Yeah, but as far as I understand that’s not a C vulnerability. It wasn’t added. C just exposes how the underlying CPU works.
If you could avoid exposing dangerous memory quirks but still retain the same power… well, you’d have invented Rust. Rust is a better language than C, I agree with that.
Edit: Yep, just double checked. Buffers live in physical memory and have to be finite, so if you advance outside of them you’ll go somewhere else. Scanf’s not special, this is just another inherent pointer issue.
Okay, but how do you code on a CPU without directly interfacing the CPU at some point? Python and JavaScript both rely on things written in mid-level languages. There’s a difference between a bad tool and one that just has limitations inherent to the technology.
Like, to echo the meme a bit, it’s not a totally straight comparison. They have different roles.
Yes, also Rust. It wasn’t an option until recently though.
The times when C or C++ is worth it definitely isn’t always, but I’m not sure I’d class much of OS programming and all embedded and high-performance computing as small. If you have actual hard data about how big those applications are relative to others, I’d be interested.
Also, it’s a nitpick, but I’d personally say a footgun has to be unforeseeable, like literal shoe guns being added to a video game where guns were previously always visible. Once you understand pointers C is reasonably consistent, just hard and human-error-prone. The quirks follow from the general concepts the language is built on.
True, but AFAIK they all sucked really bad. If you needed to make something that preformed back then you wrote in assembly.
FORTRAN might be a good counterexample. It’s pretty fast, and I’m not actually sure if it’s memory safe; it might be. But, it’s definitely very painful to work with, having had the displeasure.
That’s pure assumption and, as far as I can tell, not actually true. PASCAL was a strong contender. No language was competitive with handwritten assembly for several decades after C’s invention, and there’s no fundamental reason why PASCAL couldn’t benefit from intense compiler optimizations just as C has.
Here are some papers from before C “won”, a more recent article about how PASCAL “lost”, and a forum thread about what using PASCAL was actually like. None of them indicate a strong performance advantage for C.
Hmm, that’s really interesting. I went down a bit of a rabbit hole.
One thing you might not know is that the Soviets had their own, actually older version of C, the Адресный programming language, which also had pointers and higher-order pointers, and probably was memory-unsafe as a result (though even with some Russian, I can’t find anything conclusive). The thing I eventually ran into is that Pascal itself has pointer arithmetic, and so is vulnerable to the same kinds of errors. Maybe it was better than C, which is fascinating, but not that much better.
Off-topic, that Springer paper was also pretty neat, just because it sheds light on how people thought about programming in 1979. For example:
In the following, we shall
compare how “convenient” the languages are to code our
favourite solution to a programming problem,
play the devil’s advocate, and try to list all possible things
that can go wrong in a program expressed in a language.
Some of us, including myself, have reservations about the
validity of the second technique for comparison, the most
persuasive argument being that even though some of the features
are potentially dangerous, people rarely use them in those
contexts. There is certainly some truth in this, but until we
have experimentally collected data convincingly demonstrating
this, it is wiser to disbelieve it. Take note of the observed
fact of increased difficulty in formally proving the properties
of programs that use these potentially hazardous features in a
safe way. This is one of the reasons behind the increased
redundancy (and restrictions) of the newer languages like
Alphard
I don’t see a lot of people denying that 2 is a good metric today. In fact, in the rare exceptions where someone has come right out and said it, I’ve suspected JS Stockholm syndrome was involved. Murphy’s law is very real when you not only have to write code, but debug and maintain it for decades as a large team, possibly with significant turnover. Early on they were still innocent of that, and so this almost reads like something a non-CS acedemic would write about programming.
once you understand C++ the pitfalls of C++ are reasonably consistent
All of C++? That’s unreasonable, it’s even in the name that it’s very expansive. Yes, if you already know a thing, you won’t be surprised by it, that’s a tautology.
C is more than just pointers, obviously, but the vast majority of the difficulty there is pointers.
there are like what, 3 operating systems these days? assume those are all written entirely in c and combine them and compare that to all code ever written
Plus all previous operating systems, all supercomputer climate, physics and other science simulations, all the toaster and car and so on chips using bespoke operating systems because Linux won’t fit, every computer solving practical engineering or logistics problems numerically, renderers…
Basically, if your computational resources don’t vastly exceed the task to be done, C, Rust and friends are a good choice. If they do use whatever is easy to not fuck up, so maybe Python or Haskell.
All of C++? That’s unreasonable, it’s even in the name that it’s very expansive.
similarly, “all of pointers” is unreasonable
pointers can have a lot of unexpected results
that’s literally why java exists as a language, and is so popular
Plus all previous operating systems, all supercomputer climate, physics and other science simulations, all the toaster and car and so on chips using bespoke operating systems because Linux won’t fit, every computer solving practical engineering or logistics problems numerically, renderers…
sure, and the quantity of code where true low-level access is actually required is still absolutely minuscule compared to that where it isn’t
“all of pointers” can have a lot of unexpected results
How? They go where they point, or to NULL, and can be moved by arithmetic. If you move them where they shouldn’t go, bad things happen. If you deference NULL, bad things happen. That’s it.
sure, and the quantity of code where true low-level access is actually required is still absolutely minuscule compared to that where it isn’t
If you need to address physical memory or something, that’s a small subset of this for sure. It also just lacks the overhead other languages introduce, though. Climate simulations could be in Java or Haskell, but usually aren’t AFIAK.
Yeah, but as far as I understand that’s not a C vulnerability. It wasn’t added. C just exposes how the underlying CPU works.
If you could avoid exposing dangerous memory quirks but still retain the same power… well, you’d have invented Rust. Rust is a better language than C, I agree with that.
Edit: Yep, just double checked. Buffers live in physical memory and have to be finite, so if you advance outside of them you’ll go somewhere else. Scanf’s not special, this is just another inherent pointer issue.
exposing the machinations of the underlying CPU with no regard for safety is like, the definition of a footgun
Okay, but how do you code on a CPU without directly interfacing the CPU at some point? Python and JavaScript both rely on things written in mid-level languages. There’s a difference between a bad tool and one that just has limitations inherent to the technology.
Like, to echo the meme a bit, it’s not a totally straight comparison. They have different roles.
a footgun isn’t inherently bad, it just implies a significant amount of risk
yes, if you need the ability to code on a low level, maybe C is necessary, but the times where that is actually necessary is smol
also rust
Yes, also Rust. It wasn’t an option until recently though.
The times when C or C++ is worth it definitely isn’t always, but I’m not sure I’d class much of OS programming and all embedded and high-performance computing as small. If you have actual hard data about how big those applications are relative to others, I’d be interested.
Also, it’s a nitpick, but I’d personally say a footgun has to be unforeseeable, like literal shoe guns being added to a video game where guns were previously always visible. Once you understand pointers C is reasonably consistent, just hard and human-error-prone. The quirks follow from the general concepts the language is built on.
There were memory-safe languages long before C was invented, though; C was widely considered “dangerous” even at the time.
True, but AFAIK they all sucked really bad. If you needed to make something that preformed back then you wrote in assembly.
FORTRAN might be a good counterexample. It’s pretty fast, and I’m not actually sure if it’s memory safe; it might be. But, it’s definitely very painful to work with, having had the displeasure.
That’s pure assumption and, as far as I can tell, not actually true. PASCAL was a strong contender. No language was competitive with handwritten assembly for several decades after C’s invention, and there’s no fundamental reason why PASCAL couldn’t benefit from intense compiler optimizations just as C has.
Here are some papers from before C “won”, a more recent article about how PASCAL “lost”, and a forum thread about what using PASCAL was actually like. None of them indicate a strong performance advantage for C.
Hmm, that’s really interesting. I went down a bit of a rabbit hole.
One thing you might not know is that the Soviets had their own, actually older version of C, the Адресный programming language, which also had pointers and higher-order pointers, and probably was memory-unsafe as a result (though even with some Russian, I can’t find anything conclusive). The thing I eventually ran into is that Pascal itself has pointer arithmetic, and so is vulnerable to the same kinds of errors. Maybe it was better than C, which is fascinating, but not that much better.
Off-topic, that Springer paper was also pretty neat, just because it sheds light on how people thought about programming in 1979. For example:
I don’t see a lot of people denying that 2 is a good metric today. In fact, in the rare exceptions where someone has come right out and said it, I’ve suspected JS Stockholm syndrome was involved. Murphy’s law is very real when you not only have to write code, but debug and maintain it for decades as a large team, possibly with significant turnover. Early on they were still innocent of that, and so this almost reads like something a non-CS acedemic would write about programming.
once you understand C++ the pitfalls of C++ are reasonably consistent
there are like what, 3 operating systems these days?
assume those are all written entirely in c and combine them and compare that to all code ever written
All of C++? That’s unreasonable, it’s even in the name that it’s very expansive. Yes, if you already know a thing, you won’t be surprised by it, that’s a tautology.
C is more than just pointers, obviously, but the vast majority of the difficulty there is pointers.
Plus all previous operating systems, all supercomputer climate, physics and other science simulations, all the toaster and car and so on chips using bespoke operating systems because Linux won’t fit, every computer solving practical engineering or logistics problems numerically, renderers…
Basically, if your computational resources don’t vastly exceed the task to be done, C, Rust and friends are a good choice. If they do use whatever is easy to not fuck up, so maybe Python or Haskell.
similarly, “all of pointers” is unreasonable
pointers can have a lot of unexpected results
that’s literally why java exists as a language, and is so popular
sure, and the quantity of code where true low-level access is actually required is still absolutely minuscule compared to that where it isn’t
How? They go where they point, or to NULL, and can be moved by arithmetic. If you move them where they shouldn’t go, bad things happen. If you deference NULL, bad things happen. That’s it.
If you need to address physical memory or something, that’s a small subset of this for sure. It also just lacks the overhead other languages introduce, though. Climate simulations could be in Java or Haskell, but usually aren’t AFIAK.
What’s the point of having a function in the standard library if the universal recommendation is to never use it?
Is that the recommendation? This is the first time I’ve actually seen it discussed.
I’m wondering at this point if a new, different stdlib would be better. Or just use Rust.
To be honest, my comment probably applies more to
gets, but the point is the same.