Hacker News Reader

Comment by ajross | Hacker News Reader

That's arguably true, but only insofar as it makes sense to talk about a mutex "owning" "data". There are many kinds of data that can't be owned by the language runtime like this (think about async/shared/DMA buffers, register blocks on foreign hardware, memory-mapped database files), but that still clearly need synchronization. Even things like coarse-grained locks taken at the subsystem level (e.g. to avoid having a separate atomic in every little object) are a bad fit for this kind of API.

There's value to being opinionated about API design, but there's also cost. And Rust is supposed to be playing in the same sandbox as the lower level tools.

View on HN

Here’s what I understand from your comment, correct me if I’m wrong. You’re saying there is a runtime cost associated with a mutex/rwlock owning the data. And secondly, you think it isn’t possible to implement this pattern in other areas that need locking.

I don’t think either of those is true. Encoding the ownership in the type system makes things clearer, imposes a compile-time cost but not run-time cost. Also, there isn’t any magic in the stdlib implementation of Mutex and Rwlock other than implementing it natively for every OS. This means that it is possible to implement the pattern for the examples you gave.

That is true, but I don't believe Rust imposes a cost here. The worst case degrades to a `Mutex<()>` and then whatever's being guarded being managed separately.

From an API design point of view, I would always encapsulate the logically mutable state in some type `Foo`, even if it doesn't literally own the memory, and then expose a safe API using `Mutex<Foo>`.

Idiomatically, you would then make `Foo::new_unchecked()` unsafe, with the precondition that nobody else is accessing the same external resource, and that `Foo::new_unchecked()` is only ever called once.

Right. I advise people who don't think that anything should be owned by the lock make a marker type - call it for example struct MacGuffin - with no data in it, just to mark that we took the lock. Firstly they can now write APIs which express ideas like "You can't call this unless you have the lock" (that's a MacGuffin parameter) but then also very often they discover actually there is mutable data that they only want when they've taken the lock, and hence have MacGuffin, and hey, we can just put it in the MacGuffin type.

So long as MacGuffin doesn't have any data in it, it essentially evaporates at runtime. This is a zero size type (ZST) so we can store that in no bytes of RAM, we can use no registers at all to pass it into a function, and so on. If you're a C programmer this doesn't make sense because you don't have ZSTs, all your types take up at least one byte, but Rust isn't like that.

Yep, it's nice to use marker types for this kind of thing.

I'm not convinced, because you're conflating too many things here. Ownership and borrow checking are always compile time concepts. There is no language runtime involved. If you think of C as a magic language then start today and get that idea out of your head.

If locking and unlocking semantics are a good fit for whatever you are doing, then you can just do a lock implementation setting whatever memory mapped registers you need. There is no constraint here whatsoever.

Nobody is advocating for separate atomics in every little object in Rust. That's a Java idea that failed miserably long before Rust was even around.

You can also use a mutex that doesn't own data in Rust if you need to.

Reply on news.ycombinator.com