Bonus Interview- Bjarne Stroustrup - C++
How do we make sure that the space probes don't get the logical equivalent of the blue screen of death halfway to Mars, where we can't send a repairman? If I'm putting up a small application for my own use, or if somebody is trying to push out a little cute web app, they don't need that kind of reliability and maybe what I'm talking about, in terms of programming, is not for them. On the other hand, if you took the attitude of first to market is the only thing that happens and apply them to the steering wheel of my car, no, that's not a good idea.
- زمان مطالعه 15 دقیقه
- سطح متوسط
دانلود اپلیکیشن «زوم»
این درس را میتوانید به بهترین شکل و با امکانات عالی در اپلیکیشن «زوم» بخوانید
متن انگلیسی درس
I was trying to characterize, what is C++ particularly useful for, what did I actually build it for? And, what is its strength today after almost 30 years of evolution? And also where’s the limits to where it is. So, the way I see it is there’s a core domain for C++ and that’s what you would call traditional systems programming, but that’s not the right term, because it’s just a style of language and a style of programming. So, I go a little bit further. I said, what is it that requires the kind of services that C++ has? What is the sum of all these applications I’ve been dealing with? Where did it work and where was it essential? And I came up with a phrase infrastructure and I roughly define it as if it breaks, somebody gets hurt or somebody gets ruined. These are the kind of foundational things in our systems that must work for the system, for society to work. The question I’m trying to phrase is what matters in those areas and I’ve come up with some notions of compact data structures, very strongly typed interfaces for maintainability and for minimizing errors, heavier emphasis on algorithms or random code because we need the reliability, we need the stuff to be comprehensible, analyzable. We want to make sure that it’s actually correct and so the paper I was writing, come from that kind of thinking. What is the right style and what are the supports for that style we need for infrastructure software, or for software that must be dependable? And we can get real examples. Sort of, the keys of some of the modern operating systems, the basics of our phone system, the brakes in my car. How do we make that dependable? How do we make sure that the space probes don’t get the logical equivalent of the blue screen of death halfway to Mars, where we can’t send a repairman? How do we make sure that they actually go into the right orbits? JPL lost a probe of Mars because two groups had communicated nicely, they thought, but in fact one of them spoke Imperial measures and the other spoke the SI system, the MKS system, and the result was a misnavigation that sent more than $500 million worth of equipment into the wrong orbit. Not a good idea, it was the work of 200 good engineers, lifetime’s work down the drain. And that could’ve been avoided by ever so slightly improvement in the interfaces between the parts of that program. So, things like that I’m interested in. So, there’s a core area where I think the facilities for C++, the kind of, they are not perfect. The kind of things that I would like. The kind of things that I work on are essential. And then there is huge gray area where you have choices, they can help in my opinion but they’re not essential. Then you have areas where it probably is unsuitable to apply that kind of stringency and technique. I mean, I want to use essentially get 100% reliability. If I’m putting up a small application for my own use, or if somebody is trying to push out a little cute web app, they don’t need that kind of reliability and maybe what I’m talking about, in terms of programming, is not for them. But I think the really important thing here is to realize that there are different techniques, different languages, that apply to different areas. And we have to recognize this. We can’t have a single language for everybody, a single technique of using that language for everybody. We can’t have a single tool chain or a single kind of system. From there we can go a little bit further. We probably don’t need the same kind of training for education for everybody, The engineer, software developer, or whatever, that’s building infrastructure, say the mechanism that automatically updates the software on a cell phone, has to have a different education, different knowledge, different training, from the one that makes a little game. Because the one can actually destroy a whole day or a whole week for millions of people by a little slip. Maybe even somebody gets hurt if the 911 calls don’t get through. Bad things happen. And it’s not just the software that runs on it. It’s the update software. It’s anything in the chain of safety-critical issues that has to be dealt with. But somebody doing that has to think differently from somebody who writes a little application maybe to make a couple of bucks quickly. There’s nothing wrong with that, but they have to think different. As a matter of fact, if you apply this very stringent engineering-oriented thinking to little commercial apps, you will probably be a year late to market and it will be irrelevant. On the other hand, if you took the attitude of first to market is the only thing that happens and apply them to the steering wheel of my car, no, that’s not a good idea. These people have to think different and the way you get people to think differently is to give them different educations. We don’t have an ANSI standard programmer and we shouldn’t have. If it was, we should have several. And I think the field has to in some sense clean up its act before somebody else comes in and try and cleans it up for us. This notion that you can sort of make your own constants and have seconds after the constants, is that a feature that you added at some intermediate stage or is that simple, sort of, operator overloading? We were observing that we were getting a whole slew of little suffixes out of the various fundamental types, so the use of X is for unsigned and L is for long, I think. And I can’t remember them all anymore, and I thought, you know, you can do anything in C++ but you can’t make your own literals. And then I’d observed separately that there was techniques that were effective except that people would not use them because the notation was too ugly. The Unix example is one of those. You’ve been able to do everything in the Unix example I showed. Not quite as elegant, but you can do everything showed for the last ten years. Libraries have been available. There was a nice one from Fermilabs, back to nice places and nice people again. But it wasn’t used as much as it should because the users didn’t believe in the notation. They didn’t like it. And so we were looking at how can you basically clean up people’s source code. How can we make code look the way it would look in an ideal language? How can we make code look much as it does in the textbooks? So the Unix example is simply a way of getting your code to look the way equations look in your physics textbook. We know how to avoid that Mars Climate Orbiter problem. Everybody had been taught it in physics in high school. First make sure your units match, then do the detailed calculation. So why didn’t people do it? It was too cumbersome and it was too ugly when they did it. Or too costly if they used runtime techniques. So I, together with friends, we thought this problem was worth addressing. Tried to figure out what solutions we had. Went through several evolutions. I think the last finishing touches was done by David. And it’s now standard. This is one of the features that is not shipping widely today. But wait a year and that example will probably run on your computer too. Are there other kind of examples where you’ve let people from within a class create their own literals beyond sort of suffixes? When I started out with C++ I provided constructors which allows you to construct objects of a certain type from arguments, and that has been very effective, and people have used constructors as if they were literals but they weren’t. There was a runtime cost. So the first thing we did with C++0X, C++11, was we introduced constant expressions as a more fundamental unit. This was work between me and my colleague Gabriel Dos Reis. And we have contextual functions that can be evaluated at compile time. And we have contextual types so that you can use type-rich programming at compile time. This is very important in sort of the high-performance computing and in the embedded systems world. That was what we did to address that. Constant expression evaluation is much more general, much easier to use, and yes more pretty in C++11 than in earlier versions. So that just strengthens what direction we’ve gone before. So you could write complex 1 comma 2 to make a complex number in C++ in 1984. Today you can write the same thing and have the complex number created at compile time and therefore, say, put in ROM. I don’t know why you would want a complex number in ROM, but you might want a point, which is the same thing, point of 1 comma 2. Now you still have to write point or complex and the other thread in, of thinking in C++0X was to generalize and make safer the initialization. And so it happens that if you know the type you need, like you have a function returning a complex number, you can simply write open curly 1 comma 2, and it says, oh, 1 comma 2 is supposed to make you a complex number, and it will make a complex number and return it. And, if you happen to be at compile time, it will do that at compile time, so things work together, you get better notation, you get better performance, and anything you can do at compile time works even better in a concurrent system, because you can’t get a race condition on a constant. If it’s been calculated before the program starts, you can’t get the threading problems. So open-curly means ind a constructor for the thing that I’m about to put this in. Yes. It looks at where the destination is. Yes. And it says, is there a two-parameter constructor or a three-parameter constructor, or whatever? Or if it’s just a struct, it’ll take the first element and put it in the first element, and the second in the second. It does it for a struct as well? Oh yeah. See I should, I haven’t played in C++ in a while. That’s a great idea. I refer to this as uniform, and uniform initialization. It’ll initialize if it can. And if there’s any ambiguity, of course finds ambiguity, if you’re in a context where, like you’re calling a function and the target could be a point or a complex number, it’d say tough luck. Tell me, or tell me. So, the error checking has actually been improved. C++11 is slightly better at finding bugs than C++98 was. I come from the school of philosophy that says that the compiler is your best friend when you generate code, when you build programs, and to make it your best friend, you actually have to have more types. If everything is an integer, well what can the type system help you there? You know, it can’t. If everything is a floating point number, it can’t tell you whether it’s Imperial or SI units, and you get bugs, so you need type-rich interfaces. For that you need to be able to build cheap types, basically, and flexible types. And convenient. And types that’s easy to use, simple. And so, we worked from complex of 1 comma 2 or something like that. So curly is 1 comma 2 and type is optional or it’s only needed when it’s needed. And then finally, we can now write, if we wanted to, 1 plus 2i. Define i as the unit for the imaginary parts, and you get complex arithmetic without ever saying complex. It’s down in the definition of the i-suffix.
مشارکت کنندگان در این صفحه
تا کنون فردی در بازسازی این صفحه مشارکت نداشته است.
🖊 شما نیز میتوانید برای مشارکت در ترجمهی این صفحه یا اصلاح متن انگلیسی، به این لینک مراجعه بفرمایید.