CppCon 2017: Bjarne Stroustrup “Learning and Teaching Modern C++”

– I want to get to introducing
our opening keynote speaker. I want to say just a few words about him. A lot of you know that Bjarne
is the guy that created C++ but what a lot of people
don’t know is that he is very very interested in education. When he left his position
as a corporate researcher to become an academic researcher he also took on the
responsibility of educating undergraduate engineers in how to program. Not in how to program in C++
but how to program using C++ and he took those insights
and those lessons learned and turned them into a very
very successful textbook, again, on how to program
using C++ but how to program. And so I think that when
we think about our role as engineers on a day to
day basis, we spend some of our time as students and
some of our time as teachers. And if we are going to
be successful engineers, we are going to be successful
as students and successful as teachers and so I’m very
interested and very excited to hear what Bjarne has to say. I did say, however, that I
have a special announcement for you today and I want
to put that in, I forgot to cover that, we’ll be
announcing a lot of winners. I will announce the SCM challenge winners. We’re gonna announce the
Microsoft drawing winner and the poster competition winners. There’s another winner I’m
going to announce right now. This is announced today. It’s not us, it’s the Institute
of Engineering Technology. Engineering and Technology. This is the oldest professional
engineering society in the UK, possibly in the entire world. And they’ve announced today
their Faraday award winner. The Faraday award is the
most prestigious award that they give from the oldest
professional engineering society, previous winners
have included Maurice Wilkes, Tony Hoare, Roger Needham, Donald Knuth. Today they’ve announced the 2017 winner, who’s Bjarne Stroustrup. (audience claps) So we thought we’d ask him to
come up and say a few words. – Thanks, thanks, thanks, thanks. Very honored about that
award and very pleased. If you read the list of people
who have gotten that award, mostly they are not computer scientists. You see it’s downright intimidating. Let’s see. Am I pumping bits here? Ah, yes. So when I was asked to
give this opening keynote, which I’ve done before,
so I had a problem. It is what can I talk about? That is important to most of us, and that I know something about and that we haven’t heard a
million times at this conference Most of the time, we do what we always do. We do the reasonable rational thing. We don’t fight any fights
that can be avoided because we’re busy. This is the way the world gets stuck and no progress happens,
so I picked this quote from George Bernard Shaw. About how you have to do
unreasonable things sometimes. And that’s what lies behind
a lot of what I’m going to say here, I mean, what should
we do, what’s not right and what can we do to fix it? This is in my opinion an
immensely practical talk. And basically, I’m going
to talk a little bit about teaching programming and
then I’m going to go into something slightly more technical, not teaching technical, not how
you organize a classroom or whether you, what particular
tools you use to grade your students or find
cheaters or things like that, none of that, that’s not
what I’m talking about. This is not really an education forum. This is a forum of professionals,
people who develop stuff with a few teachers sprinkled around. So, then I’m going to go into some details about what we teach as
opposed to how we do it. And we’re all teachers, and we have some fundamental questions which is what do we teach and why? Well, what is it we want to
achieve and who do we teach and how? Those are key questions
and we’re all students. The world is moving on. It’s moving very, very fast. It’s moving too fast for any
of us to keep up on everything and so … We have to be students,
basically all the time, at least a little bit of
the time all the time, otherwise we fall behind,
and then we become part of the problem and
it’s a huge problem. If you’ve ever had a piece
of 20 year old code dumped on you and had to look
at this horrible mess of say, overgrown
illogical class hierarchies littered with … memory leaks and … pointer spaghetti and such, you can usually thank
some teacher somewhere. They don’t get it right all the time and if you’ve ever
gotten people in from … a university, good university course and they couldn’t program
their way out of a paper bag, again, that’s part of the problem and so I’m going to
try and do a little bit of rabblerousing about
we have to do better. Not because I’m criticizing
anybody in particular, but I feel not only do we
have to do better, we can. We’re not all professors,
as I said, and teaching occurs all the time. We have a colleague dropping
in, talking about it or we give a little local
course about the latest features and such, basically we … We give advice. That’s teaching. We better give good advice,
we better give useful advice, we better give advice that
brings the world forward a bit instead of stifling yet
another innocent novice and … One of the things that
bothers me a lot is that sometimes people have the
weirdest idea of what C++ is. I’ll get back to that. But basically, when you’re giving advice, when you’re giving a short presentation, when you’re giving a course,
think about what it is you want to achieve and
work backwards from that. Don’t start with what have we always done and what is easy to get started or if you are a professor,
what’s easiest to set tests for? I mean, anything that can be
tested by a multiple choice test is usually bad. All the interesting stuff
is the stuff that is … Sort of more real, harder
to quantify specifically. And … So who do we teach? Depends who we are, of
course, there are students in which case you have
some chance of knowing their background, though
I’ve taught freshman and they come in from high
school and they’re all over the place. When I started teaching
freshman, it was 60% electrical engineers and the rest
was computer engineers. Computer science was being
advanced and doing Java and I got the electrical engineers. Two years later, when
my electrical engineers significantly outprogrammed
the computer scientists the computer science department changed. But basically you have
to look at the students, the people you talk to,
the people you change and try and understand them. What’s their background,
what is it they know, what is they don’t know and
what it is that will do good for them in the future, and
of course what do they want which may not be the same. Industrial programmers,
again the same thing. Background, aims for the future and … There’s a lot of
professional nonprogrammers that are using C++. People who are physicists,
biologists, historians, they have a totally different
view of what is important. A lot of them go for the thing
that’s easiest to pick up which these days usually is
Python and then they run out of CPU power. I heard about a biologist
that was running three month runs to get his data. It could be done in just over 10 minutes. We know that because that was
done when he couldn’t do it in three months anymore. We have to help those. There are so many in the
world and they’re doing so many important interesting things. They deserve our help, and
basically the most fundamental thing about teaching I think
is that the students are by and large not like you. On average they are not
like you and on average we assume they are just like us. Hey, we’ve all been to
school, so we know what it is to learn, right? We just do what we did 20
years ago in kindergarten? No, you have to think a
little bit harder about these things, and you have to learn. Learn about the students,
learn about their needs. And I think we did a really
lousy job teaching C++ before the first standard and … Not much better later. We, that definitely includes me. Basically, the message about what C++ was was hijacked by everybody that
wanted to hijack something. C++ wasn’t supposed to succeed those days. There’s always fashions and they will … determine what people think is right. A lot of C++ got hijacked
by people who thought it was C with a few knobs
on the side and others got hijacked by people
with the oh, rah rah rah, everything is a member function,
virtual member function, preferably, and everything
should be dynamic somehow. We have gotten a lot of
really bad code out of that kind of stuff, and
it’s partly because … me and others did not
articulate clearly enough what C++ was and how
it could be used well. And this got amplified
through textbooks, through conferences and things like that. We just have to do better. People talk about getting
the message out, messaging and all that kind of stuff. I used to think it was rubbish. No, it’s right. You have to phrase the
message, you have to repeat it and you have to get it
right and then you have to get it out there, so I want
to teach good modern C++ which leaves us with a little
problem of defining “modern”, but I’ll get to that. So I have a background in this. I hate talking about things
that I don’t know anything about and by and large
don’t know anything about I define as having not done it for real. One of the problems with
education, by the way, is that the previous
occupation of most professors was student, and that, I don’t think, is just the right background
for teaching people how to do good stuff in, say, industry. Fortunately, this is
again, it’s not universal. When I’m criticizing
something with a broad brush, it’s not because I think
everybody has that problem. It’s because I have to be brief
and if I put all the caveats and exceptions in, it
gets a little bit wooly but a lot of people have that problem, a lot of people don’t. So I still teach. I give a course every spring at Columbia and I … I talk to a lot of people, give talks, interact on specific projects
so I do have some … Experience in this. And I have … I have written a couple of
books that are in wide use. I think one thing you should
note there is that they are different, they aim at different things. The swan book is aimed at
people just learning to program. The tour book is meant for
people who may have fallen five, 10, 15 years behind
with C++ and need a refresher and are so busy that they
only have a weekend to do it. That’s the design criteria for that one and the thick book is well
for people who need to know everything in quite some detail and … Well, the standard is not a tutorial. So basically when you
teach, figure out what it is that you want to achieve,
work backwards from that and write it or say it
or something like that. So … Once I decided to give this talk, and promised to give it
and still having no idea what I was going to say,
I asked a few people and I got some responses. The thing that first surprised
me was the industrialist ones, it was uniformly, great,
we want better developers. Give ’em hell, this is good. Now the teachers were … Different. Yeah, that’s great, it really
really needs attention, but you know, it’s hard, I
mean, we’ve failed so often. And what can you say that’s
in any way new or interesting? And I’m not sure, I’m trying. If nothing else, you are
not the standard educators or audience so maybe some
of the things I say are new and more helpful than they
would be if you were worn down 20 years veterans of the education system and me, I think it’s very
important but it’s practical. It is not education
theory or language theory. It is something that happens
whether we want it or not every day in classrooms, in
labs, in sort of development floors and we have to do
it and we need to balance two things, principles, ideals,
things that we would like to see done right, remember
George Bernard Shaw? And then the practical
skills that it takes to make positive change. You can’t just go and talk about it. It has to actually work and
you have to build it so … That gets back to what we’re doing. What do we teach? There are many, many ways of using C++. I mean, there’s people going
around saying things like C++ programmers do this
or this is the way C++ is written, and they’re
usually so darn certain. I’m not, I have seen enough
C++, I have seen enough C++ programmers to know that generalizing over the whole group and
then simplifying it to this is the one true right
thing is just ridiculous. Nobody knows enough, nobody
can be precise enough, nobody can give brief
advice that is wide enough to actually help everybody
so we must choose. We cannot just teach everything. There’s too much, and anyway,
we don’t know everything and so one of the things
that I recommend anybody who is in a position where
they do teaching, advising, mentoring, such, is to
think a little bit about what are your aims and
try and write them down. I find writing things down
is really really helpful. It’s one of the things
that makes us honest. And especially if you show
it to some of your friends and victims or whatever, you
can get something that is better than what you first
thought of and you have to remember that the students
will choose what’s important. They will tune out a lot of what you say. They will listen to people
who say exactly the opposite and they may believe them and
they may not believe them. That depends how well
you’re doing the job and how convincing people on, say, Reddit is. And so … You must choose because,
well, everybody does, and so here, there are lots
of aims, very reasonable aims for what you teach and
from there comes some of it how you teach it. There are a lot of people
talking about let’s get people really excited about coding. That’s very important,
especially for the young ones. But it’s not my main focus. Somebody else is better at it than I am. Not that I don’t sometimes
go and talk to high school students and middle school
students where that’s my job but it’s when I talk, it’s
not what I’m thinking about most of the time. I’m thinking about how do
we lay a foundation for careers in system development, whatever you want to call it. And also how do you … How do you … talk to people who are not
actually going to be full time developers, full time computer
scientists but are going to use a lot of that? Those are my two main focuses. There are people who say,
well, let’s get them to write a simple program and that’s their aim, that’s their whole aim. I think that very often
leads them to teach things that convinces the best and the brightest that computer science is boring, that system development is boring because they can do simple things
easy and then that’s it. I worry about that, and then of course, there are people who just
want to enroll more people so that they can get more … funding for the teaching
and that very often leads to bad teaching and to spoonfeeding. Okay, so have problems with C++ textbooks. They very often don’t focus
and if they do they don’t say what the focus is supposed to
be so it’s hard for teachers and teaching assistants
to actually know where to direct things and very
often they are slow, bottom up focused on what can be
tested, and teaching … 1990s C++ with C++ 11 syntax. If you find a book like that,
I am not going to mention any but you will know them, that’s not right. So basically, we have
to teach programming. It’s not just language features. It’s programming techniques
and it’s not just what is valid, I mean, the tests
are like how weird examples can you come up with for the
conversion of an unsigned short to an signed int is
basically uninteresting. That is not what the thing is about. That’s debugging and reading the manual. You can do that once you
manual but you have to give some advice about how to write code so that you don’t get that question. If you write good code, if you have that question, you
probably are doing something wrong, and we can’t try
and teach everything, at least not at once. We start from the beginning. There’s far too much
stuff we have to choose and we need great libraries. I think one of the things
that’s wrong with a lot of early C++ teaching is that
they teach the language in isolation, you get to
page 697 and you get vector and sort is mentioned 100 pages later. That teaches students that
the standard library is boring, uninteresting,
advanced, difficult stuff. Whereas it teaches them
that if you write your own linked list, you’re cool, if
you write your own hash table it must be better than
that standard stuff. This is wrong, wrong wrong wrong, exactly. So we need great libraries. There’s nothing much to do with C++ there but that’s a great library, that’s Trinity Library in Cambridge. So another thing is
when you teach, try not to be too clever. People always want to know the latest and we love to tell people the latest. The problem is that most
people who ask for this haven’t actually learned
the foundational stuff so we have to sort of … pull back. Teach people how to do
basic things, teach people to do simple things. When you’re illustrating
a technique, don’t use the most advanced algorithm you can find. I wouldn’t pick bubble
sort, but I wouldn’t go to the full generality of
everything the first class. And simplify things. The best example is the simplest example that illustrates a technique
or a feature or something like that, simplification
is a really important thing in teaching and always give a rationale. Don’t just tell people, do
this, it’s good for you, it builds character and you might get an A if you answer it just right. Those are very short term … Silly arguments and we need
to do better than that. You have to say why people
should follow the rules, the suggestions you’re making. And the purpose of good
teaching is to get ideas into people’s heads. Sometimes you forget that. Teachers forget that. They focus on grading,
they focus on techniques, they focus on the detailed subject matter and they forget that it’s not
just to show them something. They actually have to understand it. They have to explain what it
is, why it is and what it’s good for, not just one of the
things and don’t show off. I mean, it is so tempting to
show off to our colleague, to stand in front of a group of people and show them something. Look, this is clever,
you don’t get it, right? That means I’m smart. That is not good teaching. It is really … When you do it right, people
come out and say, yeah, what’s that, eh, simple, yeah. You’ve gotten it into their heads. Brian Kernighan had once told me, you know, if they notice you slip the knife in, you haven’t done it right. (audience laughs) He’s a superb teacher. I wish there was something
like egoless teaching but I don’t think there
is but we have to keep our egos in control,
remembering that the purpose of the exercise is to teach
somebody else something useful. I mean, yeah, we learn
something when we’re teaching, but that’s not the main point. By the way, that’s William
Occam, he of the great razor. Very good for use in teaching. So basically we have to use tools. We tend to focus on just
the compiler and a few … sort of statements and a very
limited number of libraries, especially in early teaching. And then people are
thrown into the real world where they drown in complicated
tools and libraries. I do not know how to do this. There are so many things
and so little time. But eventually people will
learn about all of these things and it’s a good idea
if people at least knew what they didn’t know and had some idea that someday when your program
is more than 1000 lines long you may actually need
some of these supports, and not just to make it
easy to do simple things but something that scales to … larger things. And again, principles and practice. You can’t just lecture about programming. It’s a practical skill. I mean, I sometimes say
you think you can learn bicycling from a correspondence course? Do you think you can learn
design without designing things, programming without writing code? Programming without reading
somebody else’s code? People do believe those
things, and that’s wrong. On the other hand, you
shouldn’t just hack. There’s a lot of people that
just learn from just looking at from what other people are
doing and what other people are saying and they just
throw code together. Cut and paste is the ultimate of that. Don’t just hack, you have
to have the principles also, but you have to have hands
on thing, and it’s not easy. It’s not quick, nobody can do
everything, but do something. I mean, throw in a robot
or get people to use a Raspberry Pi or something. This is probably obvious here. It is not obvious to a lot of universities at the first couple of years. And complexity is the enemy
of just about everything, including good teaching. Try and keep it simple,
focus on simplicity, and don’t throw all of C++ at people. They won’t be able to
understand it anyway. Basically, we tend to … Focus the students on a few things and … A lot of courses doesn’t have
graphics, GUI, web, email, database kind of libraries,
that’s some other course which most of them never take
and they come out believing almost because they’ve almost been told it that there is no graphics,
no GUI, no web, no email, et cetera. The fact that the whole
world is run by C++ using those features is completely missed. In the classroom. When I was teaching, I tended
to use the first five minutes of every lecture to talk
about some application so what does the software on
the Mars rovers look like? And how do you think Google works? And things like that, good
stuff, motivate people. Personally, the thing that keeps me going is the applications. And so … Don’t think there is one way. I really fear monocultures,
single language, single operating system, single curriculum when everybody copies from each other and gets the same things from the web. There should be many ways. The argument, that’s not
the way other people do it or if they do it at
Harvard, it must be right is not conclusive argument. You look at what others are doing, you try and learn from
it, but is there an angle of your own that you can use? Are you a university next
to a tractor factory? Maybe programming tractors
would be a good idea. My hometown was into wind
turbines and so there’s an example there, something
that can inspire people. And don’t hype. I was criticized a lot in the
early years for pointing out that in my opinion C++ could
double your productivity. Everybody else promised
orders of magnitude, transformative change. How do you get a transformative change? Little step at a time. I don’t believe you have this great genius that just solves all
your problems like that. That’s not engineering, it may
be certain branches of math but it is not when you have
to actually build a large, complicated system, have all
the pieces work together. So don’t hype. And don’t let people believe in magic. I mean, friends don’t let
friends believe in magic or something like that. I really try to explain
people how it works so you explain the
library, then you explain the techniques that the library is using, then you explain a little
bit about how it interacts with the hardware and if
you are really on a roll you can tell them a little bit
about how the hardware works. I mean, it’s abstractions all the way down and people should understand this. I meet too many people who … Do one level of the system
and then aren’t interested in how it is really done. I mean, I’m frustrated if
I don’t understand what the next level down in
the abstraction stack is. I can’t not know everything,
but at least I feel bad about it, if you feel proud that
you don’t understand hardware there’s something fundamentally wrong. If you feel proud that you
don’t know how your library is interacting with the
operating system and how often, one, you shouldn’t be proud of it. Two, you will not write
low latency software or high reliability software. This is practical. And basically we have to succeed. The main argument is that we will depend in our daily lives on the code written by the
next generation of programmers. So if we screw up, we’ll pay for it, some way or other, so we … We should focus on this, and
a certain amount of idealism is necessary, if people
just have the attitude I do my job, I do my
job in the simplest way, let’s not rock the boat. Well, a lot of people do that. For some people that’s the
only strategy that makes any sense, however, some
of us have to rock the boat and we have to make some change. That requires idealism, it
requires you thinking about what you want to do, what
would be good, what change is necessary and sometimes
a lot of work on it. I have an idea along the
lines of you know doctors have certain professionalism,
there’s things they just don’t do, and if they do they get into real trouble. Same with engineers. We build things that are at
least as essential for our world as the hardware engineers
and the medical doctors. After all, what tools do they use? And we should hold ourselves
to a very high standard there. So we have to do it better. And we have to start out
right and then keep doing the right thing. Yeah, Lego’s programming C++. So let’s try first a
little bit by analogy. How does other fields
deal with the problem of getting people on board and … Appropriate levels of complexity for them? And I like cameras and
photography so I’m going to do an analogy with photography. Remember, proof by analogy is fraud so this is just an
illustration, nothing more. You can’t just copy it,
but basically there are … There are very great similarities. Results depend on the
equipment we’re using, it depends on the user, it’s
not an automated process that just runs. That means you need teaching,
training, experience and such and there’s lots and lots
of components involved in both software and in
photography and users differ dramatically, I mean, my
main weakness in photography is that I never have
the time to take the … To take the necessary
effort to learn my equipment and to spend the hours necessary
for the light to be just right, I mean, you can’t
just pull the sun up and down as needed, you have patience. That’s my weakness, I
have it, that’s what … People have different
skills, people have different course factors, individual
needs, tastes and skills change over time, this is true for both photography and programming,
this is why I think the analogy might be enlightening. And needs differ and a system, there’s lots of parts to it. If you’re programming in
a programming environment, your libraries, language
features and such is one system and you can’t just start using
something totally different similarly with photography. So basically, there’s a
professional level camera. It weighs a ton, that
disqualifies it for me by the way. I like to travel light. It takes superb photographs
if you know what you are doing and you take your time,
I don’t have that time, I don’t have that patience. Furthermore, you can
completely drown in equipment. Boy, I mean … If you’re a gearhead,
there’s no end to the number of gadgets you can attach to your system and similarly, there are
people who never saw a library or a language feature that
they didn’t want to use in their latest project. Now. There are things that are geared
for people with less skills and less … Patience. A lot of work has gone in
to designing these things so that … Mistakes aren’t really … Disastrous, I mean, a
slightly blurred picture is still a picture and
a slightly off program is a good start. And they tend to be self contained. That is, you get a
package, you use it, and we’re not good at that in programming. I mean, people get the C++
compiler and then where’s the GUI, where’s the graphics library? The greatest problem when
I was teaching undergrads was to get the GUI library installed. This is where the real world hits. It’s painful and it doesn’t integrate well and my students were
using Macs and Linux boxes and Windows boxes and it’s all different. Same library, all the code is portable but just getting it installed is painful. These packages are done to
basically have everything in the box and it works
right out of the box. And I think I would like to see that, but there’s another aspect
of these things is that … The various manufacturers,
Sony, those were Canons or some Olympuses, some
Nikons, each has a system where you can learn and then
upgrade to the next level, not totally painless but less painless. The user interfaces are sort of standard. I can pick up any camera and
do a simple point and shoot kind of thing, it’s
standardized across there. And each has a way that
makes it easier to go up if you need more or to
go down if you are not in a professional mood. I mean, I make better pictures in the first two days if
I use a point and shoot but if I have a vacation,
the first two days I will screw up with the
more advanced equipment but then I’ll get better. You can move up and down these levels so I think we need a system. One system does not fit all. If we give the professional
camera to an amateur, one, they won’t be interested,
two, they will make really bad results and we
need some kind of idea about interchangeable parts
and things like that. And there should be distributions. I should say, give me the amateur version and I should basically
not have to say much more than I just did there, give
me the enthusiast version. I come something with 10 times
the number of knobs and dials and if you are a professional, you ask for the professional, it’s
going to be more painful but you are going to beat all the amateurs if you know what you are doing. And … So why three? Why can’t we just say
here’s the perfect system? Because there’s no perfect person, there’s no one person that … Needs all of that kind of stuff. We have a tendency of
throwing everything at it. Look at teaching, look at
what you get out of the box with a standard, you get
everything that you need to write the kernel of an operating
system, more or less. And 99.99% of people don’t ever see that and shouldn’t see it because
they’re doing something else so basically we are doing something. We are doing something. We are moving in the right direction. Modules will help. This is the way I would
like my code to look like. Give me the entry level bundle,
that’s all that’s needed, and if I want the enthusiast
level bundle, I want that instead, so I have the choice and it’s a coarse granularity. Maybe I want to say give
me the basic games package or something like that,
but we have to limit the number of things
that people can get … So that they can not
get lost in the details and then you can add
things that’s not part of the foundation, 2D graphics,
3D graphics, da da da da. Okay, and we need a better
package and build system. There are package systems out there. They’re improving, thanks
to all of you that are working on that, but the average student, the average user gets lost
in how do you download something, how do you install something? How do you know it works with
the stuff I get from over there plus the stuff I’m
getting from over there? We need to do better. I really want my operating
system to say give me the GUI from XYZ, install
it, and then I should be able to import it. If it’s more complicated
than that, it’s probably too complicated. This is not just for the novices. It’s not just for the first year students. It’s for all of us when
we’re entering a new field that we’re not experts in. I have this distinct
feeling that in most fields I’m not an expert. In most fields I do
things, I’m not an expert because I’m learning new
things and I’m getting into new things and so I’m
an amateur and I need all the help I can get. So … Let’s go to slightly
more philosophical thing and then to some code. There’s no value neutral teaching. When we teach, we say what’s important and explicitly or implicitly
we condemn something as bad, we decide what we would
like the students to learn and if we teach well,
they will believe us. If we teach badly, they won’t. But we have to decide what
it is we want people to get. In particular, if we are
teaching C++ at some level, what is it? What’s good about it,
where do we want to go? What would you like your code
to look like in five years or tomorrow? Do you want to write 1980s C++? Some people do, but I
don’t think they should so my primary interest, again, is … Get a professional attitude,
dependability, maintainability Performance is important
in most of what I’m doing. It’s not everybody’s essential thing, but it’s important in
C++ and reasonable cost. These are sort of engineering
concerns, not science concerns as such. And so … Here’s my summary of C++, as ever. I like static type safety,
precisely specified interfaces. I need resource safety. People have to understand
constructors, destructors, RAII. What is the value of
abstraction, how do you use it? How do you do generic programming
and use templates well as opposed to just use templates? And simplicity for most developers. This, I think, is key. I did a small survey of … Textbooks about C++. While writing this, and
it’s a sorry result. Do you know a textbook that
emphasizes these things? I think a lot of them
don’t even mention it. It’s outside the scope
and that’s pretty bad. And change is hard. Modern C++ is not sort of in particular. It’s the most effective
way of using C++ today given the current standard, which is now C++ 17, it was
formally approved a few days ago and inertia is the enemy of good code. I mean, professors, eh,
that’s not the way we do it, it doesn’t fit into the curriculum. Maybe we can make a
change in five years when the data structure course is willing to do a more modern thing, and students … They reject what they didn’t expect and they expect from the strangest places or from the web or from their friends and a lot of students,
I find, are convinced they are smarter than their professors. And, you know, sometimes they’re right. Especially a lot of high school
teachers have been smarter than their high school teachers,
especially in computing where computing has not been … A key thing, where we don’t
have a good solid curriculum like they’ve got in calculus. Quite often the teachers are
not as smart as the smartest student, so if you get
the smartest students into a university course or into a job, they may have a slightly jaded opinion about their teachers. That will apply to you
to because you are now in the role of the teacher and I have definitely had people, repeatedly every year in the course, freshman coming out of Texas high schools, who are absolutely convinced
they’re smarter than I am on the topic of programming. In this particular case, I’m
reasonably sure they’re wrong. (audience laughs) But … It is something that has
to be taken into account and dealt with. They don’t take it just because you say it and we have to worry
about current practice because it’s very often
what was fashionable 10, 20, or 30 years ago. By the way, this year,
similar is 50 years old, so object oriented programming
is now 50 years old. For somebody who thinks
it was cool and modern, just remember, 50 years,
we’re that far into it. So modern C++. I define it as the best practices
using the current standard possibly with two S’s. And basically I aim for
completely type safe and completely resource safe code and there’s a project which
I launched here two years ago about the C++ core guidelines that basically tries to codify this as a set of rules. This is not what C++ is. It’s a kind of way of saying … What are the good practices? How do you get started into this? I mean, it’s not something
that it’s enforced by the compiler or everybody
has to do just like that, but you really need a good
reason to do something different is the way I define it. It’s a very active project. There are … Dozens of contributors every
month and it’s a very wide … Sort of … Set of contributors, I
mean, where else do you find Microsoft and Redhat doing
something together regularly? The world is improving. So … Code examples. Programming has to do with code
in the end, so good examples are key. I sweat blood trying to find good examples and I’m never as successful
as I would like to be because they have to be simple, for teaching to a class, it has to fit on a slide
or a couple of slides. If people show me this
much code, the ideas are going to be lost in the code. You can’t get the idea there. Yeah, if you’re in a work environment, you can sit over a screen
with a guy that walks you through it, great, but … In many, many cases
finding simple examples is really really hard
and also the big examples sometimes aren’t the greatest examples. They have all the warts that was necessary for compatibility, for
the way we usually do here and the conventions that
have grown up since the 90s so finding small examples are key. And don’t separate code from explanation. If you just show the code,
people will not get the idea and they will invent reasons
why it looks like that and those reasons are
sometimes seriously weird. If you give them just the explanation, that’s sort of the standard
lecture of what is good and proper and people go to sleep. They don’t get the idea,
so a tight integration between examples and
explanation is essential. I learned that from Brian Kernighan. If you haven’t, read the
introduction, the tutorial the first chapter of KNR again. It’s a beautiful example
of how you can get ideas into people’s heads. C would have been among the
dozens of failed attempts to do what C did, if
it wasn’t for that book and that style of teaching. It’s very important, Brian is,
by the way, a really nice guy The fights between the C
community and the C++ community did not happen between me and
Brian and Dennis, that’s fact. So let’s look, I can’t
give a talk without code so I’m going to show you a
few examples of what I think. So here are two ways of writing a loop. The first is the good old
way which we have written more or less that way since the early 80s and then there is the new
fancy way with the range for. And why is two better than one? Basically, two states what I
want to do for all x and v. There’s no i, I didn’t need an i. There’s no separation
between finding the limit of my data structure and the … Use of that variable, constant,
whatever you want to call it so I can’t make a mistake
about getting the size wrong. And … You can explain this, this is easy. So it’s shorter, less
opportunities for … for mistakes. And you know, yeah, one is more general. You can express more cases. But that also means you
can make more mistakes. So remember, goto is the
simplest and most general control structure, at least non concurrent but you can’t implement
concurrency with a goto anyway. So basically just because you can do more doesn’t mean it’s better. Sometimes it means it’s worse. The fact that goto can do everything is the fundamental reason
why we don’t use it. Except in machine generated code and such. Yes. Here’s some example,
slightly more modern example. I am … Going to blink some LEDs and
there are two ways of doing it. I can give an integer, and it blinks, and I can give … A millisecond argument that
says how many milliseconds that can blink. Now if you use integers, you
have to guess about the unit. And so if you have blink number two and you just give
it something without a unit the compiler says, well … You didn’t tell me what the
unit was, so I won’t compile this for you, this is good. And the reason it’s good,
my best example is that little spaceship there, the
poor little Mars climate orbiter. It was sent up to Mars. And it was doing the
final course corrections to get around Mars in a good
orbit and it disappeared. My favorite theory at the
time was that it had been kidnapped for ransom by the Martians. There’s also people that claim
it was a Yankee go home thing We don’t want any
stinking earthlings here. However, what happened was more relevant and less interesting. There were two groups of engineers and they were transmitting a power reading through an interface that took a double. On the one hand, they had SI system MKS, on the other side, they
had the imperial system. As that double passed
through the interface, it changed its meaning with
a factor of exactly 4.65. And we never saw that
little spaceship again. That means that about the
lifetime’s work of 200 good engineers just disappeared, basically, they’d all worked for nothing. They got the salary but didn’t
deliver anything of value to the world. All the scientists that
were waiting for the data had to wait, lots of
promotions didn’t happen, lots of students didn’t
get their degrees, and … Yeah, basically, we lost
about 600 million dollars and it took 15 years more to get one up. That was the bug that you see there in … The first use of blink_led2 1500. Some bugs can be really
critical and we can do better than that, and if you want to be generic, you can do that too. Modern C++ is very flexible and
you can use that flexibility to gain precision, so here we
have some of the rules from the core guidelines,
make interfaces precisely and strongly typed,
that’s an example of that. And we can go further. Always initialize an object. This is a saying that people
break again and again and again They are so smart, they know
they are going to initialize but they declare it up here and
they initialize it down here and they get it right all the time because they are good clever programmers. No. Even if they get it right, somebody else is going to write code in
there or you have handled all three possible cases
and somebody invents a fourth case, so one of the
rules is always initialize an object, and that
always has some pressure because people have had
other ways of doing things over the years, like I
like to define all of my variables up top in the program and then I initialize
them the way I always did in Pascal and other
languages further down. After all, I would like to see
what all of my variables are. That, by the way, is a
sign that your function is too large, which is another rule. But here, always initialize the object. Now sometimes you can’t do
that, so you have to deal with exceptions like input
buffers, and in particular multiple out values have been … A problem over the years
because, well, if you have to call a function to
calculate the initial value of two or more … Variables, then you end
up having some things like that error code there. You call the channel, you
pass the error code in uninitialized as an out parameter and you don’t say so,
the reader can guess it. No, and then you have to chase
it, so in C++ 17 we have … We have … structured bindings where
we can say I want to break the result of the channel
open into its component parts and so we can now return
n values and use them as individual variables
instead of having them into an object and accessing
the object in certain ways. So always initialize objects
and return multiple out values as opposed to using out parameters which is a known bug source. One of the ways when we are
developing this is to look at bug logs, what do people get wrong? What can do we do to fix it? Again, we are not trying
to be clever here. We are actually trying to
find out the least clever ways that work. Basically. Develop an error handling
strategy early in the design. That’s a rule, it’s sort of philosophical, you can’t have a static analyzer
find it when you haven’t. Some rules are high level,
some rules are low level and checkable. Throw an exception to signal it. Exceptions are for error handling only. So if you look here … I have gotten too far down
here, yeah, I showed you … No, I did not show you this. So first we start out with
a classical style without parameters, then we use … Structured bindings and
finally we can start discussing should it have returned two values? Should it have thrown an exception? And this is where, the
tricky bit is to say, when do you throw an
exception, when do you have an error code? Is it normal to fail? Is it expected to fail? Then it’s probably an error code. If it’s considered a design
failure of an extremely rare event that fails, it should
be an exception is the answer in the core guidelines. And basically don’t be too clever. In the context of using
C++, clever is a swear word. Don’t be clever. So here is some code, slightly simplified. It’s got really bad bug in it, I don’t know how many of you spotted it, how many of you would have spotted it if it had been in code that
was a couple of screen fulls? It is trivial for the … for a static analyzer to catch this. The problem is that people
are messing with resource management, and there’s some rules. Don’t leak resources. Don’t call new and delete
explicitly in application code. It should be encapsulated
into a resource manager. And a raw reference is
non-owning, in other words, if I return *new of
something as a reference, don’t do it. Because not only have you
now sent something out so that you know that
it has to be a delete, but there’s no pointer out
there that gives you a hint that you deleted so a lot
of the guidelines have to do with resource management
and the analysis software, actually, the existing stuff
that’s shipping from Microsoft is actually catching this and
I believe that other efforts are in place to catch
things that has been … Specified in the core guidelines. Lately I have found
that people have learned to express things directly in
code a little bit too well. I see a lot of code that
has no comments in it because in the old days,
people wrote comments, pages and pages of comments
and they tended to be out of date and unreadable
and imprecise and in a lot of places, people say,
no, no, comments are bad, we won’t have them, and
now I can’t see what it is you’re trying to do. I can see what you do, but
I can’t see if you meant it and so basically the rules in the … Core guidelines example here. You express things directly in code. But there’s things you can’t
express directly in code and that is why is it there,
what is it what you are trying to do, so basically intent
should be in the comments and comments should be crisp,
not hidden in sort of somebody trying to write
War and Peace because they really wanted to be an English major. So things like that matter. And the old rule is if the
comments and the code disagree both are probably wrong. And I actually applied that
rule when I wrote the ARM, the document that started
the standardization. I tried carefully to say everything twice because then if I screwed up,
there was a high probability I would be caught. These days, when people
write specifications, they try very hard to
say things once only. Which is sort of fine except
that I can’t see if they meant it, so if I read a
statement in the standard, I don’t actually know if that
was what they meant to say and I didn’t quite understand it or … They didn’t quite understand
it and I can’t see it so comments are important. So there’s a fair amount
of philosophy rules there. I have chosen to illustrate
this with Archimedes, arguably the world’s greatest architect, sort of a philosophical
view of things, a framework for your design, a set of
all our rules is important. I’m not going to go into the details. I encourage you to go on GitHub,
find the core guidelines. Look through it, if there’s
some you like or if some you don’t like, send me an
email or put in an issue or something so that we
can improve these things. And basically, my aim is very simple. We can write type and resource safe C++. My first step is to make sure
that I can do that reliably. That requires tool enforcement. The tool enforcement is incomplete
now but we’re improving. But basically, no leaks,
no memory corruption, no garbage collector, no
limitations of what you can say. No performance degradation. This is after all C++
and write in ISO C++. I am not a fan of proprietary extensions so this is happening and
I think this is part of what it takes to teach. You have to figure out what is you teach. You have to articulate why
that is a good thing to do and so teach for the future. You have to live in it and the world shouldn’t continue to accept sort of unsafe bloatware
and probably won’t and if we don’t improve, somebody
else will try and improve it for us and we may very well not like what they come up with. So the proof of the
pudding is in the good code and there’s a lot of practical
stuff that has to be done to do this, correct,
maintainable, appropriate. Efficient is sort of my
version of Vitruvius there and we need to work on
curriculum development if we are in the education business and learners should build
a portfolio, not just go for the grades, it’s
really, really hard to get current day students to
learn as opposed to swot for the exams. Yes, it is … So … I come to the question area
and I’d like to point out I know very well that I
have not said everything that should be said about these subjects. There are several other
things to do with learning, training, teaching at this … conference, you can go
there, there’s in particular a panel tomorrow evening
where we are talking about C++ in academia, I am
a panelist and so are a couple of other people. Okay, questions. (audience claps) We have microphones up here, right? Is there one on each side? Please use them because I
really can’t hear if you don’t. This is an awfully large room. There’s so many more people
than there was last year and last year it was large already, so … – Hey, so the core guidelines,
you mentioned them, you mentioned about teaching with them. They’re quite large as they are and that can be quite
overwhelming to students to take a look at the core guidelines directly. How would you recommend
approaching teaching the core guidelines and
getting students to take a look at them? – First, I suggest that
you read the introduction to the core guidelines that
says what’s there and why, to have the rationale. Then look at the philosophy sections. Pick a couple of those and
then pick some relevant rules to focus on. As ever … You can’t teach everything. The core guidelines were
not attempting to be … Everything for everybody or to be minimal. They are orthogonal, that
is, if you look at a certain section, that section
is reasonably complete for a lot of people and you
have may have repetition on another area. I don’t mind that, if we are inconsistent, which we have been repeatedly,
people find it and tell us and we clean it up. Now … Your colleagues or students
will not need everything so you have to make up your
mind where to focus them but begin with the philosophy
that the general understanding of what we are trying to do
and then focus on something important. I don’t think we are going
to be really successful on sort of a worldwide scale. Till we have better static analysis. I want to eliminate dangling
pointers, full stop. I want guarantees that
there’s no dangling pointers in my code. After that, I can give you
code safety and resource safety and things like that, but
first we have to eliminate the dangling pointers. That is a thing that
takes static analysis, some coding discipline,
which is in the rules and some support libraries
including the parts of the standard library
and so as it is now we’re still working up to that,
but I want to scale this out to sort of four million programmers and you don’t do that without tools. You cannot talk your way into that scale and currently I am doing
more talking than deployment of tools, and that’s
unfortunate but we’ll get there eventually. The team at Microsoft that
started out with this is in a way, I think, I see Peter
sitting down there in front. Are your guys getting on to this? Yeah, yeah, see? Other places are catching up too. I know some projects in China. I hope to see some more in Clang Tidy. So … We need tools to get there and that will answer your question. People will not have to
find the rule, the rule will find you, you give it the code, and it will drop you on the rule which has what is the rule, why is the rule there, what is an example of
why the rule is there and what do you do instead? That’s where I want to go. – Thank you.
– I should take one from there. – All right, so in general,
some people would like to see in the standards
some solution that works better for beginners, I’m
thinking for example of generating random numbers. The solution we have in the
standard library is great for me, an expert, who can
use them, but some people would like just a simple
function and usually we can’t get those in the
standard because there are different, you know,
needs for different people so do you think C++ could use … Just a separate library that
just adds some of these … Simple solutions for beginners,
new string formatting, new … – Do you know the concept
of a violent agreement? This is when people keep
arguing long after they agree because they have to be so
keen on saying that they agree. That’s what I feel like now. I have been arguing for that for ages. I mean …is the random
number library that every random number library wants
to be when it grows up and as a result, people use … rand, and they get really bad results. And I think, one, the standard
committee would do well to remember that they should
make simple things simple and they should put things in the … Standard that help this. A very simple random number
function, something that can compete with rand,
should be there, and … so should, say, split, not
because I like split most but that’s when people
come in from Java and C#, that’s what they try for. They don’t understand how
to get it out of iostreams so things like that should be there. Similarly, for the beginners,
I think we should have a guaranteed range-checked set of containers and I want that portable
across all implementations, by the way. And you can take the … training wheels off later. That has something to do
with the idea of a package for beginners. When I was teaching
undergrads, there’s always one smart aleck-y guy. It’s always a guy. The women are smarter. Who wants to tell me and all of his friends in the class that really you don’t
use vector and string. You have to use pointers
and malloc because it’s more efficient. No, you shouldn’t do that,
and I would like to stop them from doing it because then
they spent their nights debugging stuff that they should
have finished at dinnertime and usually I find this guy
and ask why are you doing this? Efficiency. What optimizer options do you use? And usually the answer is, optimizer? (audience laughs) Okay, so which system are you using? Let’s pick clang. Which optimizer option do you use? I don’t know. Do you use -O2? What’s -O2? You are talking about
efficiency and you haven’t switched on the optimizer. Okay, so things like that
is important, but I think we need a package for beginners and it should definitely
have the random numbers. I love random numbers for
getting semi-realistic programs to run. There are a few little things in the GSL, the guideline support library, that helps but we hope to put that out of business by putting it into the standard and that really is help for professionals as opposed to beginners. It would be nice if there
was a beginners’ library and by the way, don’t teach,
don’t try and teach people today in high schools and … early university without
a little bit of graphics. They don’t believe it’s real computers and it’s hard to do because
we don’t have that package that’s just install
this one, use this one. Random is a relatively easy example. I have a support library for my course. It has random like that in it. – [Man] All right, thank you. – Yeah, about those packages, the bundles that you proposed for
beginners and intermediate and advanced, those libraries,
I mean, they would get some extra standing as
compared to other libraries, so is that … Do you think that should
be part of the standard or should be just libraries
that are developed by the community and elevated
into the status of being part of a bundle or how
do you see that working? – The standard is and should
be tiny compared with what’s available in the greater
world, so the way I imagine this system would be
not the standard but … packages that … educators or … early developers, junior
developers, could use and I would hope that it would
come out of the community. And there are two things
we need for that to do. We need better package and build systems. I know they are being worked on. I’m putting this in to
encourage people to do more. Not to criticize unconstructively and we need modules so
that I can just grab things and be sure they don’t
interfere with each other. I mean, you have some things
and you include a file and some … Some idiots … 10, 15 years ago, defined
a macro called min. (audience laughs) And your code stopped working
and you of course will say, oh, God, not again. A novice would be completely floored. It comes out of nowhere, so I
think package built systems, packages with related
build systems and … Modules are very essential
for doing good teaching. They’re essential for other things too. I don’t see the standards
committee doing it. The standards committee is a
little bit too expert friendly in my opinion. You’ve heard me go on about
making simple things simple. It’s not the main focus of
most people in the committee and so community. – [Man] So you would see
there being options for, like, different bundles for, say
graphic libraries, or … – Something like that, yes,
yes, but this is sort of … I’m waving my hands here. It is not a precise specification. – I had a very specific
situation not too long ago. I was working on a
project with my daughter who has a very good aptitude for coding. But doesn’t know C++. We were working on a thermostat project, has a little graphics, little robotics, kind of stuff to keep you
interested, good stuff. I did what I thought was a good thing. I created a little abstraction
for getting the temperature off the temperature sensor
and I gave her a task to do with this and I found that
the amount of C++ I needed to teach her was beyond … And she’s a college student,
so it’s not like she has no attention span, but it was
beyond her attention span and I was wondering what
your thoughts are on that in terms of getting started with something that’s gonna do something concrete, it’s gonna click some relays,
it’s gonna show something on a display, but … It seems like even the simplest
thing requires the first entire semester of C++ to do. – That is a major part of the problem and if you tell them to download
the library that makes it easier to use and install it, that’s beyond them too, so … That’s actually the deeper
reason I was going on about packaging and modules. We badly, badly need modules
and we badly, badly need modules that can do the
simple clean things simply. – [Man] Right, but I’d even
gone beyond that because we didn’t need the
modules, I already had … – Wait, wait, wait, I’m
just getting warmed up here. (audience laughs) Then we need libraries
that are actually aimed at people who are not CS majors and a lot of our libraries
are aimed at CS majors, written by and for CS majors,
that’s very, very limiting. You need things that go beyond that. I know that’s a session at
this conference on Friday that I would have gone
to if I had been in town, but I have to be somewhere else. I think that addresses something. The people that are
starting with young kids are actually getting some
of the ideas that will apply to novices of all ages,
and as I point out, I’m a novice in many fields
most of the time, so … I’m not being … (sighs) Patronizing here. And … For some people, something
like Raspberry Pi is just wonderful. And that’s an adult thing,
too, not just for kids, so we need to do more and
better, but I can’t be concrete on your example, it’s … You’re touching in the middle
of a very hard problem. – Hey, there, I’m gonna
switch gears a little bit from what we’ve been talking
about, although I think you just touched on it. At the very beginning of
your talk, you talked about how your aim is mostly
towards students who are going to be professional software
engineers in some sense, do some software work professionally, but how do you feel about programming also just as a way to teach
rigorous critical thinking to anybody, like how we
teach math to everyone, not just people who
are gonna be technical, and how would that change
your aims as a teacher? – I think that’s important. Traditionally in our culture
we have used mathematics for that, and … people aren’t getting enough
mathematics in my opinion. You can do programming
as a far more concrete and instant feedback form of
learning critical thinking, learning precise thinking. Code is very good at
hitting back where you are thinking in a wooly manner and for that I think the most important
thing is that you don’t make it too easy and in particular
you don’t make it fuzzy. There are a lot of work done so that … If you say something,
it’ll do something and … It’s not too fuzzy about what it is. If what you are wanting
to teach is precise, scientific thinking, precise things. The reason I focused more
on the software developers was that’s what I know best. I’m very reluctant to say precise things about things I haven’t
actually tried myself and so I mentioned it as an important area but I think it’s an important area that takes some serious work
by somebody who isn’t me. – [Man] Thank you. – Yes, two comments, one is on the … the guidelines themselves. I think that’s probably one
of the best things I’ve seen toward the language in
the last two decades. A problem, though, is that … I was trying to encourage
a section of my company to use it and they happen to be in Germany and they didn’t really
understand the preamble. They didn’t get the joke,
actually, about it being incomplete and such … – [Bjarne] Oh, that
one, yes, yes, we’re … Okay, I mean … There’s a joke about Germans lurking here. (audience laughs) But the … Yes, maybe it’s time
to take that one away. It’s a saying that we’re using
in some of the drafts at … In the standard documents
and in the proposals, yes, you’re right, that may
have outlived its purpose and jokes don’t travel culturally. I know that, and I think I
slipped up there, thanks. – And the, one of the
difficulties is that … I work in a highly regulated industry in the medical devices industry, and getting GSL to be
used as the library is difficult because there
doesn’t seem to be any stated direction as to when
it’s going to be proposed or entered and so having
that guidance there as to … You know, at what level it’s
going to be proposed into the main standard library or
what track it’s going to be on would be nice, especially
since some of our code is … vintage. – Yeah. There’s always the inertia
of old code and old think and that we have to
fight in every industry. I work in a highly regulated industry. Well, people don’t die if
we screw up but they may jump out of windows with the same effect when they get ruined, so we
have some of those problems and we have gotten permission
from our regulators to use things like that so it can be done. But … I hope I said it somewhere in
the guideline preamble that we need specific guidelines
for specific areas and we just haven’t gotten there yet. It is still the core guidelines, but … If you look forward 10 years
time, I would expect there would be the medical appendix
and the financial appendix that has specific rules,
supports and restrictions suitable for an industry. I know most of us have
to worry a lot about the next release and the next quarter. I try reasonably hard to think
about sort of five years, 10 years, out in the future
because it is going to come incrementally from what
we are doing today. The core guidelines is good
and it’ll be much better in the future but boy, do
I wish it was moving faster and we had better support
and better enforcement and things like that. I guess I’m somewhat impatient
with a lot of things. – [Man] The one from Microsoft
works for most things, but it’s also nice that
there’s a single header GSL light one that actually
goes back to more ancient code. – It’s … A lot of these things work together. I can’t wait till I can say without … Caveats … Use concepts for all of
your template arguments. If you want a fully generic … piece of template argument, say so. Like … Define the concept takes everything and use modules to make
sure that you don’t get infestation from other
people’s macros and type defs and that works together
with the views in the … Core guidelines. There are things that
we can do much better with what in my opinion
should have been 17, but not everything can
be 17 so let’s go for 20. And the guidelines will be better for it. And from my perspective, it’s
the same kind of thinking for getting us to the future. – He was talking about how
using libraries is important and how writing around
linked list is probably not the best idea and I can’t agree more, but, say, like 10 years ago,
when I was writing my first application, I invented my own atomic, so the bad news was it was
actually used in production but the good news is that I
learned a lot while doing this and you might have heard
they say that you’re not a real programmer if you
didn’t implement your own string class, so the question is… If we have a generation
of software developers who didn’t implement
most of basic constructs but they already learned
to use ready abstractions and libraries, should we
miss something or maybe this is the right way to go? – If we don’t teach people
how to do lower level software we get a generation of
programmers that only know how to use a library and that’s wrong. When I teach people,
when I teach freshman, I teach them how to implement vector because they need to know
about pointers and freestore and they need to screw up
so that they see it’s good to encapsulate it and those
problems don’t happen anymore. When I teach … Later on, I … Very often have people design implement a string so that they can see how it’s done and so I can show them how
they could have done it better, so yes, we need to get
down to at least the C++ memory model at some point. Otherwise we teach people magic. So you have to get down there somewhere. I thoroughly agree with that. On the other hand, I don’t
think everybody needs to write a lock-free code. If you are a computer
science major and if you are specializing in sort of engineering uses, yes you must do that, but not everybody. But yes, you have to reach
some level of abstraction that is good for what you are doing and for computer scientists I really think they should know how vector is implemented in some extent and how the string is implemented. They should know how to do
the short string optimization, for instance, then they understand. – That is great when you
teach your students, right, and imagine that you have
a junior, for example, in your company and you see
that he’s trying to implement some basic stuff, and you
tell him, don’t do this, it’s already done, yeah, that’s great. But perhaps he was trying to implement it because he don’t actually get how it works and he don’t want to be using magic. So from a practical point
of view, I don’t want him to write it, but from the
education point of view it would be convenient to do this. – In a concrete way, look
at chapters 17 and 18 of the swan book. I have done that for 10 years. That’s the vector implementation. It’s a way of trying
to address your concern for first year students,
people who have never seen a line of code before
they started chapter one so I don’t drift into
just the stratosphere. That’s not what I’m recommending. – So for, you’re talking
about Raspberry Pi. I ended up doing a summer project on one and I ended up doing it with
Python, which was nasty, but so with your talk of modules, for the Raspberry Pi
there’s a lot of stuff with setting up I2C connections, for connecting different components, setting up the different PWMs on the pins so would you have like a
module for Raspberry Pi that’s specific for the Pi or … – I think when you get to real hardware, you need dedicated facilities
to make that hardware reasonable to use, yes,
and so I would expect once we get modules in
the language, C++ 20, that was not ready for 17, I would hope that Raspberry
Pi would have a basic Raspberry Pi support module
or something like that. – [Man] That’d be great
because it’s another thing like you said where it’s good
to have interfaces and stuff which Python has a very
simple Windows 98 one. – I know, I know, I know,
hence the talk about photography, I wish I
could have done that talk for C++, but I can’t, later.
– Thank you. – Hello. So I volunteer teaching high school kids and there is always the
question, I’ll give an example. For example, when I teach
operator overloading or stuff like that, so and
I’m teaching them, like, copy constructors and operator equals, there will always be the
student that will ask about move semantics and stuff
like that because they are sitting with their laptops
and they’ll google it so how would you go about
finding the golden ratio between what’s essential to … Like, teach them modern stuff
but also not overwhelming? – I mean … That’s a very very hard question. I don’t really know. I can point to the swan book. When I had the concrete
problem, not for high school students but for freshman,
that was my solution and I think … Some of that is pretty fundamental. If you are dealing with
people who want to be software developers,
you have to go through the usual stuff, you have
to show them what a loop is, what a function is, how to build a class, how to use a class and
you will eventually … Think it’s chapter 10, get
to move semantics and such. For high school students,
you can’t quite do that. One of the problems is that
people always read ahead and they want to try something else. That was what I said
these people that decided that pointers and malloc were cool. And they wanted to be the
cool kids and corrupting the others. That’s always a problem and … I think a good teacher
will find a way of handling the students, knowing
the students and such. What you can do in preparation
is to have a curriculum that provides abstractions
that’s relevant to what they’re doing so that you can
say, yeah, you can do that but if you use this one for now, you’ll get to bed at a reasonable time. I mean, do you really want
to stay up past midnight debugging all the time? And sometimes when this happened to me, somebody heard of move
semantics just when I’m trying to teach them how to
write a string out to … to output. I say, well … We’ll get to it. They don’t like that answer
but sometimes it’s the only answer, otherwise you
end up doing bottom up and with bottom up you never
reach where you want to get there, all the time is
spent inventing macros for hiding malloc so that
you can catch errors and things like that. You don’t want to go there. So … Try for suitable
abstractions for your people and try and deal with what’s … What then happens. We are over time. So … Maybe we should stop. I will be around today or
tomorrow so try and catch me. There are people
streaming out in the back, which is fair, since
we’re 10 minutes over, so I’d like to cut it now. Sorry for you standing
up there, thank you. (audience claps) – Bash Films can shoot your
event with multiple cameras, link to presentation slides, add titles and edit your event live for
a full broadcast experience. – How is this even working? So this is an even more
interesting program to, you know, look at in a lot of ways. So let’s profile it. Give it a little bit of
time to do a profile for us. See exactly what it is that’s
making this faster or slower based on the different inputs. You can really gain a
lot of insight by looking at the profile like this. – I worked at Sesame Street. I got brought on to be
a writer’s assistant on a show called Sesame Street English, which was to teach English
to kids in China and Japan. It seems very simple, the
shows that they put together. But it’s actually really
hard to design a show that is not only for young
kids but also the parents. – Confession like this is therapeutic. I hope you all get something
out of this, but if you don’t the therapy will have been
good for me, so thank you. (audience laughs) Seven years ago I was working,
I wasn’t working at Google. It was for my previous
employer, which was a large multinational investment bank. I had what was up to that point
the worst day of my career. – And then came the anger. Anger at ourselves because
we knew we were responsible for America’s first space disaster. We wrote two more words
into our vocabulary as mission controllers,
tough and competent. Tough meaning we will
never again shirk from our responsibilities because
we are forever accountable for what we do. Competent, we’ll never again
take anything for granted. We will never stop learning. From now on the teams in
mission control will be perfect. Because as a team, we must never fail. – One other thing. We’re all in a very fortunate position, we’ve been very lucky in
our lives and so forth and I think as part of the
mission it’s also good sometimes to take that fortune and give back. (audience claps) To make sure that we take this platform and use it towards worthy causes. That’s good karma. That’s good stuff in the universe. – We understand that your
event will have needs that are specific to your organization. Please, email or call
us directly to discuss your particular event. We look forward to discussing your goals and helping make your event a success.

Comments 100

  • Modern C++ sucks.

  • Why do students not use Vector? In my college courses, data structures in particular, we were explicitly told not to use them.

  • Damnit, I am in Seattle right now. I want to go to CppCon so bad! 🙁

  • What's with Bjarne's obsession with GUIs? Anyone uses a GUI to install Python, JavaScript, Haskell packages? Anyone uses a GUI to build Cmake projects? You can, but it's much easier done from the command line. Besides, you learn a thing or two while doing it "the hard way." If you want GUIs go with Microsoft products. They have it covered. You won't learn anything beyond clicking GUI buttons though.

  • blame the teacher for a kid writing memory leaks & pointer spaghetti, yes. for not teaching rust in the first place.

  • 28:40 is soooo relate-able

  • I do think C++ should just get completely overhauled and re-written with all the best features thus far included, with easy support for additional features in the future, and an easy way to add or subtract libraries, perhaps through 'packages' or whatever they decide to call it. It's unlikely the very low level mechanics of computer programming are going to change, only the speed of hardware and quantity of memory, unless quantum computing actually takes off, of course. 😀

  • Guy trying to teach his daughter probably got triggered hard when Bjarne suggested what he needed was a library targeted at "non-CS majors".

  • let's hope for the bright future of C++

  • is Bjarne writing any new books?

  • 1:36:20 damn she's hot! o_o Tell me more about Bash Films please o_o

  • Any recommendations for a good simple GUI or gaming library to teach kids C++?

  • By the way, here is my C++11 super easy, very convenient, header only library for random
    (wrapper over C++11 <random> stuff)


  • Been waiting all week for these.

  • Will c++ be ported to a quantum computer?

  • 1:05:18 Criticizes unsafe bloatware and asserts that »the world« won’t accept unsafe bloat — yet, uses Microsoft as guide, positive example and reference. smh
    (Edited to add: gave the video a thumbs-up because of valuable rules mentioned.)

  • What's book he means?

  • One of the biggest risks in a development project is what I call "Swelled Head Syndrome"…where a programmer writes something that is simply cool/intricate…but you could have written it much simpler. I've been guilty for sure!

  • I have one question: Why your homepage is so ugly…could you please just align your pics and reshape them in same size? ^|^

  • Can somebody link me to the part where he talks about range for loops and why are they better ? Thank you

  • i think i'll wait til im bald on top and hair on the back before i attempt learning C++

  • Ohh.. the Germans… They are "inkorrekt, incompleat, and pµÃoorly formatted."

  • I can listen to him talk all day! Great presentation.

  • 51:35, if 2 is better than 1, it sure looks so easy, why wasnt this done in C++98? 20 years of code has built up and now we have to go and change every line of code that loops into a container with this code so that it looks elegant and expressive and "modern"? what about that million line code base with so many such loops? what about the testing time required? the line auto x : v isnt intutitive or expressive. what is auto? what is the range of the loop thats being traversed?

  • 这丹麦口音听的好辛苦啊,还各种停顿。要有人工翻译,有大佬出资就好了。

  • I totally disagree with Bjarne on several points.
    Some of the old style code might look messy. But that's the fault of the developer. Not the language.
    Plus. It's much easier to read the old style for() and if() type code than the cryptic garbage he's adding now.
    Furthermore. In modern C++ he keeps adding new ways to do the same task. Forcing us to commit them to memory. Then telling us in the next cycle that they're no longer supported and they changed it again!
    !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! STOP DOING THAT !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
    Pointers- auto, smart, unique,shared, weak, etc…
    Concurrency- future, await, etc…
    Cut out this crap! It makes the language HARDER to use. NOT easier!
    One pointer and one type of concurrency should work in EVERY case. We shouldn't need to screw around with multiple versions of the same task.
    The point of a language is to perform a task. Not to have a bazillion confusing ways to do that task!
    Earth to Bjarne : Modern languages don't ADD more things for developers to learn. They REMOVE them!!!
    Stop making the language more confusing than it needs to be.

  • Goethe, nietzsche wagner. c++ java

  • In my experience C++ is in most cases much too complicated to solve a user-centered-problem. And this talk shows, that it is a language from the past.

    If you have ever used other languages and you see how easy it is to get a library and install it and it works on any OS/Architecture.
    In C++ everybody tends to solve each problem again, as it is too hard to build a libary, which depends on other libraries without having issues, when updating libraries to a new version or having conflicts. Don't get me started about static vs shared, debug symbols, conventions for naming of header files etc.

    Now take a look at JavaScript: If you start a simple Angular4 project you already have a project which automatically downloads and updates >1000 NPM libraries from the Internet, without any problems. (Each library sometimes contains only a few small classes/functions which solve one problem, e.g. array-egal https://github.com/component/array-equal or content-type-parsing https://www.npmjs.com/package/content-type-parser )

    Every modern programming language has a good packaging/build system (which is also supported by local artifact servers like nexus/artifactory):
    – JavaScript / TypeScript etc. -> NPM / Bower https://www.npmjs.com/
    – Python -> PyPi
    – GoLang -> GoPM etc. https://github.com/golang/go/wiki/Projects
    – Java/Scala/Kotlin/Groovy -> Maven, Gradle, SBT, Ant, Ivy https://mvnrepository.com/
    – C# -> NuGet https://www.nuget.org/packages
    – Dart -> Pub https://pub.dartlang.org/packages
    – Julia -> Julia PDK https://pkg.julialang.org/
    – Perl -> CPAN
    – Lua -> LuaRocks https://luarocks.org/
    see http://www.modulecounts.com/

    For C++ there is:
    – CMake, which is really helpful but also some kind of workaround
    – Conan, which is in a very early state: Most of the libraries you find there are only working on windows OR linux, not all compiler flags work etc.
    – OS specific versions: e.g. apt-get, brew or vcpkg by Microsoft
    – The developers of C++ libraries tend to make it very hard for the user to use it. They don't provide Conan / VCPKG packages.

    So my advice for everyone: Make sure, that you really need C++ to solve your problem (e.g. CUDA stuff) or if you could use another language.
    Performance of JIT compiled languages can be very good as well especially without performance tuning knowledge.

  • I started with c++ decades ago and i am still there.

  • 53:53 I use this case study ALL-THE-TIME, as a $300M (or something like that) program that UTTERLY FAILED, literally BURNED UP in the Mars atmo, all on account of what? A stupid UNIT CONVERSION ERROR somewhere between ground control and one of the integrated APIs. That so say: I like to learn from my industry's and forebears mistakes. 55:00 I'm not sure I TOTALLY agree they didn't deliver ANYTHING. They delivered this: how you MUST be PRECISE, in not only VALUE, but also in UNITS OF MEASURE. It's like saying to someone; see you in a few. A few what? Nanoseconds?

  • 55:44 Yes, INDEED! In fact, not only have I cast the PERIOD, but also at times the REPRESENTATION! ABSOLUTELY!

  • 1:16:50 Sheesh, yes. How many times have I had to specify NOMINMAX. LOL! Or pick your poison. SetPort? Really?!

  • 1:29:30 Couldn't agree more. The fundamentals are of CRITICAL importance; if, as Mr. Stroustrup said, for nothing more than to appreciate, we do not necessarily need to be make those mistakes all over again.

  • 1:30:40 It is a difficult thing to comprehend for someone younger, I think it comes with age, and a bit of learned wisdom over the years. We don't know what we don't know. Does not mean the problem defies solution or cannot be solved; but there may already be a solution out there just waiting to be "found".

  • Amen to the build and package system. My job as a sysadmin for a HPC cluster involves building tons of scientific software. Save for a handful of relatively well behaved codes, it's a total crapshoot of build methods, compile time options/configurations, patching and hacking that keeps this fragile mess together.

  • We are all teachers, we are all students 🙂

  • I think the Unity engine is a great tool for teaching programming. It doesn't have to specifically be a video game application, but Unity is simple enough to add 2D sprites and create an interesting simulation without having to go over the details of how Unity is making everything work. If I ever teach an intro to programming course, I would probably use a tool like Unity where I would first start off with 2D applications, and I'd try to generalize things without going into too much detail. Although Unity provides C#, JavaScript and Boo, and not C++, my goal would be to teach students to see a language as a tool; rather than using one language over all others because it's "easier to learn". A good programmer isn't someone who is really good at one language, or even someone who is better than everyone at using C++ (whatever that means). A good programmer is someone who is willing to help others who are struggling with problems to think through their code and figure out the bugs, no matter what their skill level, as well as seek help when necessary. We can't know everything, life is too short for that.

  • The battle against satan. !!!!! #wtf retards read the fucking book. whats different??? Idiots. The ring. look for the ring. FOR ROHIRRIM.

  • AND THAN U RIDE!!!!!

  • Robots soon. Shittt. wat een strijd.

  • After months of training i did it. HA! fuck u demon.

  • Apollo was smarter than ur teacher. MEDITATIE AAN APOLLO!!!

  • Topology book. RAII the book people. READ A TOPOLOGY BOOK!!!!!

  • I started a club which uses the Raspberry Pi platform to teach coding to my peers in school after listening to this talk.

  • which Swan book is he talking?

  • Imperial Units don't even suck!

  • I like this!

  • I swear all programmers must watch his speeches, also this video too. He knows the wrong of all the books and lectures that teach only syntax. That's why half of students who majoring in CSE know syntax, but had problem on programming, at least in my country.

  • We can. Please update programming. For me.

  • Thank u stroustrup. Art for programmers.

  • Little ratboy. Read plato. Learn something.

  • Thanks Bjarne for this video. Even at age 62 I still find myself as a Student. I am old C programmer and I found myself stuck on a problem and your advice on this video helped me guide in solving a coding problem. It does help to review the programming tools you use. Thanks again.

  • This guy is perhaps a inventor of a mainstream taste developed from a BASIC, cobol, fortran or whatever background into the flavor C++… But talk about talking a LOT of zero containing bubbles…
    I just wonder what happened to the flavor that more highly intelligent person turned to; Delphi? Not much b.s. there, no. Not much of anything, Why, you may ask. Yes. Why?

  • French and c++ learn about cout.

  • c++ as c with programming skills.

  • What's swan book?

  • c++ is shapeless
    this why each year we get C++ YY

  • 24:50 OHhh, that's very importante moment.

  • I feel I ruined myself studing in university. The last year !! cmon I can finnist it.. But already 25yo and know so less :"(

  • Why are people asking for a dumb down c++? c++ is about understanding how the hardware works and writing efficient code.
    If you can't understand c++ use Python.

  • ~ 1:33:00 the question about the smart person in the room is to ask them to independently research it and present in 10 minutes. They'll opt-out at the next class

  • Cant imagine filling out paper made by the creator

  • 2:50 "Am I… Pumping bitch here?"

  • Smartest man of computers.

  • I have to disagree with George Bernard Shaw at 3:29. A reasonable man does both while an unreasonable one does neither.

  • С++ and GUI? I don't think they are from the same World. Qt is lame. I would use for it C#, JS whatever, but not C++!

  • I cant say it if is a glass ceiling or stake out in the opposite building really …I'd go with the glass celing 🙂 Teachers over the internet really get you started and leave you respectful at them, their compassion, motivation and knowledge. But finally it is application….nice though ….the snoots pointing me to the Billionaire's Pledge are trolleys …… leave me hell and priority jiffy …..:-) … why is that one looking at me?

  • …a man should have a purpose in life … lmao …..

  • … and hey the teachers on the internet, especially the guys, it is not a head start ….. thats the best thing about you guys … it is head first!! (Rufus, that makes two of us ) …:-)

  • Cross-Platform network programming please… 😍

  • He is right about people think its a good thing to not understand how the underlining system works or how the library works.
    especially people that write web stuff most annoying people i've met .

  • All we know c++ is the best. i can program from chips to workstations or gpu farms with c++ language.

  • Did anyone start such a library? I would be really interested to contribute!

  • In school we used java for programming, and we never touched stuff of the standard library. Instead we spent most of the time to learn about and implement linked list and other similar data structures in an inefficient, object oriented, non-generic style based on inheritance.

  • Why should we learn to use IDEs? I don't really like IDEs. Most are not that good, I think.

  • The only courses, where I had to use C++ at university was about graphics programming and OpenGL

  • I got inspired to use C++ again instead of rust.

  • I need to say it.. today, education on University is making student some lazy, those high level language programming they are using to teach do not convinced me, I think a true engineer NEEDs to dominate C or C++, this language is not dead, is growing a lot, is so much important…

  • Dejlig accent

  • 29:47 Implies Bjarne is a Rubyist and/or Haskeller

  • When clang has an AI interface to generate code, programmers may not be as needed.

  • Conan is an okay C++ package manager, cmake/bazel are good at building, and hab and docker are good for repeatable environments, but C/C++ really needs a more official pkg mgr because there's so much ecosystem fragmentation, at present. A package store that supports free and paid packages would be ideal.

  • I started learning programming in high school, but it was always simple maths or sorting algorhytms. At the university I started learning PRACTICAL programming on microcontrollers, databases or GUIs, and only then I got an understanding of how everything works.
    You have to have a purpose, otherwise it's useless and you'll probably forget everything anyway.

  • C++ is complex and confusing. There are several reason as follow:-

    1. To do same Task 10 million option available.

    2. Traditional thinking .. delete keyword cause for memory leak so Why u not remove this from C++ compiler.

  • I'm not keen on the packaging idea, cpan and pip already trample my OS package manager

  • "Nobody can do everything… but do something.."

  • Hey, this is a great video, Congratulations!
    I'm from BunnyInc, we are now launching the invite-only VideoBunny service.
    The first outsorcing video editing service. We have top high-quality video delivery.
    Start your new video with us postproduction.bunnyinc.com

  • i feel like enlightened from the god

  • Love his goofy and dry humor. I think that not just for C++, but every Computer Science/Engineering professor should watch this vid so that they would suck less.

  • C++ : we should be as simple as possible
    Also C++ : yeah so that's a const pointer to a const value but things change given the position of the damn asterisk and good luck with that

  • Its like walking into a church and jesus shows up

  • His accent and calm & expressive presentation made it really easy to concentrate for 1 1/2 hours. Contrary to my uni lectures where I doze off after 15 minutes

  • There is a new series coming from CrashToSmash Modern C++ . Hope that covers latest features

  • 2:15

Leave a Reply

Your email address will not be published. Required fields are marked *