News for hackers. Stuff that matters.
Re:Why I'm Choosing C++ by wuch on Monday February 01, 2016 @06:02AM (#11011040)
I have been playing with Rust for a few weeks now, so I can give a few reasons not to use it. What I will write may sound overly negative, but keep in mind that those are only reasons not to use it, in general I found a lot to like in Rust.

Unsurprisingly most of issues described below, boil down to Rust not being mature yet, and are perfectly fixable. However, fixing some of those will break backward compatibility one way or other and require effort that would be unnecessary otherwise.

Language is underspecified. In what order are subexpression evaluated in Rust? Unspecified yet, in fact it is even worse than that, it differs between borrow checkers (which I think uses the most useful one given Rust syntax: right to left behaviour), while code generation assumes left to right order. https://github.com/rust-lang/rust/issues/28160

What can you do with pointers in unsafe parts of code so that everything is still well-defined. There does not seem to be real answer to this question yet: https://github.com/rust-lang/rfcs/issues/1447

What does Rust guarantee with respect to lifetimes? In fact it guarantees much less than what would you expect. Take for example a drain method of a Vec. Drain gives you an iterator that moves elements out of a Vec. Naive implementation would leave Vec in a inconsistent state while drain iterator is still alive (because some of elements are already moved from), but restore Vec to a valid state in destructor of drain iterator. It appears to work because no one can touch Vec while drain iterators still holds mutable reference to it. Except it wouldn't work, because Rust does not guarantee that drain destructor will be executed, in fact in perfectly safe code it may not be executed at all. Surprisingly a borrow of a mutable reference may outlive the lifetime of a referent. At least it seems that Rust guarantees that those references would be inaccessible. https://internals.rust-lang.org/t/are-memory-leaks-considere...

There are quite a few language issues like this, and in general I found it very hard to find what exactly Rust semantics is.

Rust sends mixed signals about panic. On one hand it try to tells you that code that panics is probably badly broken, and may have left data structures in an inconsistent state, so you should not even try to access them. And Rust tries to ensure this to some degree by putting roadblocks on your way if you think otherwise (see Mutex poisoning, and strange things like AssertRecoverSafe you have to use when recovering from panics). Except in reality, all your desctructors already do access all those data structures. It seems that all those roadblocks are quite pointless.

Running out of memory aborts program. This is I think quite a bad decision (it seems that this may be a temporary one) for a language that have exceptions in form of panic. Not many programs need to handle out of memory situations, but handling those in programming language with exceptions and RAII is quite easy, and mostly boils down to: don't perform dynamic memory allocation in destructors. This is a step backwards from C++.

Bugs, bugs and more bugs. Rust not calling your desctructor for variables in the same scope after panic, still there after more than a year: https://github.com/rust-lang/rust/issues/14875

Borrow checker is interesting, certainly prevents you from writing buggy code, but also prevents you from writing code which is perfectly fine. In certainly does not play nice with ideas which are very common in C++ world, any STL algorithm + container would be a good example. I am still not sure if it is worth the effort, it seems to prevent problems that I don't commit in C++ anyway.

Immature state of ecosystem, libraries, documentation, ... .

In general you probably will not be as productive in Rust now as you would be in other languages. If you are interested in developing those libraries that are currently lacking, having some impact in how Rust language may look in a future, then Rust community is certainly great. But a lot of people have different priorities.

Re:Why I'm Choosing C++ by pcwalton on Monday February 01, 2016 @12:57PM (#11013625)
> However, fixing some of those will break backward compatibility one way or other and require effort that would be unnecessary otherwise.

We aren't changing anything you're talking about, except for bugs.

> In what order are subexpression evaluated in Rust? Unspecified yet, in fact it is even worse than that, it differs between borrow checkers (which I think uses the most useful one given Rust syntax: right to left behaviour), while code generation assumes left to right order. https://github.com/rust-lang/rust/issues/28160

That is not the standard question of right-to-left vs. left-to-right argument order. That is perfectly well defined: left to right. Rather, it's a very subtle question involving code nobody would write in practice.

> Except it wouldn't work, because Rust does not guarantee that drain destructor will be executed, in fact in perfectly safe code it may not be executed at all.

This is never a problem in real world code. The leakpocalypse affects nothing in the real world and was a lot of debate over what amounted to a very minor hole.

C++ has no guarantees about destructors being executed at all, and frequently due to memory management errors they aren't. Rust guarantees that they'll be executed unless you're leaking.

> Bugs, bugs and more bugs. Rust not calling your desctructor for variables in the same scope after panic, still there after more than a year: https://github.com/rust-lang/rust/issues/14875

C compilers have miscompilations too, with the same frequency. I never hit these in practice.

> I am still not sure if it is worth the effort, it seems to prevent problems that I don't commit in C++ anyway.

Yes, you do make those errors. Empirically. All large C++ projects I have ever seen have memory management errors, unless they're doing hugely expensive verification. I have not yet seen an exception.

> Running out of memory aborts program. This is I think quite a bad decision (it seems that this may be a temporary one) for a language that have exceptions in form of panic.

It's not a temporary one. Most programs don't need to handle out-of-memory conditions, and having the language do this is not a step back from C++. For programs that do, you can use libcore (recently stabilized!) and handle this yourself.

You're looking through the issue tracker, pulling out random bugs, and overstating their impact. Someone could do the same to any C++ compiler; LLVM has horrible-looking miscompilations all the time, and it's still production ready. Of the reasons not to use Rust, miscompilations (which is the vast majority of what you're talking about) aren't anywhere near the top of the list; they occur more or less with the same frequency they occur in C++. You can't complain that the borrow checker doesn't solve memory management issues you hit in C++ and then simultaneously criticize Rust by citing a whole bunch of issues that occur with much less frequency than those memory management issues.