Hacker News Reader

Comment by est31 | Hacker News Reader

> Overall Rust appears to be fulfilling one of its original design goals: allowing us to write more concurrent code safely. Both WebRender and Stylo are very large and pervasively multi-threaded, but have had minimal threading issues. What issues we did find were mistakes in the implementations of low-level and explicitly unsafe multithreading abstractions — and those mistakes were simple to fix.

> This is in contrast to many of our C++ races, which often involved things being randomly accessed on different threads with unclear semantics, necessitating non-trivial refactorings of the code.

View on HN

And when it comes to finding bugs in your unsafe concurrency primitives in Rust programs, there is a tool called Loom that helps with that, as a powerful complement to what ThreadSanitizer provides: https://github.com/tokio-rs/loom . You can think of Loom sort of like QuickCheck, except the permutations are all the possible thread states of your program based on the memory orderings that you have specified.

Here's a video that briefly talks about what Loom can do https://youtu.be/rMGWeSjctlY?t=7519 (it's part of a larger video about memory orderings, and is a great introduction to the topic).

Loom is really awesome, though it is focused on exhaustive testing, so not suitable for code that has a lot of possible interleavings (e.g. due to a ton of threads, or a large body of code).

There is a new project out of AWS called Shuttle [1] which is like Loom, but it does random exploration instead of exhaustive exploration, which enables massively distributed testing of really complicated stuff.

[1] https://github.com/awslabs/shuttle

Excellent, thanks for the tip! I have to ask though, given a tool that does exhaustive testing (like Loom), shouldn't it be trivial to adapt it to do random testing merely by randomizing rather than enumerating the test inputs? Is there something more going on that I'm not seeing?

Is there anything like Loom or Shuttle for C/C++?

Or are these only implemented in rust and they'd still work with C/C++?

I vaguely recall that Loom is an implementation of a research paper that purported to implement this for C/C++, but I don't know if that implementation is available/production-ready. In the worst case it should be relatively straightforward to back-port Loom to C++, since Rust also uses the C memory model.

The paper is here (referenced in the Loom docs.):

http://demsky.eecs.uci.edu/publications/c11modelcheck.pdf

This page describes how to get the code (linked from the paper):

http://plrg.ics.uci.edu/software_page/42-2/

I found some data races in an ODBC driver using TSan. Running it through CDSChecker is one of my round tuit projects.

Loom's style of exploration (and that of aws shuttle mentioned below) can be quite effective. Coyote is the equivalent library in the .NET world (formerly known as P#) and comes with a bunch of exploration "schedulers" (from random exploration to probabilistic and even an experimental reinforcement learning based one). They released an animation rich video today (see https://news.ycombinator.com/item?id=26718273). Worth watching even if you're using Rust as Loom and aws shuttle are conceptually similar.

Yeah, that was the part that stood out to me as well. I thought it was pretty intuitively obvious that having your data races confined to small segments is better than allowing them to be spread over the entire codebase, but a lot of people act like it negates the advantage of working in a safer language entirely. It's nice to have empirical reports that back up the "obvious" though. Sometimes "obvious" things turn out to be wrong. It's always good to check them.

> but a lot of people act like it negates the advantage of working in a safer language entirely.

I think the underlying assumption is more along the lines of: the easy cases are made easier, and the hard cases are made impossible — or rather the concern being that rust hasn’t made the problem easier, it just moved it around.

Which is often the case with claims of solving long-lasting problems

> the concern being that rust hasn’t made the problem easier, it just moved it around.

This is largely the point of Rust, though. Rust takes problems that used to manifest at runtime and causes them to manifest at compile-time, shortening the feedback loop drastically. It also takes problems that used to manifest all over the codebase and roots their causes in the leaf nodes of the program that are making use of `unsafe` blocks. Moving around the problem turns out to be a valuable way of addressing the problem, without ignoring any of the essential complexity or making anything fundamentally impossible.

> shortening the feedback loop drastically

More importantly, forcing you to catch them. Problems at runtime don't get caught until that condition actually occurs during an actual run of the software.

Not all forms of moving things around are equivalent. Some just keep the problem as diffuse as it was before, others restrict the problem to a smaller area.

Rust does not, by any means, solve this problem. What it does do is help you prove that the problem is contained to a smaller subset of your codebase that you can audit/reason about much more easily.

> Which is often the case with claims of solving long-lasting problems

This is what I really like about programming language theory: problems which are literally undecidable in existing languages, can be made almost trivial by designing a language with that problem in mind. The trick is to have the language keep track of the information needed for the solution, so we're not dealing with "black boxes".

A classic example is annotating every variable with a type, and propagating that information through compound expressions; lets us spot problems like trying to use a number as a string, without worrying about the halting problem (we do this by forbidding some technically-valid programs, like `1 + (true? 10 : "hello")`).

Of course, this leaves the much harder problems of (a) keeping the burden imposed by this extra information down to a reasonable level, and (b) trying to make the language compelling enough for people to actually use. Rust is one of the few examples of going all-in on a new ecosystem, and it actually being embraced. Often it's easier to "embed" a new language inside an existing language, to piggy-back on the existing tooling (e.g. the variety of auto-differentiating languages built in Python)

Yes, people often forget that TMs really are "pure runtime". Sure, the program itself is static, but nothing else is. There's no imposed semantics on anything beyond the pure mechanism of the interpreter itself. This lessens the 'problem' part of the Halting Problem drastically if you just impose some constraints on the programs, e.g. static types, maybe a borrow-checker, etc.

> I think the underlying assumption is more along the lines of: the easy cases are made easier, and the hard cases are made impossible — or rather the concern being that rust hasn’t made the problem easier, it just moved it around.

Except in this case the article linked demonstrates that they've made things dramatically easier overall. Data races went from being endemic in a C++ codebase to being in a few low-level and easy-to-locate areas.

The easy stuff stayed easy, the bulk of the rest was folded into the easy case, and even the hard cases appear to have been made easier to find, identify, and fix.

Once you pick away enough at the straws, the needle becomes a lost easier to find.

D takes a different approach. D has a type constructor `shared`. Only types marked as `shared` (or `immutable`) can be accessed by multiple threads. If you've got concurrency problems, the scope of the problems are immediately reduced to the subset of the code typed as `shared`.

In addition, `shared` types cannot be accessed directly. They can only be accessed via library routines, such as D's atomic library.

Having an `immutable` data type also helps, as immutable data can be accessed by multiple threads without need for synchronization.

It's not that different. Rust has Sync + Send + &refs and &mut refs.

Does this approach have to deal with potential recursive locking?

Not by itself. But D's prototype ownership-borrowing system can help with that.

Did D ever get around to specifying a formal memory model? Because when I tried to adopt D a couple years ago I felt there was a lot of ambiguity surrounding the semantics of `shared`, particularly when interfacing with C and C++ code.

I ended up just casting to and from `shared`, and that seemed to work, but it was pretty verbose for highly parallel code and I was never quite sure under what circumstances doing that might violate the compiler's invariants.

Also such casting appeared to eliminate most of the benefits, since what appeared to be local data might have been cast and had a pointer sent across to another thread elsewhere. In the end `shared`, @nogc antics (the language internals weren't fully annotated), closures requiring the GC (compare to C++), and the friction of interfacing with highly templated C++ code such as the Eigen library caused me to abandon the attempt. I sure learned a lot in the process though!

> memory model

It's very similar to C++'s memory model, but it isn't necessary to understand it to be effective with D.

> casting

Casting to/from shared is necessary, but is restricted to @system code.

That description seems very similar to how Rust's type system enforces thread safety. You have two kinds of references: shared and exclusive. The shared references will typically only provide immutable access, whereas if you have an exclusive reference, you can mutate it. Beyond this, there are some special types that can be mutated through shared references, e.g. the AtomicI32 type. Similarly, the Mutex type lets you lock it given only a shared reference and provides you with an exclusive reference when locked.

> D takes a different approach. D has a type constructor `shared`. Only types marked as `shared` (or `immutable`) can be accessed by multiple threads.

This is called Sync in rust. There is also a Send trait for types types which can be moved (but not necessarily shared), as there are data items which can not leave their origin thread e.g. thread locals or non-atomically refcounted instances.

It's interesting, I read this and had the opposite conclusion -- people are still writing racy code using Rust.

I mean, it's great that you can write threaded code with more confidence, but it seems the conclusion here is that Rust gives you more confidence but not absolute confidence and you still need the normal stable of runtime detection tooling. The second fix[1] in particular resembles the bad pattern of "sprinkle atomics it until it works" you see in non-Rust languages.

(If this comment raises your hackles, please read the "no true Scotsman[2]" fallacy before responding.)

[1]: https://hg.mozilla.org/integration/autoland/rev/044f8b594c71 [2]: https://en.wikipedia.org/wiki/No_true_Scotsman

> What issues we did find were mistakes in the implementations of low-level and explicitly unsafe multithreading abstractions

(emphasis mine)

The important thing is that the races only happened in unsafe { } blocks. It's well-established that these blocks should basically be treated like C++ in terms of scrutiny, but (I believe) you can still have roughly "absolute confidence" in non-unsafe code.

It's true and interesting that unsafe { } blocks deserve sanitization analysis like C++ code does - and maybe this hasn't been discussed enough - but I don't think it's fair to suggest that the enormous language benefits are outweighed by some false sense of confidence. The proportions in that idea just seem way out of whack.

AFAIU one can create races in safe code using atomics with insufficient ordering.

(That notwithstanding, I agree that Rust is a vast improvement over C/++.)

"races" is too general. Safe Rust cannot have data races. It can have race conditions.

I'm not convinced this is a meaningful distinction in this context. Assuming a developer consciously creates a race which turns out to be a bug, how does it matter whether it was a data race or some other kind of race condition?

Data races are undefined behavior, and race conditions are a logic error. Yes, both are bugs, and both are important to fix. But they have (at least to me) different severities.

(Worth noting that data races have a clear, unambiguous definition, and therefore are possible to eliminate through tooling. Race conditions rely on program intent or logic, and are therefore... not really.)

Also, I think it matters, generally, that we are truthful about what Rust prevents and does not prevent.

> Worth noting that data races have a clear, unambiguous definition, and therefore are possible to eliminate through tooling. Race conditions rely on program intent or logic, and are therefore... not really.

This is a great point. Usually people only talk about the two categories in terms of vague severity, but this is a hard distinction which has lots of implications for how each can be dealt with.

> Data races are undefined behavior, and race conditions are a logic error. Yes, both are bugs, and both are important to fix. But they have (at least to me) different severities.

Some data races really are benign while some other race conditions can lead to data corruption or security vulnerabilities. Why then should data races be strictly more severe than data races? What am I missing?

Any kind of race can lead to data corruption, or being very drastic, possible death if the outcome is related to any computer system with direct influence over human lives like factory automation systems.

This is why I think Rust discussions about this subject don't focus enough that the language only validates a very tiny subset of races.

In general data races can break program invariants, violate memory safety and do other undefined behavior.

In C++, data races can cause jumps to invalid addresses and the compiler will assume data races away and delete checks that are redundant in the single-thread case. The equivalent problem happens in most languages where data races are possible (two Python threads writing at once can put a value into an impossible/illegal state)

An ordinary race condition is just an observable effect of the nondeterministic progress of threads, so it's possible to have a truly benign race condition, so long as the correctness of the whole program doesn't depend on the outcome of the race. I/O causes nondeterminism anyway, so it's not really practical to avoid all ordinary races.

It's possible to have truly benign data races too, e.g. lazily computing a hash of immutable data.

One of the reasons that Rust's ownership model is necessary is most data races cannot otherwise be statically detected. The compiler will be unable to apply any of its theoretical UB optimizations. It will emit its native code and you get the defined behavior of the underlying hardware. There will be no undefined behavior. If that ever changes, the bigger news, I imagine, will be that we solved the halting problem.

"This context" is the original article, which is discussing data races specifically.

Actually "this context" was my previous reply. Much like I can nest a lexical scope inside a program and have it reach up and past the containing scope, I can create a new context in a discussion. Please at least try to understand my position before you reply.

And deadlocks

You can also create races in safe Rust by using the file system, no concurrency primitives necessary (just create a new file, close it, and re-open it assuming it's still there..).

What safe Rust doesn't have is _data races_.

The thing is, the linked data race occurred in safe code! Looking at https://bugzilla.mozilla.org/show_bug.cgi?id=1686158, the patch changes a non-atomic read through a &SwCompositeGraphNode and a subsequent non-atomic write through a &mut SwCompositeGraphNode (both safe code). They should not race, and the only reason they do is because there wasn't proper RwLock-style synchronization when handing out &SwCompositeGraphNode and &mut SwCompositeGraphNode. Also I think merely creating an unsynchronized &mut is undefined behavior in of itself (in addition to allowing UB in safe code). If so, even the fixed code is still technically wrong (just not miscompiled or misbehaving in practice).

Looking at https://searchfox.org/mozilla-central/rev/ee9dab6aa95f167a34..., SwCompositeGraphNodeRef has "safe" methods (line 286) with unsafe blocks, handing out raw &SwCompositeGraphNode and &mut SwCompositeGraphNode to the interior of an UnsafeCell, without locking or runtime checking. I think SwCompositeGraphNodeRef "caused" the UB because it was invoked by some other caller to create a &mut SwCompositeGraphNode without proper synchronization... And worse yet, SwCompositeGraphNodeRef has Deref and DerefMut implementations that implicitly create & and &mut, unsynchronized, at call sites that merely look like method calls...

My theory as to how this happened:

- C++ atomics are not mutable through a const *, so the authors used &mut to indicate mutability (even though Rust's &mut has stronger guarantees, according to the Stacked Borrows memory model, "all other pointers to this object are invalidated, except the pointers we reborrowed from"... though Stacked Borrows may change to make async fn and intrusive linked lists sound, I suspect it won't make this code sound).

- Maybe they thought "we want to prevent calling &mut SwCompositeGraphNode methods from a &SwCompositeGraphNodeRef". IDK. Semi-related: SwCompositeGraphNodeRef is a #[derive(Clone)] Arc<UnsafeCell<SwCompositeGraphNode>> but !Sync, so only threads that own a SwCompositeGraphNodeRef can create either &SwCompositeGraphNodeRef or &mut SwCompositeGraphNodeRef. And I don't see &SwCompositeGraphNodeRef floating around, so I don't see why access control would matter.

(EDIT: The commit introducing the unsound abstraction is https://hg.mozilla.org/integration/autoland/rev/8d47e408c578, filed under the bug at https://bugzilla.mozilla.org/show_bug.cgi?id=1670328.)

hopefully it will lead to a new borrow checker logic

Is that "sprinkle atomics" stuff in bindings/interfaces for non-Rust external code? I'm not particularly surprised that you would have to wrap atomics around externalities and not doing that would be a source of errors if the external code has race conditions.

It's not, I linked the patch. (I don't fully understand the patch so I'm happy to be corrected here, but it looks like it was an ordinary "use atomics incorrectly" bug.)

It's unsafe code (see the "unsafe" keyword?). So, Rust relies on the people who wrote it to do so correctly, and in this case they mistakenly assumed they could get away without one variable being atomic. The patch corrects that, making it atomic.

So Rust did exactly what we wanted here, the problem code was narrowed down to the code that was explicitly unsafe and needing caution. Humans remained fallible, and the ThreadSanitizer spotted their mistake.

Another fallacy to keep in mind is that two things not being perfect does not make them equal failures.

You are correct. The "no data races" slogan of Rust is accurate but you need to mention the way it needs to be understood. First, data races are only a subset of all race conditions. Second, like the other safety guarantees, the statement only applies to safe Rust. Once you use unsafe Rust or any non-Rust language (and both usually happens in any non trivial program), this guarantee stops being a guarantee. But it doesn't mean it vanishes into nothingness. It instead becomes a quantitative statement: it's much easier to avoid data races in Rust than in unsafe alternatives.

I think when you build Rust projects that require reliability, it's foolish to lean back and believe that marketing slogan and not look for concurrency bugs. However, once you found them, it's way easier to fix them, at least when your unsafety is well encapsulated like how it's recommended.

I think the big challenge for Rust in this context is to improve its tooling so that using it is comparable or even easier than the C++ tooling. This blog post is proof that such tools should be ran otherwise you are missing out on important bugs.

I think the mistaken idea that Rust allows "bug free" multithreaded programs will lead to a new generation of difficult to maintain multithreaded software... The lesson of multithreading should be that you need to use less of it and only when necessary, not that you should write more. It is similar to what we got when Java was the darling against C++: you now have lots of Java programs that don't have the same problems as the C++ version, but still are difficult to understand, leading to the well known FactoryFactory syndrome.

I honestly don’t see “less concurrency” as a viable route for modern software. We’re adding cores at a much faster rate than we’re improving clock speeds. Anything that makes concurrency easier is a win in my book.

There are several ways to increase concurrency: low level and high level. UNIX, for example, provided high level tools for concurrency (processes). You can also use message passing if you want. It is possible to write more concurrent software using safer primitives, instead of relying in low level techniques such as Rust/C++ multithreading.

I don't think it makes sense to call processes either higher or lower level than threads. Splitting your program into multiple processes often comes with many other drawbacks than spitting it into multiple threads. You can even still have data races if you use files or mmap to communicate between processes (you can also get something similar even with pipes/sockets for complex chunked data structures).

Even worse, there are very few, if any, tools that could help you analyze a multiprocess program for any kind of race conditions.

When processes crash or corrupt memory they don't bring all the application down with them like threads, hence why security focused architectures have moved back to processes after the thread adoption hype.

They literally found tons of new races conditions in c++ and 2 in the rust code with their tool. And you want to act as if that little factoid makes rust useless somehow. Anyone who thinks a language is flawless is not thinking correctly and that includes rust. However it does a much better job at this than C/C++ yet somehow you twisted that into declaring rust as useless.

I'm not sure how you got from "Rust gives you more confidence but not absolute confidence and you still need the normal stable of runtime detection tooling" to "useless".

It's probably because "Rust gives you absolute confidence" is a strawman, I don't think anyone was arguing that, and often people's first reaction to a strawman is to construct a contrary strawman.

> you still need the normal stable of runtime detection tooling

I hate that you're getting so many downvotes, and I wish people wouldn't do that...but I still want to disagree slightly :) I think it's interesting to note that the meaning of "you" changes between the two cases. If I write safe code on top of parking_lot, you could argue that I don't really need to do runtime race detection. Rather, someone in the ecosystem needs to do it, but once it's been done, the shared code is improved for everyone. If my code is all safe, I can benefit from these global improvements over time, and I can be confident that I haven't introduced any new races myself.

"No true Scotsman" is a fallacy because "true Scotsman" doesn't have a clear definition and can mean anything. But "unsafe" has a very precise meaning in Rust. Safe Rust didn't have any data races, which is exactly what we would expect.

Reply on news.ycombinator.com