In my opinion, catching a panic is a mistake in the first place. They’re meant to be unrecoverable anyway, so if you want the program to keep running, use Result types or return errors some other way.
Mutex poisoning isn’t really about catching panics. Mutex poisoning happens without any relation to
catch_unwind.The thing with mutex poisoning is that it is multi-threaded. One thread holds the mutex and panics. It does not catch the panic, the thread simply dies and releases the lock while unwinding. That marks the mutex as poisoned.
Another thread then tries to get the mutex and sees the poison error. This other thread is still alive since panics do not propogate among all threads. Panics only kill the thread the panic occurs on.
I guess I would have thought panic would trigger other threads to abort as well. Admittedly I haven’t really done any multithreading in Rust. I was thinking it was more like
exit(status)in C/C++ that immediately terminated the entire process. Exception unrolling in general I feel is something that should be avoided if you care at all about error conditions.If panics are isolated to one thread, then I guess mutex poisoning is a better default error case than a deadlock where the process never exits. An entirely independent thread might still be unaware the rest of the program has crashed though.
Panics in Rust default to stack unwinding, which I believe is similar to how exceptions work in C++. I don’t believe an unhandled exception in C++ kills other threads either? Could be wrong.
Panics can also be configured to instead abort the process immediately, which is the exit(status) way. Obviously with that solution you can never observe a poisoned mutex so you would never need a poisoning mutex in that situation anyway.
I don’t believe an unhandled exception in C++ kills other threads either? Could be wrong.
An unhandled exception in C++ causes the program to call std::terminate, which calls std::abort by default and brings down the entire program. I don’t think there’s a way to prevent an unhandled exception from doing that without causing nasal demons, but I could likewise be wrong. According to the standard, regarding the function function called by
std::terminate:A
std::terminate_handlershall terminate execution of the program without returning to the caller, otherwise the behavior is undefined.A quick check suggests that GCC’s
std::terminatecallsstd::abortif your terminate handler does not, but I don’t know if that applies to other implementations
I don’t quite get it, why should a mutex that was released because the thread panicked stay as it everthing is going just fine?
Because many (some would say most) use cases don’t have an actual window in their critical section where
- The invariant the Mutex protects is broken.
- There is a possibility of a panic.
Usually that’s simply not the case. You only need poisoning if you have the mutex unlocked and you are temporarily breaking its invariant before reconstructing the invariant and there’s a possibility of panic before that reconstruction occurs.
Even if your use case is like that, there’s a fairly good argument that
mutex.lock()should just automatically panic and should not requiremutex.lock().unwrap(), because most applications would never try to handle a poisoning error. They would just propogate the panic and try to shut down ASAP. If you really wanted to handle that poison, we could just have a different method for that, likemutex.try_poison_lock()or whatever. But the standard library unfortunately chose thatmutex.lock()should return a poison error.Ok, that goes more straight to the point than the article. I see the point now.
Mutex poisoning has been a recurring topic of debate in the Rust community so I think the article probably comes at it from a bit of an in-group perspective where it expects the reader to already have a bit of this context :)
It’s a good article and I agree, poisoning is a niche use case and shouldn’t be the default. That said it should document explicitly that it doesn’t handle panics in the critical section and that if you need that you should use a variant that does support poisoning. Basically the opposite of the situation as it currently exists where the default is poison with a note saying if you don’t need it use the non-poisoning (nightly only) version.
-
The only real argument for not poisoning is “performance” (at the expense of safety). Provide a real-world use-case where you think poisoning is supposedly the bottleneck, and I will (try to) show you that it isn’t.
-
I like how everyone is pretending writing your own extension trait is not a thing, where you can just add methods like panicking-lock() (with whatever short name you like), but also for backed-off lock (with try_lock() which is one atomic op and time-outs).
I’m having trouble finding any popular alternative implementations of
Mutexthat support poisoning.parking_lot,tokio, andfuturesall have non-poisoning (possibly async) mutexes.In fact, there’s almost no argument for requiring poisoning. There are some fringe cases where poisoning is useful, and in those cases, you could just as easily create a
Poisoning<T>pointer that implements it.As the article says, no other built-in pointer type in Rust support poisoning, yet the same “invalid state” argument can be made for all of them. A
&mut Tcan become temporarily “invalid” (in a sound manner) during a critical block that panics, for example. That pointer could be mutating data that’s eventually dropped during unwinding, does not hold its invariants at this time, and triggers aDropimpl that’s unsound as a result. At the end of the day, the correct way to handle this is to somehow either fix the state during unwinding so that it no longer violates an invariant, or mark it as poisoned, regardless of pointer type.The only real argument for poisoning by default is because it already is the default. It’s a strong argument of course (it would likely be a breaking change to remove it), but it’s not a very solid argument for why it should be a default to begin with.
I’m having trouble finding any popular alternative implementations of Mutex that support poisoning.
I don’t think performance is the bottleneck. The real reason to not use poisoning mutexes is convenience and developer experience. The standard library should take such considerations heavily into account. A rare use case such as poisoning being requried should not be the default for mutexes in the standard library.
The real reason to not use poisoning mutexes is convenience and developer experience. The standard library should take such considerations heavily into account. A rare use case such as poisoning being required should not be the default for mutexes in the standard library.
What does that even mean?
The interface concerns are as trivial as it gets:
#![feature(nonpoison_mutex)] #![feature(sync_nonpoison)] use std::sync::Mutex as PMutex; use std::sync::nonpoison::Mutex as NoPMutex; trait LockExt<'a> { type Guard; fn l(&'a self) -> Self::Guard; } impl<'a, T: 'a> LockExt<'a> for PMutex<T> { type Guard = std::sync::MutexGuard<'a, T>; fn l(&'a self) -> Self::Guard { self.lock().expect("non-poisoned") } } impl<'a, T: 'a> LockExt<'a> for NoPMutex<T> { type Guard = std::sync::nonpoison::MutexGuard<'a, T>; fn l(&'a self) -> Self::Guard { self.lock() } } fn main() { let m_p = PMutex::new(false); let m_no_p = NoPMutex::new(false); *m_p.l() = true; *m_no_p.l() = true; dbg!(*m_p.l()); dbg!(*m_no_p.l()); }This is basically is the one thing that shouldn’t consume any mind tokens when it comes to synchronization primitive choice.
I’m not really sure what you’re trying to say here. Are you saying that this is not a problem and everyone should just add this LockExt trait to all their code bases that uses Mutexes? Or…?
Every non-trivial code-base already has a
utilsmodule (sometimes a whole crate even) containing, in part, extension traits (and sometimes macros) to make some usages of some types nicer, in a way that is specific to the codebase (even in this case, a codebase might want a debug or trace log message for example, alongside the panic).It’s a part of the language for a reason. It provides full idiomacity. And this is not limited to (poisoned or otherwise) synchronization primitives.
Talking about a method returning a
Resultlike it’s a big deal must be some sort of splitting-hairs maxxing.And btw, as it just happens, such traits actually already exist, in a crate that is well-known to the ecosystem.
in a way that is specific to the codebase
Well I would say this is not at all specific to the codebase - as I said above, it applies to basically all code bases that use mutexes, since poisoning is so rarely relevant.
These two statements are kind of strange to me:
Talking about a method returning a Result like it’s a big deal must be some sort of splitting-hairs maxxing.
and
And btw, as it just happens, such traits actually already exist, in a crate that is well-known to the ecosystem.
If this is such a small deal, then using an external dependency is a terrible solution. That’s like the whole left-pad situation, a dependency to solve an extremely small problem, but something that affects many, many code bases. That’s exactly the use case that the standard library should cover, exactly so that you don’t need to use an external dependency for such a trivial thing!
And yea you could say “duh just write the extension trait in your utils module” but this is just not acknowledging the problem at all and just sidestepping it. That’s not helping anyone.
-
Notice the lack of any benchmarks for any real load from a real use-case.




