So you can’t get a Rust program to segfault without trying really hard. I haven’t observed a single segfault in the normal Rust code I wrote in the past 8 years.
The code used in cve-rs is not that complicated, and it’s not out of the realm of possibility that somebody would use lifetimes like this if they had just enough knowledge to be dangerous.
I’m as much a rust evangelist as the next guy, but part of having excellent guard rails is loudly pointing out subtle breakages that can cause hard to diagnose issues.
No, I’m saying that when people run into strange bugs, sometimes they put together an issue (like the person behind cve-rs), and sometimes they quietly work around it because they’re busy.
Seeing as I don’t often trawl through issues on the language git, neither really involve notifying me specifically.
My lack of an anecdote does not equate to anecdotal evidence of no issue, just that I haven’t met every rust developer.
Sure. I haven’t seen a proper segfault from any modern, post-C/C++ language in ages. I’ve never seen a Go program segfault, or a Nim one (although, there are comparatively few of those as a sample size).
So, it seems to me that - purely from the perspective of a user of programs - Rust still seems about as safe as any other modern language - since I’ve seen no other modern (say, created in the past decade) compiled language segfault. Even the C segfaults seem to be largely becoming rare occurrences, which I have to chalk up to better tooling, because I highly doubt that there’s been some magical increase in general C programmer quality in the intervening years.
Yes, the problems rust is solving are already solved under different constraints. This is not a spicy take.
The world isn’t clamoring to turn a go app into rust specifically for the memory safety they both enjoy.
Systems applications are still almost exclusively written in C & C++, and they absolutely do run into memory bugs. All the time. I work with C almost exclusively for my day job (with shell and rust interspersed), and while tried and tested C programs have far fewer memory bugs than when they were first made, that means the bugs you do find are by their nature more painful to diagnose. Eliminating a whole class of problems in-language is absolutely worth the hype.
Everything is slower than C (I haven’t seen a benchmark yet where a language bests C; even hand-crafted ASM ceded the high ground decades ago when compilers got better than human assembly programmers), but then, C compiler technology has had literally 40+ years to mature.
Go and Java (once warm) do pretty well, but absolutely give up execution speed for coding simplicity and (in Go’s case, anyway) speed. Nim is young; I’m curious to see how it matures. They’re having a bit of a performance crisis at the moment, but assuming they get past that it seems like a fair middle ground between Go’s simplicity and Rust’s bare-metal performance. Then again, manual memory management was absolutely my least favorite thing about C and is what eventually drove me away; worst. Boilerplate. Ever. Even worse than Go’s error handling (which they almost fixed and looks like will be addressed within the next free releases). Anyhoo, going back to that shit is going to be a hard pill to swallow.
Rust is still having its honeymoon, and is the hip language of the decade now. We’ll see!
Arguably when LLVM gets a bit better, Rust can be even faster than C because rust can be optimised in more places safely than C code can. The issue is that LLVM wasn’t written with that in mind, so some performance is left on the table.
I don’t know if you’re talking about panics and abort or about crashes caused by memory safety errors. The latter class is very unlikely in safe rust, other than as rare compiler bugs. Panics and aborts are your call. You can easily write code that doesn’t panic or abort.
As a user. I don’t write Rust, but lots of programs I use do and, as I said, they seem to crash about as much as any other compiled language tools I use are written in. I almost never seesegfaults; I can’t say I’ve ever seen one in a Go program, and I use a bunch on those.
If we’re only talking about segfaults, the only language I can remember seeing doing that has been C, or C++. If not doing segfaults is what makes a language “safe,” then it seems to me most modern languages are as safe as Rust. If we include crashes, then as I said, I see Rust programs crashing about as much as any other proglang.
This is definitely into the territory of misinformation.
I don’t write Rust, but lots of programs I use do and, as I said, they seem to crash about as much as any other compiled language tools I use are written in
I already addressed this before. Regular crashes are almost always (I can’t remember any exceptions) due to panics or aborts chosen by the user - especially due to unwraps. Using that to equate Rust programs’ stability to ‘any other compiled language tools I use are written in’ is very disingenuous - because it’s just as easy to handle those errors and prevent a crash at all.
If not doing segfaults is what makes a language “safe,” then it seems to me most modern languages are as safe as Rust
You are unnecessarily conflating issues here. ‘Most modern languages’ are not a replacement for what C, C++ and Rust can do. Go most famously had to retract their ‘systems programming language’ tag, for example. If a GC language meets your requirements - then by all means, use it. But it’s not without reason that many companies have rewritten even their web backends in Rust. Memory safety without GC is a very big feature that a lot of professionals care about. It’s not something to dismiss as trivial.
And while at it, you neglecting what segfaults represent. It’s just a benign example of memory safety bug. It’s benign because it gets caught causes the program to crash. There are a whole lot of them that causes the program to continue running - causing serious vulnerabilities. This is why even the US government and agencies recommend memory safety languages and especially Rust if performance and other limitations matter.
If we include crashes, then as I said, I see Rust programs crashing about as much as any other proglang.
I really don’t want to repeat the reason twice in a single comment and 3 times including in my previous comment. But the only way you are going to make Rust crash as much as ‘any other prolang’ is to neglect idiomatic Rust. That isn’t surprising because crashing anything is possible if that’s your intention.
I think you’re missing the point where I said I don’t write Rust. I’m merely making an observation - as a user who’s more than usual, probably, aware of which language any given tool is written in, of what I observe.
I know I’ve seen a Rust program segfault, and I’m certain that I saw this only once. I’m equally confident that almost every other segfault has been from C/C++ code, but not other languages.
What I’m hearing you say is that crashes aren’t a risk, per se, as long as they aren’t memory related and so the crashes I see from Rust programs don’t count because they aren’t segmentation faults. Did I read that right? And I also hear you claiming that, e.g., NPEs are security issues, even if the runtime catches them and safely exits the program in a controlled manner (“crashing out” safely) - is that right?
That’s misinformation. There’s no overestimation. The problem is so bad that even the US government advocates the use of memory safe languages (including GC languages).
I have used C and C++. You need laser sharp focus to avoid memory safety errors even after you learn what causes them and how to avoid them. It’s significantly easier to write programs in Rust because any lapse in care to avoid memory safety bugs are caught by the compiler.
What I mean is that even if you use GC languages like Java or Go you will still encounter annoying bugs. I’m not saying that memory safety isn’t important.
You said bugs caused by ‘memory problems’. And that Rust programmers vastly overestimate them. Those aren’t generic logical bugs that you get in Go or Java. And Rust never claimed to solve logical bugs.
Yeah. The verdict is still out on whether having a deeply surly compiler will help me focus on iterating and understanding the client’s needs.
I run Python CICD controls on main with at least the same level of prissiness (as Rust comes with), but at least Python knows how to shut up and let me prototype.
I’m currently not convinced that Rust’s opinionated design hits a useable long term sweet spot.
But I think if Rust adds a debug flag --fuck-off-i-need-to-try-something, it could genuinely become the next Python, and the world would be better for it.
Edit: And if I just missed the --fuck-off-i-need-to-try-something Rust flag, someone point me at it, and I’ll gladly give Rust another run.
That flag exists, it’s called unsafe for if you need to tell the borrow checker to trust you or unwrap if you don’t want to deal with handling errors on most ADTs.
You can always cast anything to an unmanaged pointer type and use it in unsafe code.
I’m sorry to hear that. I think at one point in my past, about half my job was tracking down nil dereference errors in Ruby. And probably a quarter was writing tests for things a good type system would catch at compile time.
And I’m sorry to hear about that Ruby experience. I authored one of the Ruby stdlibs, and similar issues with the language, and the inevitable encroachment of Rails into every project, eventually drove me away from it.
I was, however, excluding interpreted languages from my comparison. Dynamically typed languages are a different matter and can’t hope to be any kind of safe - but that’s not the game where they excel.
If you can make safe Rust segfault you’re doing something wrong.
To be fair, you are doing something wrong if you’re app segfaults no matter what anguage you wrote it in…
Despite that, some languages make it easier to be wrong than others.
I don’t think so, since memory safe languages are supposed to prevent you from doing that, so it would be the language implementation’s fault.
It actually is possible to segfault in safe Rust, although it is considered a bug. Proofs of concept are shown in this cve-rs crate.
If you want an explanation of why this happens, I recommend this video: https://youtu.be/vfMpIsJwpjU
So you can’t get a Rust program to segfault without trying really hard. I haven’t observed a single segfault in the normal Rust code I wrote in the past 8 years.
The code used in cve-rs is not that complicated, and it’s not out of the realm of possibility that somebody would use lifetimes like this if they had just enough knowledge to be dangerous.
I’m as much a rust evangelist as the next guy, but part of having excellent guard rails is loudly pointing out subtle breakages that can cause hard to diagnose issues.
If that’s so trivial to trigger, people would be doing so everywhere. Have you seen it in the wild or heard of anyone doing so?
If someone did, why would I hear of it?
So you’re just spouting some cooked up theory. Not a practical scenario.
deleted by creator
You don’t have any evidence, much less anecdotal evidence. Things don’t become true just because you insist.
No, I’m saying that when people run into strange bugs, sometimes they put together an issue (like the person behind cve-rs), and sometimes they quietly work around it because they’re busy.
Seeing as I don’t often trawl through issues on the language git, neither really involve notifying me specifically.
My lack of an anecdote does not equate to anecdotal evidence of no issue, just that I haven’t met every rust developer.
IME Rust programs crash at about the same rate as other languages. “Rewrite everything in Rust” hasn’t made much of a difference for me, so far.
A crash is different to a SEGFAULT. I’d be very surprised to see a safe rust program segfault unless it was actively exploiting a compiler bug.
Sure. I haven’t seen a proper segfault from any modern, post-C/C++ language in ages. I’ve never seen a Go program segfault, or a Nim one (although, there are comparatively few of those as a sample size).
So, it seems to me that - purely from the perspective of a user of programs - Rust still seems about as safe as any other modern language - since I’ve seen no other modern (say, created in the past decade) compiled language segfault. Even the C segfaults seem to be largely becoming rare occurrences, which I have to chalk up to better tooling, because I highly doubt that there’s been some magical increase in general C programmer quality in the intervening years.
Yes, the problems rust is solving are already solved under different constraints. This is not a spicy take.
The world isn’t clamoring to turn a go app into rust specifically for the memory safety they both enjoy.
Systems applications are still almost exclusively written in C & C++, and they absolutely do run into memory bugs. All the time. I work with C almost exclusively for my day job (with shell and rust interspersed), and while tried and tested C programs have far fewer memory bugs than when they were first made, that means the bugs you do find are by their nature more painful to diagnose. Eliminating a whole class of problems in-language is absolutely worth the hype.
Go, Java, and Nim (in most cases) are all memory safe but are generally slower than C or C++ due to the ways they achieve memory safety.
Rust’s memory safety approach is zero-cost performance wise, which makes it practical for low level, high throughput, and low latency applications.
Everything is slower than C (I haven’t seen a benchmark yet where a language bests C; even hand-crafted ASM ceded the high ground decades ago when compilers got better than human assembly programmers), but then, C compiler technology has had literally 40+ years to mature.
Go and Java (once warm) do pretty well, but absolutely give up execution speed for coding simplicity and (in Go’s case, anyway) speed. Nim is young; I’m curious to see how it matures. They’re having a bit of a performance crisis at the moment, but assuming they get past that it seems like a fair middle ground between Go’s simplicity and Rust’s bare-metal performance. Then again, manual memory management was absolutely my least favorite thing about C and is what eventually drove me away; worst. Boilerplate. Ever. Even worse than Go’s error handling (which they almost fixed and looks like will be addressed within the next free releases). Anyhoo, going back to that shit is going to be a hard pill to swallow.
Rust is still having its honeymoon, and is the hip language of the decade now. We’ll see!
Rust is roughly similar to C in most of these benchmarks and beats it in a few: https://benchmarksgame-team.pages.debian.net/benchmarksgame/fastest/rust.html
Arguably when LLVM gets a bit better, Rust can be even faster than C because rust can be optimised in more places safely than C code can. The issue is that LLVM wasn’t written with that in mind, so some performance is left on the table.
I don’t know if you’re talking about panics and abort or about crashes caused by memory safety errors. The latter class is very unlikely in safe rust, other than as rare compiler bugs. Panics and aborts are your call. You can easily write code that doesn’t panic or abort.
As a user. I don’t write Rust, but lots of programs I use do and, as I said, they seem to crash about as much as any other compiled language tools I use are written in. I almost never see segfaults; I can’t say I’ve ever seen one in a Go program, and I use a bunch on those.
If we’re only talking about segfaults, the only language I can remember seeing doing that has been C, or C++. If not doing segfaults is what makes a language “safe,” then it seems to me most modern languages are as safe as Rust. If we include crashes, then as I said, I see Rust programs crashing about as much as any other proglang.
This is definitely into the territory of misinformation.
I already addressed this before. Regular crashes are almost always (I can’t remember any exceptions) due to panics or aborts chosen by the user - especially due to unwraps. Using that to equate Rust programs’ stability to ‘any other compiled language tools I use are written in’ is very disingenuous - because it’s just as easy to handle those errors and prevent a crash at all.
You are unnecessarily conflating issues here. ‘Most modern languages’ are not a replacement for what C, C++ and Rust can do. Go most famously had to retract their ‘systems programming language’ tag, for example. If a GC language meets your requirements - then by all means, use it. But it’s not without reason that many companies have rewritten even their web backends in Rust. Memory safety without GC is a very big feature that a lot of professionals care about. It’s not something to dismiss as trivial.
And while at it, you neglecting what segfaults represent. It’s just a benign example of memory safety bug. It’s benign because it gets caught causes the program to crash. There are a whole lot of them that causes the program to continue running - causing serious vulnerabilities. This is why even the US government and agencies recommend memory safety languages and especially Rust if performance and other limitations matter.
I really don’t want to repeat the reason twice in a single comment and 3 times including in my previous comment. But the only way you are going to make Rust crash as much as ‘any other prolang’ is to neglect idiomatic Rust. That isn’t surprising because crashing anything is possible if that’s your intention.
I think you’re missing the point where I said I don’t write Rust. I’m merely making an observation - as a user who’s more than usual, probably, aware of which language any given tool is written in, of what I observe.
I know I’ve seen a Rust program segfault, and I’m certain that I saw this only once. I’m equally confident that almost every other segfault has been from C/C++ code, but not other languages.
What I’m hearing you say is that crashes aren’t a risk, per se, as long as they aren’t memory related and so the crashes I see from Rust programs don’t count because they aren’t segmentation faults. Did I read that right? And I also hear you claiming that, e.g., NPEs are security issues, even if the runtime catches them and safely exits the program in a controlled manner (“crashing out” safely) - is that right?
Rust programmers vastly overestimate how many bugs are caused by memory problems
That’s misinformation. There’s no overestimation. The problem is so bad that even the US government advocates the use of memory safe languages (including GC languages).
I have used C and C++. You need laser sharp focus to avoid memory safety errors even after you learn what causes them and how to avoid them. It’s significantly easier to write programs in Rust because any lapse in care to avoid memory safety bugs are caught by the compiler.
What I mean is that even if you use GC languages like Java or Go you will still encounter annoying bugs. I’m not saying that memory safety isn’t important.
You said bugs caused by ‘memory problems’. And that Rust programmers vastly overestimate them. Those aren’t generic logical bugs that you get in Go or Java. And Rust never claimed to solve logical bugs.
I’m more poking fun at Rust enthusiasts who make it seem like programs are automatically bug free just because they’re written in Rust.
Haskell enthusiasts made similar claims as well.
I don’t know, I’ve caused a whole lot in C/C++. I haven’t actually written anything in Rust either, so I’m somewhat unbiased.
Amen. But they have to, because that’s where they spend most of their cognitive focus, and it’s Rust’s “killer feature.”
Yeah. The verdict is still out on whether having a deeply surly compiler will help me focus on iterating and understanding the client’s needs.
I run Python CICD controls on main with at least the same level of prissiness (as Rust comes with), but at least Python knows how to shut up and let me prototype.
I’m currently not convinced that Rust’s opinionated design hits a useable long term sweet spot.
But I think if Rust adds a debug flag
--fuck-off-i-need-to-try-something, it could genuinely become the next Python, and the world would be better for it.Edit: And if I just missed the
--fuck-off-i-need-to-try-somethingRust flag, someone point me at it, and I’ll gladly give Rust another run.That flag exists, it’s called
unsafefor if you need to tell the borrow checker to trust you orunwrapif you don’t want to deal with handling errors on most ADTs.You can always cast anything to an unmanaged pointer type and use it in unsafe code.
Thank you. I’ll check it out next time.
I’m sorry to hear that. I think at one point in my past, about half my job was tracking down nil dereference errors in Ruby. And probably a quarter was writing tests for things a good type system would catch at compile time.
And I’m sorry to hear about that Ruby experience. I authored one of the Ruby stdlibs, and similar issues with the language, and the inevitable encroachment of Rails into every project, eventually drove me away from it.
I was, however, excluding interpreted languages from my comparison. Dynamically typed languages are a different matter and can’t hope to be any kind of safe - but that’s not the game where they excel.