I think Rob Pike even said it's easy to see where a program fail in one of his talks?
But to me the superb thing about exceptions, is that error handling can be done where it makes sense. I.e. we can try{ problem-code }catch(problem){ handle problem } in a single location. Otherwise we end up peppering the entire code base with a ton of error checking far down the call stack, where we really cannot do much about the problem anyway (unless we are writing command line tools where error handling is just writing the problem to stderr).
Exceptions gives us a nice way to let problems bubble up to the surface, while also stating what the problem was, and where it occurred. That is great IMO.
Work is ongoing on some of this, but there are popular libraries in Rust (like `anyhow`) that let you attach backtraces to regular errors, add context, etc. Propagating erros to callers is handled with the standard `?` operator, which means "short-circuit and return this if it was an error, otherwise give me the successful result". This has the benefit of making early exits explicit, without interrupting the visual flow of straight-line code.
Because Rust uses the type-safe explicit Result/? approach for all non-bug failures in the program, the implicit panic (that behaves similarly to RuntimeException in Java) is reserved for assertion failures and crashes only.
`catch_unwind` is not guaranteed to work in Rust. There's a setting to disable it and always hard abort() the whole process on every panic. Rust is serious with panics being for programmer's bugs only, and not trivialities like "file not found".
I have to say I really enjoyed Rust when it was in its infancy (version 0.1 - 0.2 or thereabouts), but have since fallen off. It used to be so simple and so clean, and unlike anything else. Today it's just way to complex for me :-)
Not necessarily. With exceptions, it is easy to be a cause of error and just throw the exception, then expect up the stack to handle it. Which of course has no idea how, it didn't control the cause in the first place.
Forcing error handling as near as where error can happen prevents this.
The Rust/Go approach always makes me laugh. Normally in engineering or anything where reliability matters, panicking is understood to be a bad thing to do and people go through extensive training to ensure they don't do it. Somehow these language communities decided that panicking and giving up on the spot is a smart behaviour.
Oh, come on, that's a straw man.
Just because panic! exists as a "abort-program-with-a-message", does not mean it's somehow encouraged above using idiomatic error handling.
Just as you can do the same thing in languages with exceptions. Sometimes exiting the program is the right thing to do.
Sometimes exiting the program is the right thing to do.
Yes, but it's very rare that the code where something went wrong is in the position to decide that. The survival of the entire process is not a decision to delegate to every possible line of code or library author.
Consider a very common case where I benefit from exceptions every day - my IDE. IntelliJ hosts a bazillion plugins, of varying quality. These plugins like to do very complex analysis on sometimes broken and incoherent snippets of code, that may be in the process of being edited. In other words it's a nightmare to correctly predict everything that can go wrong.
Not surprisingly, sometimes these plugins crash. And you know what? It doesn't matter. A lot of the code is just providing optional quality-of-life features like static analysis. If one of them goes wrong, IntelliJ looks at the exception and figures out which plugin is likely to blame, it examines the type of error and maybe gathers editor context, it can report it easily to a central server that then groups and aggregates exceptions based on stack traces. Meanwhile as a user, it doesn't bother me because it's fine to just not have that analysis show up in the editor.
If every time an IDE plugin encountered an unexpected situation it aborted the entire process it'd be insane. The plugin ecosystem could never scale that way. People would be afraid of installing/upgrading plugins and that in turn would discourage people from writing them or adding features to them.
What if a plugin developer used `System.exit(-1)` in their catch block. How's Intellij going to handle that?
A very popular eclipse plugin did this in past and would bring down the entire IDE when a particular exception happened.
Yes, in theory, there are all sorts of ways you can still trash the process with bad code. But in practice, the sorts of bugs that programmers really make in GC-d memory-safe languages are the ones that don't. So, exception based error handling really does come in very useful and Rust probably got it wrong here.
Stated like that, who can really disagree?
I remember when I was writing a bunch of Go when the language was still very new (2009 - 2011). One of the most popular use cases for the language was making websites. All sorts of unexpected problems caused the entire website to go down, due to unexpected panics here and there. The suggested solution from the Go team was to just restart the web-server whenever it was killed by a panic. Surely that cannot be the best way to do it..
This is not really true. If you are indexing into something that may fail you use the `get` method which returns an `Option` if the index is out of bounds. The index operator is just a shorthand for `v.get(i).unwrap()` pretty much.
https://doc.rust-lang.org/std/primitive.slice.html#method.ge...
The panic mentality comes from people who have spent most of their life writing C++, in which if anything goes wrong like an out of bounds index, memory might be corrupted in arbitrary ways, and in which you don't have a GC to clean up after you. Writing exception safe code is much easier in type safe GCd languages, and many programming errors end up being recoverable.
Agree, this can happen. Perhaps the bad attempt at fixing this in Java for instance - checked exceptions, made people dislike exceptions ever more. The caller "has to handle" the exceptions or re-throw them of course. Even though RuntimeException's can come from anywhere at anytime, so "guard" provided by checked exceptions just made a complete mess of things. People are lulled into thinking that methods without the 'throws BlaBlaException' signature are safe and so on.
I guess no language is 100% on everything, but I've always felt that exceptions are one thing I really like; especially when a language manages to do them correctly.
In other words: there’s a bug in the code, and that bug has now caused an unrecoverable error, panicking the thread. Now the thread has died (or maybe caught the panic to present a friendly error message). Either way, the user is now aware of the bug, and disaster has been avoided.
Of all situations where your app might want to create a backup of its state, why would you choose to do so precisely while unwinding a crashed thread, where all assumptions, bets and invariants are already off?
And what would the helpful error dialog even say? „A problem has occurred and the app will now shut down“? From the user’s point of view, is that really an actionable or helpful error dialog?
Yes, literally. This is already much better than anything that gets the spinning ball of death of macOS going. You can even continue if you are running an event-driven app where the error may have happened as part of an event handler (and thus limited to a very specific part of the software).
To give my own experience: I develop https://ossia.io and use this catch-all method. In 7 years I've gotten numerous thanks from the users for it not crashing but being able to carry forward in case some issue crops up in a sub-sub-sub-module. Not a single time I remember this to cause some memory corruption later.
(backing up state is done up to the previous user action but while in my case it works, it's not always practical)
I suspect in reality most of the problems this would catch in OSSIA wouldn't end up as panics in a hypothetical "Rust OSSIA" because of the different attitudes to exception throwing/ panic vs "normal" error flow in these languages and libraries - unless you got really happy slapping "unwrap()" on things when you shouldn't, but sure, it would solve this problem.
As to memory corruption - the problem isn't strictly "memory corruption" but unstable system state. If my underlying cause is that somebody's dubious Leslie simulator blows up when I frob the gain control on it too quickly, restoring exactly the state in which it blew up last time doesn't help me on its own. I need some way to say OK, that was crazy, no Leslie simulator until I save the project and then we can take it gently, which again is somewhere the thread solution is nicer.
Reality is not that simple, if you worked in this industry you would know. For example I was building a web scraper years ago and the WebView would crash since is C/C++ , instead of doing it's job and show a web page or a broken web page it crashed my entire program,. The solution was to split my program in a parent program and a child program so this bug does not bring my entire thing down, and I can crash the issue and record the bad url that crashes and try again or just skip it.
I would hate to use Rust libraries that would crash my entire program if they for some reason are bugged. In my experience I found bugs in many popular libraries. So in Rust if I import a say library to resize an img and say the img is corrupted and library is shit it will crash my entire program? I would prefer a higher language where I can try=crash the image resize function and if shit goes wrong I can show the user a relevant message , or fallback to some other resizing method.
What you're describing is the `catch_unwind` mechanism that Rust does have. Because panics are implemented with unwinding (by default), you can catch them. But it's not the normal error handling mechanism; it's the "oh god an assert just failed, or we just OOMed or something, who knows, most bets are off" mechanism. If you have a main loop that's sufficiently isolated from individual tasks, such that you think you can do something useful with the fact that one of your tasks just vanished in a puff of smoke, then catching a panic coming out of a task might be a reasonable thing to do. That often makes sense in server code, where your main loop might want to keep trucking, or at least gracefully shut down other connections. But for most library code, the right thing to do is to allow most panics to propagate and crash the caller.
YEs it happen to me many times to hit bugs when working in real world, bugs in image libraries, bugs in regex libraries, bugs in pdf libraries, bugs in html/xml parsers so from my experience working with c/c++ and higher level languages I prefer the higher level languages, less bugs, almost no complete crashes and better error reports from the exceptions. I never had the tiem to try Rust but I am not tempted so far.
Nope, we just check the return result because libraries usually don't crash and have well-defined error cases. Having a decent type system helps catching all the possible outcomes. In a few years coding Rust, i never had a single crash due to a library panic, only from explicit unwraps i applied in my own codebase.
Panics are not intended for errors, but for unrecoverable failures. For example, in rust std a failing memory allocation will crash your whole program, which is in most cases what you want to do. For the remaining cases, there are other non-fallible methods.
For example: String::reserve vs String::try_reserve or HashMap::insert vs HashMap::try_insert.
Sure, there are bugs in every code but unexpectedly panicking is considered a bug in a library so in my not too extensive experience with rust libs, these are not the norm at all. So simply writing code where you yourself don’t panic should give you quite a high chance of not hitting this case ever.
Yes so would you like say Firefox to just crash when one of it's many dependencies crashes?
You are suggesting but I am not sure if I understand correctly that only memory errors cause panics? So what if the library reads a file and unexpectedly it shit happens with the file, it will crash the program because the developer maybe forgot to return a special error code in this case,
I'm not super up to speed on JS, but I might draw an analogy to Python. Handling a result in Rust is similar to catching an exception of a known type in Python, a very common thing to do. On the other hand, catch_unwind is (loosely) similar to writing a bare except clause that catches every conceivable exception. You can do that, and sometimes it's correct to do that, but in most cases it's a bad idea. You don't want to accidentally suppress errors that indicate a bug in your program.
This days doing backend dev I am forced to move stuff in a different process but most stuff I use I prefer to use binaries then libraries , for example for resizing an image instead of using the built in image library that crashes sometimes and brings the script down I install image magic on the server and write a script for resizing an image then call that script and check it's output , sometimes I had to use the timeout linux program to kill the program if it gets stuck on some input file.
If I were to create that image resize library in Rust I would attempt to catch everything , including panics and return it as a error result(so only system crashes would be uncaught)
* Entering an infinite loop can bring down everything. A separate thread might not, but since Rust provides no way to kill a thread without it cooperating, there is no way to stop a stuck thread without bringing down the whole process.
* Stack overflow is an instant abort, not a panic.
* Double panic, where panicking calls a destructor that itself panics, is an instant abort.
- Extremely common operations with dedicated syntax, where introducing error handling would be burdensome. Things like array indexing or arithmetic overflow. In these cases, you usually want an alternative, fallible way to do the same operation.
- Cases where most callers will probably convert the error into a panic anyway. One example of this might be .split_at() on slices, which is bounds-checked just like an array access. Most callers would probably just .unwrap() the out-of-bounds case, and callers who don't want it to panic can easily check before the call, so it's more ergonomic to panic.
- Cases where the only plausible reason for failure is a bug in the caller. For example, the .borrow() and .borrow_mut() methods on RefCell will panic if a write overlaps with another read or write. The caller is almost always expected to statically guarantee that that doesn't happen, usually by making all borrows short-lived. (And here again there are fallible alternatives available.)
An interesting example of something that doesn't panic, but which probably should, is taking a mutex. The standard mutex in Rust includes a "poisoning" mechanism, which almost every caller just .unwrap()s. I think the majority opinion these days is that poisoning should just be removed, but given that it's around I think most people wish it just panicked instead of returning a Result.
That’s essentially it, yes. My code should never actually panic. If it does, it means the state of the process has become deeply diseased, and attempting to “clean up” is likely to just make things worse. Of course, if it’s safe Rust, then it still won’t write past the end of a buffer or anything disastrous like that, but buggy code is still buggy code and there’s lots of stuff Rust won’t stop you from doing.
One of the more extreme things I’ve done in production Rust code was add a “watchdog thread”. It has a channel that takes unit and receives on a timeout, and the thread doing the actual work is expected to send it a message once a minute. If it doesn’t receive a message within a minute, it hard aborts the process. The default setup is run under a service manager like systemd to make sure it gets restarted, and that failures are actually logged somewhere.
This is meant to solve the problem that safe Rust is a Turing complete language, so is subject to the halting problem. The type checker can prove that you won’t read past the end of a buffer, but it cannot prove that your code will ever actually finish running. Which means, if you have a project like a web scraper that needs high uptime, you need to prevent it from getting stuck somehow.
There are always bugs(logic bugs where Rust can't protect you) so why not have a clean interface?
A good developer will crash the program, as soon as possible, if and only if it has a bug. If you want to write a program that never crashes, then you need to write a program with no bugs.
The reason you don’t want buggy code to limp along after it detects a bug, is that crashing isn’t the worst possible thing.
The worst possible thing is getting stuck in an infinite loop or a deadlock.
You would, but for other programs with other requirements it would actually be beneficial. There's no single right answer, and you should pick the library that follows your particular requirements for that particular program.
I swear, sometimes it seems like Rust people are from another planet. What do you think "unwrap" does? It's not used in every library, but certainly in many of them.
There’s no need to talk in a condescending manner.
What they said was correct - an “unwrap” outside of test/prototyping is considered a serious bug. They Rust loving strawmen you’re creating never claimed that every line of Rust ever written is perfect and bug free.
In the end you don't use Rust because it's so easy and nice to use (unless you come from C/C++). You come to Rust because you want meticulous control over performance, and you don't want to sacrifice safety to attain that.
If that's not why you're using it, I agree you're probably better off choosing Java, it's plenty fast and comfortable to use, especially if you pick modern tooling.
That statement really resonates with me. If you use a library, you’re responsible for what it does, just like how you’re responsible for your own code.
Are you mistaking Rust for some other language? Error handling in Rust is mostly just using the `?` operator.
The Rust book explains - https://doc.rust-lang.org/book/ch09-02-recoverable-errors-wi...
something.map_err(…)? is quite readable in my opinion and that is the worst case, when your method returns a Result<..,..> but the called method has an Optional return type. Otherwise it is just a single ?.
Sure, I do believe that exceptions are superior but we do have to understand that rust is a low-level language, period. It is very expressive considering its nature, but it will never be as productive as a managed language in my opinion - we have this distinction for a very good reason. If you want maximal control over what happens “behind the scenes” you loose some automatism that could improve productivity.
Remarkable that people with such little knowledge feel comfortable talking so much.
Many programmers are writing code for sunny weather only, with error handling being something you might add as an afterthought if your code starts to feel a little too brittle.
In my eyes error handling is just as important to do correctly as getting the core of the functionality done, because error handling is a core functionality of any program, especially if we speak of libraries others are meant to use.
Error handling is what differenciates engineering from coding.
Consider serde-json, a widely used library to serialise and deserialise json. You asked me to find “one Rust repo”. Ok here it is - https://github.com/serde-rs/json/search?q=unwrap&type=. Of the 22 uses of unwrap, nearly all are in test code or in comments. Of the remaining 3 or 4, they seem safe to me. But maybe they’re not. Could you think of some json input that could trigger a panic from one of those unwraps?
I’ll put my money where my mouth is. I’ll donate $100 to a charity of your choice if you can find that.
But if you can’t, at least have the honesty to admit that you misspoke when you said not even a single repo without “excessive” use of unwraps exists.
It's just like the IndexOutOfBounds exception in Java. Many functions can theoretically throw it, but most libraries and programs do not catch it because usually if it is thrown it means that something happened that the programmer did not expect and therefore the program should crash.
The assertion that "If a library panics on user data, that's a pretty serious bug" remains true.
If a library is panicking on invalid user data, it is because they are abusing panic, which is a serious bug. Or they just didn't realize that their code could panic, which is also a serious bug.
It panics just like my `my_vector[2]` example does. What did you think `my_vector[2]` did? Libraries use `my_vector[2]` too. I don't get why we're changing topics from one commonly used panicking operation to another.
I've been programming in Rust since it came out, and a couple of those years professionally, and I don't think I've ever seen anyone use catch_unwind. Maybe once in a test case?
To be concrete, let's talk about an example of a panic. Say you want to access the 3rd element of a vector. There are two cases:
1. You're not sure whether the vector actually has three elements on not. In this case, you call `my_vector.get(2)`, which returns an Option, and you handle the case where it's present and the case where it's not. This is standard error handling.
2. You are sure that the vector has at least three elements. Perhaps you just checked its length for some other reason, or you are careful to maintain this invariant, or you just constructed this vector by pushing 5 elements onto it. In this case, you would typically use `my_vector[2]`, which panics if the vector is too short.
For #2, the thing to notice is that this function literally never panics, under any input whatsoever if it is written correctly. Should that fact really clutter up its type signature, either by forcing it to return a Result type or by forcing it to have a "throws panic" marker?
EDIT: This is for a function that uses a possibly-panicking operation, `my_vector[2]`. There are also the functions that define a potentially panicking operation, like the vector indexing function itself. You could put a marker in the type signature of those, that would be reasonable. Though it would only be for users; the compiler wouldn't care.