 00:03:37.525 Haskell is a pure language, 74 00:03:37.525 --> 00:03:40.835 so you don't have to worry about side effects in weird parts of your code. 75 00:03:40.835 --> 00:03:43.775 Or it has lazy evaluation, so you don't have to worry about things being 76 00:03:43.775 --> 00:03:45.635 evaluated, even if you don't end up using them. 77 00:03:47.018 --> 00:03:52.410 Or it let's you express complicated ideas in a small amount of code. 78 00:03:52.410 --> 00:03:55.680 And the first time you see that set of arguments, it might be compelling. 79 00:03:55.680 --> 00:03:59.090 But the 5th, or 10th, or 100th time you see them, you're like, okay, I get it. 80 00:03:59.090 --> 00:04:01.110 I know Haskell has those things, what else can I do? 81 00:04:01.110 --> 00:04:02.850 >> No, that's awesome, yeah. 82 00:04:02.850 --> 00:04:08.000 And I think it's really cool that they kinda broke it down. 83 00:04:08.000 --> 00:04:09.840 They had some different reasons for choosing Haskell. 84 00:04:09.840 --> 00:04:16.060 So do you kind of want to kick off, what reasons and why they chose Haskell? 85 00:04:16.060 --> 00:04:19.420 >> Sure, as I mentioned, there's kind of the bog standard reasons for 86 00:04:19.420 --> 00:04:21.320 why people typically choose Haskell. 87 00:04:21.320 --> 00:04:24.330 But one thing that I like about this article is they didn't dwell on 88 00:04:24.330 --> 00:04:24.930 that too much. 89 00:04:24.930 --> 00:04:27.780 They said yeah, these are all characteristics of Haskell. 90 00:04:27.780 --> 00:04:29.590 These are reasons why you might wanna choose it. 91 00:04:29.590 --> 00:04:33.450 But they ended up focusing more on what things 92 00:04:34.520 --> 00:04:38.400 that are only possible, because of those kind of fundamental characteristics, 93 00:04:38.400 --> 00:04:42.080 but more compelling reasons for their particular business case. 94 00:04:42.080 --> 00:04:44.980 And I think the one that they start with is control flow, right? 95 00:04:44.980 --> 00:04:46.180 >> Correct, yep. 96 00:04:46.180 --> 00:04:51.690 >> And so by control flow, what they mean is that in a typical programming language, 97 00:04:51.690 --> 00:04:54.940 something like Java or C#, or 98 00:04:54.940 --> 00:04:58.680 go the control flow is built into the language itself. 99 00:04:58.680 --> 00:05:01.450 So if a statement is above another one, 100 00:05:01.450 --> 00:05:03.575 you know that it's going to get executed first. 101 00:05:03.575 --> 00:05:05.708 So from top to bottom left to right, basically. 102 00:05:05.708 --> 00:05:09.170 And in Haskell, that's not necessarily the case. 103 00:05:09.170 --> 00:05:14.080 You can write code that looks the same, but one version of it might be 104 00:05:14.080 --> 00:05:16.960 synchronous, and another version of it might be asynchronous. 105 00:05:16.960 --> 00:05:21.930 So this comes up in JavaScript all the time where the synchronous 106 00:05:21.930 --> 00:05:25.090 function is really easy to write, but you block the main thread for a long time. 107 00:05:25.090 --> 00:05:27.360 If you're doing something with a web UI, 108 00:05:27.360 --> 00:05:31.340 then your UI stalls while that synchronous thing is happening. 109 00:05:31.340 --> 00:05:33.590 So you wanna write stuff in this a sync fashion. 110 00:05:33.590 --> 00:05:35.480 But then you have to do like nested call backs. 111 00:05:35.480 --> 00:05:37.850 So you do that a sync await, or something like that, right? 112 00:05:37.850 --> 00:05:41.350 So the syntax, you have to change the way that you write your code, 113 00:05:41.350 --> 00:05:44.030 in order to change how the control flow works. 114 00:05:44.030 --> 00:05:49.078 But with Haskell, the code that you write looks the same, 115 00:05:49.078 --> 00:05:52.386 but you can change how it is executed. 116 00:05:52.386 --> 00:05:53.112 >> Right, right, right. 117 00:05:53.112 --> 00:05:58.699 So pretty much higher up in the program is where is kind of the deciding factor of, 118 00:05:58.699 --> 00:06:02.361 does this sync, is this asynchronous or synchronous? 119 00:06:02.361 --> 00:06:03.050 >> Exactly. 120 00:06:03.050 --> 00:06:06.230 >> And the further you get in to the code, it just looks the same. 121 00:06:06.230 --> 00:06:11.435 You don't have to be worried of, am I writing this the right way? 122 00:06:11.435 --> 00:06:12.220 >> Mm-hm. 123 00:06:12.220 --> 00:06:14.940 >> Because in the end, I want this result. 124 00:06:14.940 --> 00:06:19.540 The code you write can be executed however you want it to be executed. 125 00:06:19.540 --> 00:06:22.950 You just tell that once you start to write the executable, and 126 00:06:22.950 --> 00:06:24.490 make that code come true. 127 00:06:24.490 --> 00:06:28.290 >> Yeah, and that's especially useful for the Semantic team, 128 00:06:28.290 --> 00:06:31.708 because what they're doing is parsing and analyzing source code. 129 00:06:31.708 --> 00:06:36.390 So they wanna be able to parse, let's say, a Python file and 130 00:06:36.390 --> 00:06:40.370 analyze it as if it was executing, but not actually execute it. 131 00:06:40.370 --> 00:06:43.710 And so that becomes really powerful to have all these tools that Haskell provides 132 00:06:43.710 --> 00:06:48.010 to be able to manage the control flow in the same way that you manage regular data 133 00:06:48.010 --> 00:06:49.250 in other languages. 134 00:06:49.250 --> 00:06:51.620 >> Right, right, no, I think that's awesome. 135 00:06:52.690 --> 00:06:54.730 So the other thing they kinda talk about and 136 00:06:54.730 --> 00:06:58.010 the reason they chose it was runtime correctness. 137 00:06:59.080 --> 00:07:03.160 And they gave a lot of really great things we've kinda experienced, 138 00:07:03.160 --> 00:07:08.220 but what were some of the things you took away from that aspect to the article? 139 00:07:08.220 --> 00:07:10.121 >> I really liked that aspect to the article, and 140 00:07:10.121 --> 00:07:11.550 I'm glad that they brought it up. 141 00:07:11.550 --> 00:07:14.192 Because when you read about, let's say Elm, 142 00:07:14.192 --> 00:07:16.717 another language that's in the same vein. 143 00:07:16.717 --> 00:07:20.730 They bring that front and center all the time, it's on their home page. 144 00:07:20.730 --> 00:07:22.830 They say you can write programs with confidence, and 145 00:07:22.830 --> 00:07:24.580 not have to worry about runtime errors. 146 00:07:24.580 --> 00:07:27.052 And I wish Haskell marketed more in that direction, 147 00:07:27.052 --> 00:07:30.771 because it is really powerful to not have to worry about those runtime errors. 148 00:07:30.771 --> 00:07:34.684 And trust that, when you write the program and it compiles, 149 00:07:34.684 --> 00:07:39.430 it will generally do what you want, or what you intended for it to do. 150 00:07:39.430 --> 00:07:41.120 >> Right. >> And you don't have to worry, crap, 151 00:07:41.120 --> 00:07:43.550 well, I forgot that this value could be null. 152 00:07:43.550 --> 00:07:46.000 So then you get a null pointer exception in production, or 153 00:07:46.000 --> 00:07:49.330 you forgot I changed the name of that method last week, and 154 00:07:49.330 --> 00:07:50.910 I forgot to update it here at this call site. 155 00:07:50.910 --> 00:07:53.800 So now I'm getting a method not found error, right? 156 00:07:53.800 --> 00:07:57.228 It saves you from those really tedious kind of rote problems. 157 00:07:57.228 --> 00:08:01.090 And so that you can focus on developing new features, and shipping those out to 158 00:08:01.090 --> 00:08:05.010 your users, rather than fixing some Bug that popped up in your issue tracker for 159 00:08:05.010 --> 00:08:06.082 the hundredth time. 160 00:08:06.082 --> 00:08:09.113 >> Yeah, one thing I wish Haskell was magical. 161 00:08:09.113 --> 00:08:12.425 And it can be we just aren't necessarily using it right now. 162 00:08:12.425 --> 00:08:17.332 I like to put TV is that kind of a way to get away 163 00:08:17.332 --> 00:08:21.980 from accidentally like misaligning like 164 00:08:21.980 --> 00:08:26.912 a value from, JSON or something like that. 165 00:08:26.912 --> 00:08:31.846 Because we've kind of, some of our parsing, for our JSON parsing we 166 00:08:31.846 --> 00:08:36.695 use the applicative stuff and that kind of ensures that like, hey, 167 00:08:36.695 --> 00:08:42.248 this could end up being different because you misalign the keys or whatever. 168 00:08:42.248 --> 00:08:45.955 So I think as a group of developers I would like for 169 00:08:45.955 --> 00:08:49.032 us to kind of lead lean more into, okay, 170 00:08:49.032 --> 00:08:54.200 let's not use a plug it in just so we can make sure we keep our sanity. 171 00:08:54.200 --> 00:08:58.488 Because I mean, a few months ago, I was just banging my head against the wall 172 00:08:58.488 --> 00:09:03.245 trying to figure out like, wait a second, what's happening here and sure enough, 173 00:09:03.245 --> 00:09:05.535 it was just a miss of flip-flopped key. 174 00:09:05.535 --> 00:09:08.791 It was like [SOUND] >> That brings up a good point that just 175 00:09:08.791 --> 00:09:12.030 because Haskell gets rid of a large portion of runtime errors, 176 00:09:12.030 --> 00:09:14.487 it doesn't mean that it gets rid of all of them. 177 00:09:14.487 --> 00:09:18.907 And in fact, it's not even desirable to get rid of all of them because you'd have 178 00:09:18.907 --> 00:09:21.192 to pay so much upfront in order to do that. 179 00:09:21.192 --> 00:09:25.343 But there are somethings like you mentioned with the applicative syntax 180 00:09:25.343 --> 00:09:29.290 where they allow you to sort of accidentally introduce problems that 181 00:09:29.290 --> 00:09:32.711 are very hard to detect and there are easy ways around that. 182 00:09:32.711 --> 00:09:35.845 In our case writing those pressures in a more 183 00:09:35.845 --> 00:09:40.344 monastic style makes it clearer which things are going where. 184 00:09:40.344 --> 00:09:44.541 So you can look at a line and say no, we're pulling this field it's named price 185 00:09:44.541 --> 00:09:48.679 but we're pulling it from object where the key we're looking for his name, and 186 00:09:48.679 --> 00:09:50.490 that obviously doesn't line up. 187 00:09:50.490 --> 00:09:54.974 >> Right, right, right, yeah, so I mean I do think a lot of the things 188 00:09:54.974 --> 00:09:59.086 that Haskell provides for runtime correctness is are great. 189 00:09:59.086 --> 00:10:02.638 So I'm really glad that they kind of echoed that because like you said, 190 00:10:02.638 --> 00:10:05.015 Elm, that's something they boast all about. 191 00:10:05.015 --> 00:10:05.808 >> Yeah. 192 00:10:05.808 --> 00:10:08.551 >> But housing you don't really hear about that until you're in it. 193 00:10:08.551 --> 00:10:11.986 And you're like, this isn't gonna let me change a function 194 00:10:11.986 --> 00:10:15.801 name over here without updating all the colors, that's not possible. 195 00:10:15.801 --> 00:10:16.500 >> Yep. 196 00:10:16.500 --> 00:10:19.293 >> Because that compiler will not be happy with you. 197 00:10:19.293 --> 00:10:21.482 >> [LAUGH] >> That's awesome. 198 00:10:21.482 --> 00:10:25.164 The other thing they kinda touched on was kinda the research. 199 00:10:25.164 --> 00:10:31.169 So academia is kinda leaning more towards Haskell as a language that, 200 00:10:31.169 --> 00:10:37.074 these new research methods are kinda written in these algorithms and 201 00:10:37.074 --> 00:10:39.232 all this various stuff. 202 00:10:39.232 --> 00:10:44.340 What is that useful for GitHub? 203 00:10:44.340 --> 00:10:48.801 >> You touched on it a little already in that a lot of computer science papers 204 00:10:48.801 --> 00:10:53.207 are written using Haskell or a language that's very close to Haskell. 205 00:10:53.207 --> 00:10:57.959 But that kind of typical computer science paper that you read with the font that 206 00:10:57.959 --> 00:11:01.643 everybody knows and the to column layout and all that stuff. 207 00:11:01.643 --> 00:11:04.838 The code that you look at in those papers is usually Haskell. 208 00:11:04.838 --> 00:11:10.015 And if you can take that code from an established research paper and 209 00:11:10.015 --> 00:11:15.939 more or less copy paste it into your library, and start using it right away. 210 00:11:15.939 --> 00:11:17.599 That's a huge win versus, 211 00:11:17.599 --> 00:11:21.942 if you were using basically any other language you would have to think. 212 00:11:21.942 --> 00:11:25.706 Okay, how can I put this behavior that I want that they're talking about 213 00:11:25.706 --> 00:11:28.003 in this paper into the language that I have? 214 00:11:28.003 --> 00:11:31.568 And you have to change the semantics, you have to change the actual shape 215 00:11:31.568 --> 00:11:35.556 of the code itself, you have to make sure everything works the way you expected. 216 00:11:35.556 --> 00:11:37.933 So it's just more difficult to do that. 217 00:11:37.933 --> 00:11:42.382 Now, that's not a common problem for work-all-day programmers like us, 218 00:11:42.382 --> 00:11:43.895 we're writing web apps. 219 00:11:43.895 --> 00:11:45.959 >> Right. >> We're not looking into research papers 220 00:11:45.959 --> 00:11:47.847 to figure out, well, how do we parse JSON? 221 00:11:47.847 --> 00:11:49.860 It's kind of a solved problem at this point. 222 00:11:49.860 --> 00:11:52.952 [LAUGH] >> Correct, you're right, which is nice, 223 00:11:52.952 --> 00:11:57.691 but it is also helpful if yo're trying to be on the cutting edge of technology and 224 00:11:57.691 --> 00:11:59.336 the growth of technology. 225 00:11:59.336 --> 00:12:03.320 Research papers have Haskell, that's where the code is. 226 00:12:03.320 --> 00:12:04.092 >> Yeah. >> That's what it is. 227 00:12:04.092 --> 00:12:05.742 Those papers are still coming out, 228 00:12:05.742 --> 00:12:08.166 programming language theory is not a solved field. 229 00:12:08.166 --> 00:12:11.947 And the Semantic team at GitHub Is on the cutting edge of this, 230 00:12:11.947 --> 00:12:14.057 they are analyzing source code for 231 00:12:14.057 --> 00:12:18.949 millions of software projects across many different programming languages. 232 00:12:18.949 --> 00:12:23.396 And they wanna do it quickly, effectively, safely and the research is kinda showing 233 00:12:23.396 --> 00:12:26.811 them how to do that and they're able to crib from it effectively. 234 00:12:26.811 --> 00:12:28.316 >> And I think that's really awesome. 235 00:12:28.316 --> 00:12:30.924 So I definitely applaud them for- >> Yeah, me too. 236 00:12:30.924 --> 00:12:32.654 >> This article. 237 00:12:32.654 --> 00:12:37.138 They also kinda talked about the things I didn't like, which is a helpful 238 00:12:37.138 --> 00:12:41.498 thing when you're analyzing, okay, do I jump in and go with Haskell? 239 00:12:41.498 --> 00:12:47.991 Or do I lean towards the Java or C#, or something a little more mainstream? 240 00:12:47.991 --> 00:12:50.435 >> Yeah, I always like it when I see an article like this. 241 00:12:50.435 --> 00:12:52.365 And they include the stuff they didn't like. 242 00:12:52.365 --> 00:12:55.998 Otherwise, it's really easy to look at it and discount it offhand, and think well, 243 00:12:55.998 --> 00:12:59.637 they just really like this and it's their pet language, and they want it to succeed. 244 00:12:59.637 --> 00:13:01.288 So they're not gonna say anything bad about it. 245 00:13:01.288 --> 00:13:02.151 >> Is it their cat language? 246 00:13:02.151 --> 00:13:03.218 >> [LAUGH] >> Or their dog language? 247 00:13:03.218 --> 00:13:04.007 >> A cat or dog, right? 248 00:13:04.007 --> 00:13:05.235 Are they cat or dog people, do you think? 249 00:13:05.235 --> 00:13:06.837 >> Maybe fish, they could be fish people. 250 00:13:06.837 --> 00:13:07.959 >> Fish people. 251 00:13:07.959 --> 00:13:09.317 Lizard people. 252 00:13:09.317 --> 00:13:12.444 >> The Aquaman. 253 00:13:12.444 --> 00:13:15.355 [LAUGH] >> All right, 254 00:13:15.355 --> 00:13:20.754 I think it does add credibility to this document or blog post. 255 00:13:20.754 --> 00:13:21.507 >> Yeah, this article. 256 00:13:21.507 --> 00:13:22.618 >> Very unofficial article. 257 00:13:22.618 --> 00:13:25.294 >> Yeah. [LAUGH] >> Articles are a very good term, I should use that more often. 258 00:13:25.294 --> 00:13:30.399 But they talk about kind of the weak tooling aspect. 259 00:13:30.399 --> 00:13:36.317 This is something we've come to deal with. 260 00:13:36.317 --> 00:13:38.476 >> Yeah, it's a problem we've definitely run into. 261 00:13:38.476 --> 00:13:41.993 >> Right, and so do you wanna kinda explain what we do to kinda counteract 262 00:13:41.993 --> 00:13:43.073 this and- >> Sure. 263 00:13:43.073 --> 00:13:44.375 >> Acknowledge it? 264 00:13:44.375 --> 00:13:46.545 >> First, just kinda for a statement of terms. 265 00:13:46.545 --> 00:13:50.786 I think that when they talk about tooling, they're talking about like 266 00:13:50.786 --> 00:13:54.616 an ID integration, something like IntelliJ or Visual Studio. 267 00:13:54.616 --> 00:13:59.941 Where when you're working on code, you can mouse over a certain part and it'll show 268 00:13:59.941 --> 00:14:04.816 you perhaps the documentation for that thing or what parameters it expects or 269 00:14:04.816 --> 00:14:10.227 where in your code base it's used, stuff that isn't earth shattering on it's own. 270 00:14:10.227 --> 00:14:14.689 But a bunch of small niceties that add up to a really positive experience. 271 00:14:14.689 --> 00:14:18.605 And they're right to say that Haskell is kinda weak in this regards, 272 00:14:18.605 --> 00:14:20.774 there are certainly tools that help. 273 00:14:20.774 --> 00:14:24.049 The one that we use most often is GHCID. 274 00:14:24.049 --> 00:14:27.507 >> Another one I was just thinking about offhand, so 275 00:14:27.507 --> 00:14:30.562 I didn't catch you off on this but Interro. 276 00:14:30.562 --> 00:14:32.937 >> Yeah. >> Interro offers, 277 00:14:32.937 --> 00:14:36.274 the awesome IDE experience. 278 00:14:36.274 --> 00:14:39.937 But it takes over people's machines and like, crashes them. 279 00:14:39.937 --> 00:14:43.966 >> [LAUGH] Yeah and we've run into problems with Haskell IDE engine where 280 00:14:43.966 --> 00:14:46.953 the promise of it is excellent but when you run it for 281 00:14:46.953 --> 00:14:50.930 more than a couple hours it starts heating up all the RAM it can find. 282 00:14:50.930 --> 00:14:51.866 So- >> It's hungry. 283 00:14:51.866 --> 00:14:54.702 >> I think that Haskell is moving in the right direction and 284 00:14:54.702 --> 00:14:57.370 they seem to have landed on the same solutions as us. 285 00:14:57.370 --> 00:15:01.563 Which is you have your editor up maybe the left hand side of your screen and 286 00:15:01.563 --> 00:15:04.107 your compiler output up on the right hand and 287 00:15:04.107 --> 00:15:06.813 you kind of stitch things together yourself. 288 00:15:06.813 --> 00:15:10.239 >> Obviously, not ideal but definitely workable. 289 00:15:10.239 --> 00:15:12.364 And just to kinda color that a bit, 290 00:15:12.364 --> 00:15:16.833 from my point of view, I think this comes down mostly to lack of resources 291 00:15:16.833 --> 00:15:20.814 because developing all of that tooling just takes a lot of work. 292 00:15:20.814 --> 00:15:24.631 And for these other languages like Java or JavaScript, you have millions of people 293 00:15:24.631 --> 00:15:27.591 working in these languages, and that's not true for Haskell. 294 00:15:27.591 --> 00:15:31.529 So I'm excited to see GitHub talk about this because if they get people interested 295 00:15:31.529 --> 00:15:35.410 and excited about Haskell and working in it, then the tooling is naturally gonna 296 00:15:35.410 --> 00:15:37.829 improve as more people get in with the community. 297 00:15:37.829 --> 00:15:39.370 >> No, I think that's a very good point. 298 00:15:39.370 --> 00:15:41.818 So way to go, GitHub. 299 00:15:41.818 --> 00:15:43.196 >> [LAUGH] >> Awesome, well, 300 00:15:43.196 --> 00:15:46.325 they also kind of talk about I think that's a very good point. 301 00:15:46.325 --> 00:15:48.817 [LAUGH] So way to go, github. 302 00:15:48.817 --> 00:15:51.199 >> No dependent types. 303 00:15:51.199 --> 00:15:53.280 Why do you think this is an issue for them? 304 00:15:53.280 --> 00:15:56.910 >> So we already talked a little how they kinda borrowed from current programming 305 00:15:56.910 --> 00:15:59.667 language research and apply it, to their problem domain. 306 00:15:59.667 --> 00:16:04.532 So they're already in a kind of unusual situation as far as that goes. 307 00:16:04.532 --> 00:16:09.250 And I think that's why they feel this pain from the lack of dependent typing, 308 00:16:09.250 --> 00:16:14.118 because they're pushing, Haskell's type system to its limits and Haskell, 309 00:16:14.118 --> 00:16:18.258 while it's well-revered for having a strong useful type system, 310 00:16:18.258 --> 00:16:20.390 it does not have dependent types. 311 00:16:20.390 --> 00:16:23.904 And there's lot of things you can do to try to pretend like you have them or 312 00:16:23.904 --> 00:16:26.726 give you some of the same benefits of dependent types, but 313 00:16:26.726 --> 00:16:28.821 at the end of the day, it doesn't happen. 314 00:16:28.821 --> 00:16:32.921 And I can't speak personally to this cause I haven't run into this particular 315 00:16:32.921 --> 00:16:35.540 limitation in the type of development that we do. 316 00:16:35.540 --> 00:16:39.389 And I think that most developers would not run into this problem, they wouldn't 317 00:16:39.389 --> 00:16:43.539 think, well, I wish I had dependent types here, it doesn't come up that often. 318 00:16:43.539 --> 00:16:46.743 But for them it does, I don't wanna discount their experience here, 319 00:16:46.743 --> 00:16:49.093 obviously they wish they had this and they don't. 320 00:16:49.093 --> 00:16:53.574 So the you know the only languages that do have this they're they're few and 321 00:16:53.574 --> 00:16:55.895 far between stuff like Okta and Idris. 322 00:16:55.895 --> 00:16:59.726 They have this but they're they're a little less production ready than Haskell 323 00:16:59.726 --> 00:17:01.542 is they're not used quite as widely. 324 00:17:01.542 --> 00:17:06.056 >> Right, that's interesting, yeah, I don't think we've really 325 00:17:06.056 --> 00:17:09.950 thought of dependent types has never really been a thing. 326 00:17:09.950 --> 00:17:13.636 >> I think that it's a great idea and I'm really interested in it but 327 00:17:13.636 --> 00:17:17.522 day to day there's not too much where I'm like, this would be so 328 00:17:17.522 --> 00:17:20.151 much easier if I had dependent types, drats. 329 00:17:20.151 --> 00:17:25.860 >> Awesome, well, they also kind of talk about kind of the lack of infra glue. 330 00:17:25.860 --> 00:17:28.633 >> Yeah, I like that terminology they used, infra glue, 331 00:17:28.633 --> 00:17:31.827 it sounds like something you buy from a TV channel or something. 332 00:17:31.827 --> 00:17:36.014 >> Right, here is your infra glu, glue, wow, I can't even talk. 333 00:17:36.014 --> 00:17:39.445 >> But yeah, so they mentioned that for the actual meat and 334 00:17:39.445 --> 00:17:42.809 potatoes of their program, Haskell does a great job. 335 00:17:42.809 --> 00:17:46.620 But for the stuff where they have to worry about deploying or making containers or 336 00:17:46.620 --> 00:17:50.184 stuff like that, It's a little more difficult. 337 00:17:50.184 --> 00:17:55.111 And that's mostly because it's similar to the tooling problem where just not as many 338 00:17:55.111 --> 00:17:56.693 people have been using it. 339 00:17:56.693 --> 00:18:00.579 So there's not a canned solution but I go I want to deploy this to AWS or 340 00:18:00.579 --> 00:18:02.734 I want to make a container out of this. 341 00:18:02.734 --> 00:18:06.100 You can definitely do those things but you're probably going to have to do them 342 00:18:06.100 --> 00:18:09.415 yourself rather than just picking something off the shelf that says, okay, 343 00:18:09.415 --> 00:18:11.978 yeah, I'll play this Python program to look at whatever. 344 00:18:11.978 --> 00:18:13.118 >> It's a little more boilerplate. 345 00:18:13.118 --> 00:18:17.410 >> Yeah, and again not a show stopper and not terribly difficult to get around, 346 00:18:17.410 --> 00:18:19.666 but just annoying that it's not there. 347 00:18:19.666 --> 00:18:24.829 >> Yeah, that's fair, but one of the more interesting 348 00:18:24.829 --> 00:18:29.647 ones that I feel like we've encountered a lot is, 349 00:18:29.647 --> 00:18:32.990 the issues with lazy evaluation. 350 00:18:32.990 --> 00:18:35.913 I think for us we've had various scripts and 351 00:18:35.913 --> 00:18:41.139 you know things as you know our code is trying to figure out what it's gonna do. 352 00:18:41.139 --> 00:18:45.043 Until it actually needs to execute something it just builds up this giant 353 00:18:45.043 --> 00:18:47.225 thunk that hangs out all over the place. 354 00:18:47.225 --> 00:18:51.933 And you're like, everything slows down and you're like what's going on? 355 00:18:51.933 --> 00:18:54.441 >> And then you put one exclamation point in one place and 356 00:18:54.441 --> 00:18:56.121 everything works again, ta-da! 357 00:18:56.121 --> 00:19:01.063 >> Yeah and they kinda talked about their language extension, strict data? 358 00:19:01.063 --> 00:19:02.269 >> Strict data, I was like, yeah. 359 00:19:02.269 --> 00:19:03.039 >> Static data, no? 360 00:19:03.039 --> 00:19:07.433 >> Strict data, yeah, and so that's kind of how they interacted it, but 361 00:19:07.433 --> 00:19:10.670 why do you think for them that's kind of a big issue? 362 00:19:10.670 --> 00:19:14.432 >> I think that lazy evaluation is a double edged sword in Haskell for sure. 363 00:19:14.432 --> 00:19:19.055 And the main reason for that is that Haskell is pretty much the only 364 00:19:19.055 --> 00:19:24.786 programming language that I can think of off down my head that is lazy by default. 365 00:19:24.786 --> 00:19:28.019 Every other programming language is strict by default. 366 00:19:28.019 --> 00:19:32.515 So as programmers, we're not used to identifying and solving, 367 00:19:32.515 --> 00:19:37.434 problems that stem from laziness, because we rarely encounter them. 368 00:19:37.434 --> 00:19:39.318 In other languages, you have to explicitly opt in. 369 00:19:39.318 --> 00:19:41.647 So like in Python, you can make a generator, or 370 00:19:41.647 --> 00:19:45.317 in JavaScript you do all the acing stuff to pretend like you have laziness, 371 00:19:45.317 --> 00:19:47.376 whereas in Haskell, it's everywhere. 372 00:19:47.376 --> 00:19:51.908 So people just aren't used to looking for those types of problems and so 373 00:19:51.908 --> 00:19:54.789 they're bound to crop up every now and then. 374 00:19:54.789 --> 00:19:59.723 That being said, they're usually not terribly hard to diagnose and 375 00:19:59.723 --> 00:20:02.963 fix once you have a couple tricks in your bag. 376 00:20:02.963 --> 00:20:06.165 But the first time that you run into them, they can be completely mystifying 377 00:20:06.165 --> 00:20:08.779 because you can look at code and it looks totally reasonable. 378 00:20:08.779 --> 00:20:13.936 But for whatever reason, the way that you wrote it either generates too many thunks, 379 00:20:13.936 --> 00:20:18.100 like you mentioned, and you don't evaluate them at the right time. 380 00:20:18.100 --> 00:20:19.595 Or just something like that, I mean, 381 00:20:19.595 --> 00:20:21.805 can be really difficult to figure out what's going on. 382 00:20:21.805 --> 00:20:26.810 And again, this comes back to the lack of tooling, if there was better tooling to 383 00:20:26.810 --> 00:20:31.685 analyze the laziness of your program this could be an easier problem to solve. 384 00:20:31.685 --> 00:20:34.916 >> So maybe with time, maybe with time, maybe with time. 385 00:20:34.916 --> 00:20:39.427 The last one, this one hits home for us is they kinda of said, 386 00:20:39.427 --> 00:20:44.483 it's gotta a notorious interpretation of being too hard to learn. 387 00:20:44.483 --> 00:20:48.854 As bunch of junior developers who just minus you and 388 00:20:48.854 --> 00:20:53.226 Cody, learning Haskell in the last few years it had 389 00:20:53.226 --> 00:20:57.816 it's a little bit of growing pains in the beginning. 390 00:20:57.816 --> 00:21:02.878 But once you understand the concept and the paradigm of functional programming, 391 00:21:02.878 --> 00:21:06.985 it becomes a lot more attainable to really conquer the language and 392 00:21:06.985 --> 00:21:09.060 have a good understanding of it. 393 00:21:09.060 --> 00:21:13.697 Without you we're not the deepest understanding but you can have a general 394 00:21:13.697 --> 00:21:17.608 idea what's going on, so what did they have to say about that. 395 00:21:17.608 --> 00:21:22.261 >> I think they say something very similar to what we think and hope I've said but 396 00:21:22.261 --> 00:21:23.335 if we haven't yet 397 00:21:23.335 --> 00:21:28.366 we'll say now that Haskell does have this reputation of being difficult to learn. 398 00:21:28.366 --> 00:21:29.818 But it's not that difficult and 399 00:21:29.818 --> 00:21:33.583 it's especially not that difficult to get to a working knowledge, where maybe you 400 00:21:33.583 --> 00:21:36.942 don't have a super duper understanding of everything that's going on. 401 00:21:36.942 --> 00:21:41.678 But you can get worked on and you can feel effective and be effective, 402 00:21:41.678 --> 00:21:44.781 you don't have to learn category theory, or 403 00:21:44.781 --> 00:21:49.941 really understand what is a monad in order to program in Haskell day to day. 404 00:21:49.941 --> 00:21:54.113 And they say, especially with some type of, mentor or 405 00:21:54.113 --> 00:21:59.115 tutor on the team, you can get people up to speed really quickly. 406 00:21:59.115 --> 00:22:01.767 And we've seen that we hired somebody a month ago and 407 00:22:01.767 --> 00:22:04.489 they got up to speed basically within the first week. 408 00:22:04.489 --> 00:22:08.115 Now, granted they obviously don't understand all of Haskell, everything, 409 00:22:08.115 --> 00:22:10.668 I mean, it's a huge language, there's a lot to know. 410 00:22:10.668 --> 00:22:13.498 >> Right. >> But they know enough to be effective, 411 00:22:13.498 --> 00:22:16.841 and write code that works, and submit it for PR and 412 00:22:16.841 --> 00:22:20.759 get it into production within a week or two of being hired. 413 00:22:20.759 --> 00:22:24.601 >> Right. >> It's not insurmountable. 414 00:22:24.601 --> 00:22:27.457 >> Yeah. >> And I think another attribute for us, 415 00:22:27.457 --> 00:22:32.870 attribute to our ability to learn Haskell, has been the idea of promiscuous pairing, 416 00:22:32.870 --> 00:22:37.910 which for those listeners, you guys check it out, there's a cool paper on it. 417 00:22:37.910 --> 00:22:41.169 So we've been doing it for a little while now and we find it very effective, 418 00:22:41.169 --> 00:22:43.158 t allows it knowledge transfer really well. 419 00:22:43.158 --> 00:22:49.394 So I did a shameless plug for Jason, [LAUGH] he's all he's all about it. 420 00:22:49.394 --> 00:22:53.565 >> Yeah, he's all about it and it really ties into what this semantic team at 421 00:22:53.565 --> 00:22:58.276 GitHub said, of having a tutor or somebody who knows the loops to show the new person 422 00:22:58.276 --> 00:23:01.729 promiscuous pairing, just forces that every time around. 423 00:23:01.729 --> 00:23:03.323 >> Everyone must be a learner. 424 00:23:03.323 --> 00:23:04.431 >> [LAUGH]. >> A teacher, 425 00:23:04.431 --> 00:23:06.907 I'm whispering into a microphone. 426 00:23:06.907 --> 00:23:09.593 >> [LAUGH] >> That's kind of weird, all right, and 427 00:23:09.593 --> 00:23:12.828 obviously, this program is pretty large semantic. 428 00:23:12.828 --> 00:23:18.708 Like as far as the depth of what it can do, but it's only 20,000 lines of code. 429 00:23:18.708 --> 00:23:22.153 Yeah and that's amazing because as they point out several times 430 00:23:22.153 --> 00:23:25.547 in their article if they had written this in, let's say Java. 431 00:23:25.547 --> 00:23:29.260 They may have written 20,000 lines of just boilerplate, 432 00:23:29.260 --> 00:23:33.189 whereas the entire Haskell program would fit into that same space. 433 00:23:33.189 --> 00:23:37.429 So yeah, it's phenomenal that they can get so much done with so little code. 434 00:23:37.429 --> 00:23:41.795 >> Yeah, I think it's just incredible, so I do think this is a really great article 435 00:23:41.795 --> 00:23:45.605 and I would definitely encourage all of our listeners to check it out. 436 00:23:45.605 --> 00:23:46.911 >> For sure. 437 00:23:46.911 --> 00:23:49.575 >> But do you have anything else you'd like to add on it? 438 00:23:49.575 --> 00:23:53.121 >> I don't think so, I just wanna reiterate that it's great to see another 439 00:23:53.121 --> 00:23:55.969 big company like GitHub using Haskell in production, and 440 00:23:55.969 --> 00:23:58.543 generally coming away with a positive experience. 441 00:23:58.543 --> 00:24:02.372 Obviously, they had some things that they didn't like too much, but overall, 442 00:24:02.372 --> 00:24:04.135 they were really positive about it. 443 00:24:04.135 --> 00:24:06.587 And it seems like if they had to choose it again, they would. 444 00:24:06.587 --> 00:24:11.755 >> Yeah, we'll see the octocat soon with some sweet Haskell tattoos. 445 00:24:11.755 --> 00:24:14.155 >> Yeah, the purely functional octocat. 446 00:24:14.155 --> 00:24:15.185 >> Yeah. 447 00:24:15.185 --> 00:24:16.718 >> [LAUGH] >> Well, awesome, 448 00:24:16.718 --> 00:24:19.098 well Taylor it's been really fun talking. 449 00:24:19.098 --> 00:24:20.619 >> It's been great talking with you Kelvin, 450 00:24:20.619 --> 00:24:21.938 thanks for being on the show with me. 451 00:24:21.938 --> 00:24:23.567 >> Of course, always, any time. 452 00:24:23.567 --> 00:24:27.131 >> And thank you for listening to the Haskell weekly podcast, 453 00:24:27.131 --> 00:24:29.277 this has been episode number 13. 454 00:24:29.277 --> 00:24:33.742 >> If you liked what you heard find out more at our website haskellweekly.news. 455 00:24:33.742 --> 00:24:38.116 Also please take a minute to read and review us on iTunes it helps us out a lot. 456 00:24:38.116 --> 00:24:41.382 I think right now Kelvin and I are the only people that have reviewed us. 457 00:24:42.489 --> 00:24:44.705 >> Out there reviewed. 458 00:24:44.705 --> 00:24:47.212 >> This episode like all the other ones we've done so 459 00:24:47.212 --> 00:24:50.621 far was actually recorded in our own office, the ITProTV studios. 460 00:24:50.621 --> 00:24:55.761 So I just wanna give a shout out to ITPro for recording this whole thing, whoo-hoo. 461 00:24:55.761 --> 00:24:59.781 >> And at ITPro what we do is provide IT training that's both effective and 462 00:24:59.781 --> 00:25:00.726 entertaining. 463 00:25:00.726 --> 00:25:05.793 So if that sound interesting to you please go find out more at ITPro.TV. 464 00:25:05.793 --> 00:25:08.751 Thanks again for listening and we'll see you next week, bye. 465 00:25:08.751 --> 00:25:09.630 >> Adios.