Rust Programming Language Bay Area Rust, 3 Jan 2017

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: Bay Area Rust Meetup September 2014

Description

Topics about Rust.

Help us caption & translate this video!

http://amara.org/v/2Fh9/

A

Joining us, everybody I'm so happy to see you all here. This is the sep tember rust meet up.

A

As you know, this is a web related meet up a continuation of the one we had two months ago today we have two illustrious rust community members talking about web technology, Allen Andra day talking about his oxidized j/s framework, which is a library for doing compression and then the incomparable Chris Morgan, who I'm sure everybody is aware of talking about TP and the future of rust and the web and miscellaneous web things I'm, so excited and I hope you are too so without further ado. Let's get started. Alan come on up.

B

C

Nobody started hold on.

C

C

Cool one second: can you hear me all right, yeah, okay, I hear some ecos. Okay.

C

So I'm connecting to the internet will select guest.

D

C

All right we're ready to roll cool. Thank you, everyone for being here. My name is Allen Andrade and I'm going to talk about oxidized GS, which is a tool to do parallel. Javascript compression ah before I start. These talk I would really like to give some background. Cuz I know, there's a lot of difference in expertise here. So I just you know, want to share I. Remember when I was in high school I started learning about see, I mean that was pretty late and I. Remember that I I stopped when I got two pointers.

C

I was like what fails is right.

E

C

Just like drop that book and then in college I met Ruby every every project I would do in Ruby after that, I started doing ruby on rails and that's what I'm doing right now later on I jumped into JavaScript. I really enjoy the Emperor framework and after all, these things Ross happened and has been amazing. Thank you for that cool. Let's talk about the problem, mmm so um when you have a rails app, it tends to grow like a lot right. So when you deploy your app, it takes a lot of time.

C

Let's assume these rails app. That I'm talking about is using the rails acid pipeline when you deploy. It normally takes like 10 to 15 minutes plus extra time, and that extra time is, you know, uploading your assets somewhere. Whatever.

C

Also, the reason why we want to pursue a faster time in these deployments is because we want a faster release cycle right.

C

Okay, so let's talk about potential potential solution to this I want to cut down the 15 20 minutes deployed down to five hits its path possible 10 if it's possible all right. Let's unwrap that looks paralyzed with Bryce. That's the first thing that I thought: I was learning, Ross I mean I'm, still, learning I mean and I thought I mean it would be amazing if I can solve a problem and also teach myself. Some rust and I didn't know how to start.

C

How do you proud paralyzed, you know acid compressions is so starting from the from from the fundamentals. How does a tool that we currently use compressed and minified javascript? So the first thing that I, the first thing that was natural to me, was to use Ross and all without you guess, which is the mini fire compressor that probably the best one out there.

C

Let's talk a little bit about all the fiber, yes, I'll be five years. It's a it's, a compressor mini fire, as you can see right there and it consists of different parts. Aqua vitae jizz has a parser, so that means it will yield some AST.

C

Also, it has a code generator which will cried some JavaScript from the ASD. mmm We have the compressor, which is it uses a transformer API, which is something that we will see on the next bullet point and it optimizes it that it optimizes AST into smaller one, a Mangler which is it gets it reads: it gets read of variables that you can use it compresses your variable names to the minimum possible.

C

We have a scope analyzer which augments the AST. I really don't know what that means. You know it says it augments, the ASC with information about where variables are defined and reference, probably people that work with parses parsers know pretty much. What's that right, I, don't know tree Walker, so very simple, you to an API so that you can walk the ASD and then we have the tree transformer, which is the API that the compressor uses.

C

So that's a no Verve you folk aqua, fiber. Yes, next up I want to talk about the architecture, so we have manifest manifesto or whatever you call it any. It is in a JSON file and this manifests the Ross code. What it will do, it will separate separate your files into chunks evenly distributed among your cpu cores so that you distribute the work load of compressing into your color course cpu cores after we have these chunks of paths.

C

As you can see there, we use ugly fat, yes and we use those back those paths to feed the aquifers process. At that point, we run ugly five years only with a compress option. We do not mangle right there. After we aqua Phi, we use channels to synchronize the writing of that of the output of ugly fat.

C

Yes, otherwise we would get like gibberish and overlapped outputs in in file that right now, I don't know if it's good to put in a file, probably better to keep it in memory and that's one of the things that I'm still stuck with I, don't know what will be the best option.

C

So after we Olga fire, we use the channels we on. If I unify in these temporary JavaScript and after we unify it, we run all the fly again with a mango option.

C

Finally, these discipline five years will yield the result of oxid I oxidize years, which is a final product there. There are a couple problems that I see with these implementation, and these, like the you know, the first idea that I got about aquafine and how to make it sorry about oxidizer, yes and how to make it parallel. So one of the things is, I feel we we lose a lot of time in in parsing, right, I'll, go if I parses and then generates JavaScript and then spit it spits it out.

C

Then we parse it again and we do the same thing. That's really wasteful. It sounds to me.

C

Okay, let's talk a little bit about the results. Let's try that let's try that I, don't think this is still a reliable architecture so that we can, you know, make yeah. This is awesome or a do project or we have good improvements. However, when I got a version working of these architecture, I did a simple example of.

C

C

um That's a pretty quick road map or a pretty quick explanation about what oxidized g is does and, let's think about. What's next for this project, irem I would really like to know if we are on the right path. He's Eggman script coming out with new standards that do not require compressing.

C

Other or other ways of doing these same tasks, I still don't know in my and the knowledge I have about that regards is reduced, at least at this point also, I was I was in a previous meet up here. I was talking to somebody a reduced which is not here today. um Well, that's the higher cna. I don't remember his real name, uh but he was talking about wow. Ok, so you will pass it and then play with that. You see and I was like yeah.

C

That's what ugly five years does, but I don't know like, should I ride the parser for javascript in rows. That sounds like somebody else have done it before right, I, don't know and I mean I thought about like I c or c++ implementation, so that we can it's pretty easy to talked to talk with with a C or C++ interface right. So I don't know why. I cannot find that. I was talking thinking about okay, so um spider monkey is that is that number, the name?

C

Sorry I was okay, spider monkey should have it and I don't know. I cannot find that thing that parser, that at least can give me some AC um I talked about that already and then, if it happens that we find the c or c++ implementation can we feed uglify with these asd in a javascript representation, so that you know we don't have to rewrite what up with ideas does over it and and the last question would be who we compress minify pars and do everything ourselves, I don't know.

C

So if there is anybody out there writing a JavaScript, parser I would really really like to help with that, and let's see how that goes.

C

That's pretty much it about oxidized Gia's, it's a pretty small project and it has been amazing to work with with with Ross on it before I, before I grabbed obvious. More presentation. I would really like to say thanks to see planet club Nick, which I mean who wrote the Ross for ruby, skye that that's the guy that brought me into this community and after after that guide, I was hooked and I. Cannot.

C

You know I'm at work, thinking about Russ, so I don't know if that's good, I love this community and I'm super happy to see people joining new people every time joining to these meetups keep doing that and I'm pretty sure you alot of these community also IRC people. There are super helpful. Thank you, everybody in the IRC channel. I'm probably if you remember me it's because I was bugging you really stupid questions and also about the Bay Area Russ hack nights, I organized that uh that uh that meet up and we've all.

C

We have only had one meetup ever that happened at github and was was pretty good. I discovered that they that hack nights are not great, because people you know are tired, so I was thinking about moving it to another different day, probably saturday, and also I'm interested in switching the you know. The hack day is not hack night anymore or the hack day in San Francisco than one in the east bay. You know like switching every week or every two weeks. So if anybody is up, you know willing to help on that I'm.

C

Looking for somebody that is in San, Francisco I have the East Bay already covering.

C

Thank you very much for listening for your attention.

C

Do you have any comments or questions.

F

Here whether you know, if you're just trying to reduce the amount of information you have to send down the pipe to the browser you could get over, you could do that by simple. You know, zipping or something of the data, so is, are you and if you want to minify this stuff, you could do that as a separate step to compression so I guess what I'm wondering about is? Did you try as a baseline test, for instance, saying okay?

F

How long does it take to to to simply zip the files, and how does that compare and- and you know you could also probably do some kind of parallel zip to to get the parallelism as well.

C

Yeah these problem or up I guess the solution is uh I.

C

Guess is strictly targeted to rails users, and that's that's weak it like when you, when you deploy a rails app, it will try to compress minify and give you one JavaScript that it's already like minified to it fullest and then, if you want you can use if that there, the thing that I'm trying to get faster is the process of minifying that JavaScript. So my understanding to your question is I. Don't think I am solving the same problem.

C

Your are you know asking about I think, that's that that's solvable, there's a different solution for the post, deploy and I am in the face of pre deploy deploying stage. Does that make sense.

A

C

Well, yes, that's it! Thank you very much.

A

Alan that was so amazing. The last trust hack night was a super awesome and I can't wait to the next one. Alright next is Chris Morgan talking about all kinds of stuff related to the web and rust come on up Chris.

G

There well speciously, entitled miscellaneous web things. This talk for any of you that have gone watching. The video of my talk at strange loop is substantially the same as far as the slides are concerned, but I will be taking a much more strongly rust focus and there are a few extra goodies here as well.

G

Okay, first of all, I suppose I should have just a little bit of background about me. Yeah I, am a web developer by trade. I've been interested in the web since probably about the age of 11 that took me ages to understand, CSS and but I've been working with. Python is my preferred language for about six years, and I've been using rust for the last year and a year and a quarter or so and rust has been absolutely spoiling Python for me.

G

Okay, we had this model that everyone follows. You have a certain amount of time in which you wish to get a product to some people, and you will make money from that. That's the commercial model that we have this time that we have is very limited because otherwise, you'll run out of money and also with venture capitalists are present. They don't seem to mind whether you make money or not so this whole time thing it's a bomb. I.

G

Will be talking substantially and specifically about web servers here, client-side stuff, I do have a vision for where rust can fit in in the client side as well, but for the moment, let's be satisfied with server side because well the front-end guys they're stuck with javascript anyway, for the moment at least okay. So who among here, is not a practicing web developer?

G

Let's say about a third of you: people using dynamic, fully dynamic languages.

G

Well so I mean for professional web development yep there, that's probably around about half a bit more people using the m based languages professionally, eight, so that's will call it about a quarter. People using compiled languages. Note I'm saying nothing about the type systems here go and see their type systems are rather weak, so compiled languages. I see four hands, so that's the fewest of the lot, and yet this is where rust fits in, incidentally, as the coloring of the logo ongoing debate as to whether that should just be plain black.

G

If you know the stickers that brian has back there they're all black, the colored version of the logo may be on the way out. Anyway, so we have this situation where most people are using the dynamic languages.

G

So they use these dynamic languages. Why? Because they think that they can use them to produce the result on this ticking time bomb the fastest, and so they use that to get the results to the people and make money. The problem is that this causes various bugs and so on, and it's also slow. So you end up with needing lots and lots of machines, and this all ends up causing problems and they don't end up making money out of it.

G

Simplification. Of course, we then look at what people do after this step. You look at say: Twitter, for example, they say: we've got this problem. We we want to get rid of some of the bugs that we have. We want to not need so many machines and we want to make money. How do we do this? Well, the solution that they used was get rid of Ruby, which they had been using and use Scala. Instead, of course, it's an open question: why aren't they making money still but never mind?

G

Okay, so we had this whole performance aspect of it, because yeah performance is likely to be important to someone. That's using rust for web development.

G

Initially these dynamic languages they're very appealing because they had they don't have this compile time. However, they do have development server, startup time, which is frequently 5-10 seconds 30 seconds, even once you're getting to a big project and once it's all started, the performance is slow.

G

However, with rust, once you've got compilation /, which yeah it takes some time. You know it's really really nice having a server that you press the enter key and nothing sir. Oh yeah, it forgot to put in a print message: it's already started. The server before I had even finished, lift my finger from the enter key. You really can have the entire HTTP stack, starting in just a few milliseconds, and once it's all started, it is fast.

G

So we get them these sorts of graphs, this being a new relic thing that large pale blue chunk there, which is taking in excess of 300 milliseconds there. That's all the Ruby code. Of course, this was an example where people have been doing bad things that are computationally intensive, and so in that talk, New Relic were demonstrating how to get that considerably smaller, but still they're. Only looking at cutting down to 50 or 100 milliseconds in rust I would be confident that that stuff you'd be looking at more one or two milliseconds.

G

It really is that much faster. You can reduce the costs a lot and, as everyone knows, especially in the e-commerce world response time is actually connected with revenue.

G

People then complain, but this compilation phase it's a real problem, so we shouldn't use rust because it has this compilation phase. Of course, that's not that's merely an excuse. Everyone has their test suites, which run really really slowly in the rails.

G

World I understand it's perfectly normal to have test suites that may take 8 20 minutes to run, whereas in rust, you're less likely to have tests that take that long to run, or maybe you will our package manager they're very good at taking a long time as well npm under any of these things, they tend to do a lot more work than they need to for a lot of things. Of course, we're not going to escape from that with cargo either so, but never mind we.

G

The problem is people use these excuses, but really they are just excuses. We'll come up with ways of making us things slow. However, we do it we're very good at doing things inefficiently, but then this compilation phase what is rust doing in there time when it's compiling stuff that people complaining about I would argue that it is saving you time, because there are various classes of errors that it's picking up on, which you have made. It may not be saving your time up front.

G

It certainly does give the impression of making things take longer, but in the long term, I believe that it is actually faster. Let's take a completely synthetic made-up example, which just happens to show that rust is better, so well, yes, I won't bother reading it all out there. You can look at it, but the the end result even accounting for this two hours of hallways sword. Fight without a legitimate reason is that rust is a full minute faster.

G

Isn't that marvelous, but then, who are the people that are saved time, because I'm claiming that rust is rust, has good performance in as fast for more people more than just runtime performance users, the response time in whatever they're using is decreased, so the and there is that inverse correlation between response time and revenue and happiness as well. People are likely to be happier if it's faster.

G

There are also very interesting questions of ethics. There I can talk more about that later on, if you're interested in it developers maintenance. Once you have the type system of rust, you can use it to avoid various problems where, before you would have had a complete start up, your development server run your tests or whatever find something's gone wrong, and you also in dynamic languages, often end up writing tests, which takes quite a long time for things that the compiler would have caught in the first place.

G

So you in rust, you simply wouldn't have needed to write the tests, so fewer users get annoyed if you have fewer bugs in your system, and so you spend less time dealing with your customers and software developers, don't like dealing with customers. Do they.

G

Also in operations- and yes, servers you'll end up with lower resource usage. To take the example of Twitter, when they were merely switching from Ruby to scala and rust would have been even further along that way. They were expecting performance improvements of about a factor of 10 by the change they actually found that they got performance improvements of about 50 there, so it was considerably better than they'd expected. So that means that for every 50 machines that they had before, they can reduce to just adding one.

G

Of course, those sorts of calculations only really matter if you're dealing at significant scale, but never mind, they're still worthwhile considering, and so that naturally saves you money as well. We're used to the saying that developer time is more important than machine time, but especially once you're working at a big company that can actually pretty quickly become not true and also a lot due to this the type system there.

G

The security is improved because there are various classes of problems that simply can't be represented in the type system, and so you end up having to spend less time dealing with security vulnerabilities telling people about it, loss of reputation all of these sorts of problems. So now, let's consider why you need this good type system. I'll. Take us the case, study, HTTP, headers and I'll be looking specifically at how tp does it actually HTTP headers their string, lee typed in almost every library out there?

G

There are a few that aren't, but this very significant majority of ones and practically all the ones that are actually used significantly a string lee typed. This is wrong. Headers are not strings semantically, they are not strings. So this makes me sad. This was in fact the reason why I started writing. Rusts HTTP libraries when I first started writing rust code, because I was afraid that if I left someone else to write the libraries that would go string typing take for example, then this the do not track header.

G

So it's a very simple header: it has a value 0 or 1 in a language like Python the code block at the top there.

G

It's very simple thing: you have a dictionary as it would be in Python keys are strings values of strings, the value might not be there, and so that's how you track your three states there, but with the with the way things are arranged in TP, we can have it all tied into the type system. So the very fact that we are passing in this value DNT.

G

That is enough to specify what the return type of that method will be, and so we have a semantically correct header type there, and so it is either some opt in some opt out or none.

G

This, then, is the implementation of that particular header type. It's as you can see a pretty simple sort of thing, very basic enum you go specifying how the header is paused and how it's so past from a stream. Actually, it's not a stream at a slice of vectors there and that's because it's already come out of the stream, because these are all done lazily for various reasons which I can talk about later and also how the header is then put back into your stream.

G

Some people at strange loop were wondering why these are so verbose. They wouldn't actually need to be quite server birth, so quite so verbose. If we didn't care about performance, but as it is caring about performance say you want to use a writer on the on the formatting method, rather than taking it to an intermediate string. So it ends up slightly more verbose that way, and that macro there. That comes down to something like that. As a matter of fact, so you have that associated types stuff there, it's actually not using associated types yeah.

G

At present it is using generics there's the reason. I can't use associated types yet is because of this business here. So we had the constraint. We would need to apply a constraint there, which can't be represented yet, but that will come so this, then, is the overall stuff there. We have the trait the basic marker traits are the DNT type here, which is used as the parameter to the get method and that so, when we pass in there as the header market DNT, it then says well, the return type will be the returns.

G

I will be its header type, which was that enum, and so you can very neatly encode into the type system, this notion of strongly typed headers.

G

The really good thing is that it simply can't go wrong there, because it is all it's all in safe code, actually I think that there might be a little bit of unsafe, curd there for performance reasons, but it can be represented entirely and safe code.

G

Now I won't, I won't lie at all. This is something that you could do in dynamic languages, but it would be hazardous and the full extent of it would not be neat to do in a dynamic language. The reason I say it would be hazardous is because you don't have the type system to catch you when you make a mistake with it all, and so this type system it allows you to do things like this, that people are just normally too scared to do it.

G

Lets you do these things, but be confident that it is actually correct and also with the way that it's implemented in TP there. There are no privileged header types, you can have multiple headers using the same name and it will decode the raw value as the appropriate type. Whenever you ask for it so say if you have, if you had two different libraries with their own, do not track header types, that's no problem, it can cut with that, and that is actually a part that would be harder to do in a dynamic language.

G

Just like to re-emphasize headers are not streams oak there on that good. Let's then, take just a quick look at an example of some of the sorts of things that will be possible in rust. I haven't gone implementing the significant majority of this stuff and note that the / web- it's dealing with the full web stack so client side as well as server side.

G

/ / web is a language in the Haskell and ml traditions. So it is thus and somewhat unreadable to the layman, who is used to these languages like Ruby, Python JavaScript, and that's why it's going to fail. Well, that's why it has already failed and that's why I believe that rust has a much better chance. Rust is pragmatic, rust is familiar, and so, although its type system is not necessarily as good as some of these things say, Haskell it has a better type system than rust, but rust is approachable.

G

Rust is something that the layman can use, but let's go back to looking at some of these things, so code injection attacks. They include in that in HTML injection sequel injection. Any of those sorts of things that you can imagine by using a strong type system. It can simply avoid these problems. Invalid HTML.

G

G

Yeah, they are all the sorts of things that we can represent and rust. Take, for example, I want to see someone experimenting with compiling templates, semantically and using the power of the type system, something like this so inside here we have a this.a, the value there. It's just a plain string. It will need to be HTML escaped, be it will need to be HTML escaped, but then again what?

G

If it's, what if it is already HTML, it might be HTML, so you're going to need to accept something which is declared to be safe, HTML or not see.

G

This is the place where all your escaping systems in Django rails. This is the place where they fall apart, because almost invariably, the only type of escaping that they know about, is HTML escaping once so in c in location, see there what we're expecting there is actually a URL. It would be wrong, semantically incorrect for it to be anything other than a URL and the escaping rules are different. If you go applying HTML escaping rules there, it will break in D as well.

G

If you go applying HTML escaping rules, it will break and what's more something that will be passed by it as perfectly clean HTML will be bad JavaScript. That's where you get those injection things. We have all this all the ability to have this semantic information which were just ignoring, and that's not a good thing. The auto escape that existing frameworks have is simply not enough for meaningful security. So these sorts of ideas. These are things that I would like to see. People experimenting with see what you can do.

G

You can get similar sorts of things with sequel, these security, bugs that we have here. They simply don't need to exist in a language like Python or Ruby or JavaScript. They are comparatively unavoidable in a language like most of these security, bugs they simply don't need to exist. We need to get rid of them with our type system, there's an article that I have linked they're, structurally fixing injection bugs it goes into these topics in some more detail.

G

So we have a type system. Let's use it for correctness, because correctness is a very important thing as well, but we've also got to remember rust. Its goal is pragmatism. It is if we want these ideas to permeate through web development in general. We need to embrace the pragmatism. We need to realize that we've got to go for something that people can use, even if it might not be quite as good as we could achieve. If we wanted to go carefully into all the theory things rust is not a skill.

G

We do not have full loan ads and I'm happy for us not to have monads for the moment. Eventually we can have monads, but for the moment monads would be very bad marketing for us. I.

G

Simply don't have time here they go into all the good stuff in rust, rust ownership model, it's something that I've been sorely missing when I work in Python, because you you end up just not being able to reason objectively about something that I've grown used to being held a reason objectively about using the type system for errors rather than exceptions, and so on. It's a very nice thing to have.

G

We've also seen how r us can integrate with dynamic languages say the quick version of that is you look at what you Hooda cats and Dooley have done with skylight, where the majority of their product there for tracking rails applications is written in rust because they decided rust was simply the best language that they could find for it, and there are many other great things about rust as well.

G

So the summary of all this is, if you want to get rid of all your bugs all your bugs, like a most of your bags, lots of your bags, some of your bugs a few of your bugs any one or two. If you want to get rid of some sorry, some of those things, if you are to reduce the number of machines, you need to make things faster, you make money, make your users happy, throw away your dynamic languages, use rust for web development, a.

G

Good quote here from the Bible, of course, it is not reasonable to expect you to just throw away all your Python Ruby code bases you will need to keep them going. You may and find that in your experimentation with rust things, don't turn out so well.

G

You find that some of the increased verbosity some of the increased memory capacity that you've got to deal with I'm in your own memory capacity, not computers, because you'll find memory usage goes down a lot too, but just some of these things, rust is conceptually more complex than these other languages. Make no mistake of that so yeah keep going with these other things. Try rust out as well, because you may find that one of them suits your purpose is better than the other.

G

Now briefly, just about TP and a month ago, I was here for the last meet up and then on the following monday. I went cycling off into the distance on my recumbent tricycle towards st. Louis for strange loop, I made it I actually only made it as far as Lucerne Valley and before I had a mishap which resulted in a we.

G

All being bent and I straightened it enough that I could continue but decided didn't quite trust it enough to go continue going through the desert, where there was no one around so I went back to the nearest town. Hired a car went via the Grand Canyon for a couple of days, but during all that time, I found myself simply disinclined to do anything to do with computers, which I haven't felt for many many years, who's very interesting, so yes, TP and so on. It all just stalled while I had this holiday.

G

It needs to get past this stage. So, but there are some pretty tricky questions to do with how precisely pausing in particular should be handled. I would be interested to go talking with various people about afterwards, perhaps at the table over there, or something like that. If you're interested in helping out with that feel free to come over and I, think that that's it is anyone have any questions.

G

Complaints, yeah, oh sorry, yeah should have I should have specified what TP was and so. First of all, we had rust, HTTP, the HTTP library, and, but it was largely experimental. I was figuring rust out as I went there and got to the stage where it need to needed to be replaced. So TP is the redesigning of the designing of an HTTP library from the ground up, thus naturally prone to second system syndrome, and all that.

G

Anyone else hang on IRC con talking to me. No one talks to me. Oh.

E

D

F

Yeah I'll get a question about the rush. Compiler um I agree with you that it's nice to have a system which will do type checking for you and make your empties for you. On the other hand, it's also nice to have a system which is responsive for development like a ruffle or or like a fast interpreter.

F

So I guess my question is I, don't know anything about the internals of roughs compiler. Would it be possible to build something which ran quickly but did not make all those tests, and then you could use that for development and every so often like the testing, people do say. Okay now now now run the full set of checks to see if I'm doing any stupid to the.

G

Extent that you say no and to a more limited extent, I hope, so I'm just waiting for someone else to do it. Yeah there are various things that can be done to improve performance in those sorts of things, say, background compilation, that's something that can be done yeah also, if you do turn off all optimizations.

G

There are various other tricks that you can do to speed things up somewhat, but still certainly compile time. Performance I would say, is just about the weakest point of a whole lot, but that's acknowledged to be a problem with rust as a whole, so there will certainly be ongoing work towards mitigating those problems.

G

B

H

B

I guess the traditional difference between dynamic languages like Python or Java Script and static compiled languages, was when you're prototyping in a language and you're, sometimes you're messing with the the data model and you're you're, changing it a lot a lot of times in traditional languages like C or C++, it's a little cumbersome to make changes like that. There's it's a because they're, verbose and they're part of a lot of changes in different areas. How does a out of busts kind of mitigate this? What.

D

G

We'll find out with something with some of these lots of things. I am hopeful that they will work well. A lot of it will come down to the design of any framework making so making it designed to cope with these sorts of things as welders can be done, but these are points where it may not turn out as well as I expect it to that's. Why I give this advice, keep going with your dynamic development, but try out Russell, try and see what a good type system can do. Did someone speak to me.

G

Hey, that's a good point to that. We have. There is parallel combination now. So, if, if you go splitting things into smaller crates, they can be built there, they can be built in parallel, so some of these problems are reduced by parallelization.

I

How does it compare to see code for thats a fairly I guess, there's that many large code bodies in rust, but it one.

B

Of the things of.

I

C++ that I noticed is that, as C++ projects grow, the compilation in link time grows dramatically, and this really impacts that whole debugging make a change, compile, run a test feedback cycle for developers.

F

I

This is actually one of the things that make me like see again is being able to quickly like a change in the code and see it running quickly and Scala is another language. It has that problem. Very bad was Rihan.

G

Analysis, color is reckoned to have really really bad compile time problems I think that I get the impression that it.

G

I

F

I

Just turning off the optimizations makes everything a lot faster. That's actually fine for the.

G

Development, oh yes,.

I

Sudden, but if it's just inherently painfully slow, that's a problem, a.

G

Significant the second event of madrone.

I

And that things like Linux are huge C code bases that are still successful. Is it's not that painful to built that word C++? They would probably take several times longer. It's.

G

It's worth bearing in mind with them, so with see the standard model that you'll have actually I should change this black, so black I meant black, not blue B is blue rather than black k is black. The the sea model is to have you have lots of different files, and you then go you go building each of these files individually.

G

So one two three four five: six, if you change just one, you might be able to do that and then you go to link them all with rust or no with C++ is own, but with a lot of these things, you end up with the big monolithic model, where you just have one thing that needs building splitting things into crates can mitigate that a lot so that you are working with smaller boundaries, so they are just back to working with ya.

G

Oh capers scales don't work now, but you there is not so much inherently good about see, as you might think, there are various things in rust that can and will be improved over time where there simply hasn't been time to experiment with improving them. Yet because at present there is one rust compiler of any seriousness at all, but well how many C compilers are there about there?

G

There has been an enormous amount of work into optimizing that these are things which I, certainly hope will be in be improved significantly and rust, as it gets more popular and yeah I'm backing rust because I believe it I believe it is most likely to succeed.

G

And someone say something in here.

G

Yes, my concerns for passing the plaster there is this scary thing in HTTP, which has a number two in it. Http two is rather difficult to manage because with HTTP to you, go implementing full multiplexing and dependency management and price stream priorities all over the one stream managing that in the same way, which that and having it map onto some task model or whatever that still keeping it flexible is hard deciding. Also how precisely to balance performance with correctness.

G

That's a fairly tricky thing as well say: Giants, HTTP parser, which is used by node engine X, and so on. It's a monstrosity which needs to die. It is a really really horrible piece of code. It's a marvel that it works at all its head. In my first read-through of the code, I found at least two rather severe bugs it's not a nice thing. What it's worth there.

E

Are 30 forcing library used to not fully parse the request method? So if you all.

G

E

Qulet parts of the post, hmm I think it's been textin, but that.

G

Was one of the ones that I picked up on my first rate through where it simply wasn't checking things properly, but yet deciding whether to go for an iteration based model or call backs or whatever there are or phantom types. There are various approaches that can be used with pros and cons. Then yeah I have a couple of prototypes. That I can show people afterwards, but John rim won't be here. For that sorry, John at least John. You can know that some of these things are things that I discussed vaguely with you a month ago.

F

And one of the motivations for peugeot was dealing with the long wait times that Google was having it million line systems and so I guess I've got two questions and one is do those kinds of how do those problems play out in the rust implementation and can rust steal any of the ideas that go uses to mitigate those and.

G

You run from the rust team who actually knows about this stuff, want to do something say something about it because I don't know much about these errors of it. Take take a microphone or come here or something yeah I.

A

Don't know much about the link strategy that go uses, but one of Russ.

E

A

Values is to integrate with the system tools, so, in that respect, rust wants to use the system linker. So, unlike we don't want to write our own link here, we just want to use gold o LD that comes with the system, so we kind of beholden to whatever the link your performance is by default.

D

All these weird header files you have on this guy parsing, it's more just actual surfing, you might think matter- should be doing six and so I know how that works. The scale it's not properly going condemn us wit met with microphones.

G

The other question is from.

H

Thank you for a really entertaining talk. I was crazy, very hurt.

G

That microphone is.

H

Not tonight.

G

H

H

G

See I came to rust wanting to write a web framework because I determined that the language was the only one that was even decently suited for the goal. The dream framework that I had pictured.

G

The problem was that well, there was no HTTP library there, so I've been busying, busying myself shaving that yak first.

I

Yours is the only one right now, no.

G

There are, there are a couple of other one's name: conduite has been doing something in the area and hyper as well. I'm, not sure on the detail was a piper at all. I I gather that he popped up while I was away mostly and conduit. I haven't looked much at at all.

G

They're, probably popping up at least to some extent, because I've been I haven't been getting on and finishing tp2 a user book state because the president, you can't use it at all, because it doesn't have any of the stuff to actually handle streams.

A

So I've heard t be described as rusts first vaporware.

I

Okay, so is it going to ship.

G

It's one of these things where whatever I am so well: okay, if it's gonna be vapor where I'm going to say yes anyway, under its.

D

G

I am intending to be getting back to working on it, but intent. What is intent.

G

But dear, we can talk about it further afterwards. I certainly want to get it to something, that's usable and by the way, if anyone happens to want to hire me for working on rust stuff rather than Python stuff, I'd love, it.

G

Or sponsor TP development yeah anyway, any other questions.

G

No one seems to be saying anything.

A

Thank you, Chris I was super entertaining.

A

Do you mind if I say a few things about some of the issues that just came up there Bryce on your question about using HTTP and production? I wanted to note that the cargo has a centralized repository that is web-based and it's not online yet, but it's going to come online, the next month or so. This is going to be it's almost entirely rust and it's going to be exposed to the web.

A

It's going to be rushed doing the web in production for the first time that I'm aware of and it's going to be scary, but it should be rock solid.

A

Really I wasn't even aware of that. That is crazy on the topic of compile, speed, I thought, maybe I could inform you about some of the developments that are happening there and that might be expected in the future.

A

Everybody knows this is kind of Russ Achilles heel that we're slow to compile in general, rust compilation is divided by two phases: the analysis which is all rust and optimization and cogeneration result lvm and general compilation is dominated by llvm recently, a parallel code generation patch landed, which takes some the lvm step and divides it into some number of threads. This is not on by default, but it can provide speed.

A

Ups, the same off that did that Stewart perm Steiner has an incremental compilation patch, which is supposed to specifically address the development cycle, so it only recompile things that have changed, and it does kind of interprocedural analysis to just figure out exactly the code. This change, and only recoge in that, and that you provide a lot better experience that the dynamic language of are familiar with beyond that gets a lot harder, though, to get improvements. What you'll see probably past that is the next lowest hanging fruit is simply improving.

A

Russ I, our generation, the code that we're feeding to the lmm back end right now, we're kind of notorious for the poor quality of IR that we send llvm. We really lean heavily on llvm to do optimizations and kind of perform optimization miracles to make the code fast. If you look at clangs output versus a Russ output, clangs ir looks really nice. It's treating llvm with a lot of respect, russ just dumped garbage on llvm and expects us to perform miracles. So some low-hanging fruit there and that's probably the next place- will go beyond that.

A

It's probably to take an entirely new back in specifically designed for performance to make rust really fast a compilation time, and there are a few people that have ideas about that, but that's kind of a multi-year project, but I do want to emphasize that the compilation performance is not as dire as it may seem right now, there's a lot to do there and there's a lot that can be done.

A

uh Finally, that was awesome, Chris and awesome. Andre Allen sorry I want to mention that uh there's a whole bunch of going on in the rest community right now, a lot of events around the world. If you go to the rust website, there's a new link that goes to the community calendar.

A

There are regular events in both Paris and Seattle, there's also starting to be some spun up in Tokyo and there's kind of talks at conferences periodically throughout the year lots going on, let's going what's happening, especially as we ramp up the one point, though it's very exciting, so keep an eye on that. Thank you, everybody for coming. It's really been a pleasure. Thank you good night.

A