[zvrba]

> Ok, so here is the article Klabnik cites, comparing various Rust HTTP clients:

Interesting. I'm being led to believe that the actual problem lies in HTTP itself: it is too convoluted to be safely implementable and should be replaced with something else for security-critical applications. (Or in general.)

[jude-]

I don't think this is true. In fact, today, I privately forked one of the aforementioned libraries (httparse) and removed all of the unsafe parts in under an hour (without introducing any new memory allocations, I might add). All the tests pass.

It's not insurmountable. It just takes a little bit of elbow grease and self-discipline ;)

[kragen]

While this may or may not be true, it's irrelevant to the use of “unsafe” code in Rust. Rust doesn't attempt a compile-time guarantee that your code is simple or bug-free. It attempts to guarantee that mutable aliasing, use-after-free, double free, and resource leaks don't happen. No wire protocol requires such things or is even relevant to them.

[thu2111]

But this is a problem Rust creates for itself. HTTP is safely implementable in most modern languages without any unsafe code because they're all garbage collected. For instance, Java has a mechanism also called Unsafe that lets you bypass the language's safety features, but I've never encountered an HTTP client that uses it because performance is fine without that. And the JVM developers have been working for years on safe but performant constructs that can replace the uses of Unsafe that do exist in the community, mostly, cases where very precise control over memory and concurrency constructs are needed in high performance servers.