[filwit]

So Rust has some cool safety features, especially for concurrent code. But, and perhaps I'm just uninformed, I never really understood the safety benefit of Rust's 'never nil' design. Nil is a useful modelling tool, even in Rust where it exists via Option<>/None, correct? Perhaps by forcing you to be extremely explicit (and enforcing `match` always handles all conditions) you gain some arguable safety, but at what cost? It's certainly not easier to use and reason about, IMO. And it seems just as likely you'll end up crashing your program due to a bounds-check error (which may happen more often since Rust encourages indexing over references due to this very design.. at least, so I've read).

It seems to me the design was chosen more as a way to ensure memory lifetime could be better predicted by the compiler rather than any strong argument for safety.. but then, I'm not well read on the subject, and It's very likely there's good safety arguments for it I'm not aware of.. either way, in my experience nil-deref errors are rarely a painful thing. They happen often, but are also fixed quickly.

[pcwalton]

> Nil is a useful modelling tool, even in Rust where it exists via Option<>/None, correct?

It's not that null is not useful. It's that most pointers can never be null, so nullability is the wrong default. And it is useful for the compiler to force you to handle the case in which pointers are null.

> Perhaps by forcing you to be extremely explicit (and enforcing `match` always handles all conditions) you gain some arguable safety, but at what cost?

There's basically no downside to having no null pointers. With constructs like Option::map the code is usually even less verbose than the equivalent code with null.

> It's certainly not easier to use and reason about, IMO.

You never have to worry about your program failing whenever you type "." or "∗". With null, the semantics of the language are that an exception can be thrown [1] whenever those constructs are invoked. That's pretty much objectively easier to reason about.

> It seems to me the design was chosen more as a way to ensure memory lifetime could be better predicted by the compiler rather than any strong argument for safety

Huh? Lifetimes are totally independent. We could have had null pointers with the lifetime system (and there were languages like Cyclone that had both). The system exists precisely because of safety.

We also get some really nice optimizations out of it that are impossible to get in C. All pointers in Rust are dereferenceable per the LLVM definition, which opens up some really neat optimizations like loop invariant code motion on loads.

> in my experience nil-deref errors are rarely a painful thing. They happen often, but are also fixed quickly.

Not in my experience. They show up in production all the time.

[1]: Or you could do what Nim does, and make dereferencing null undefined behavior instead of guaranteeing that an exception is thrown, but that strikes me as worse than what Java does.

[filwit]

> it is useful for the compiler to force you to handle the case in which pointers are null.

Well I agree that it's very useful (and we have that in Nim), but..

> With constructs like Option::map the code is usually even less verbose than the equivalent code with null.

I'm still not convinced of this part. It certainly hasn't been the case with the, admittedly small amount of, Rust code I've seen. However, I'll look for more comparisons in the future (or offer Nim comparisons to Rust snippets anyone posts). Point is, nil is still a useful and commonly used tool. So the argument for verbosity and conveniences is relevant, IMO.

> With null, the semantics of the language are that an exception can be thrown [1] whenever those constructs are invoked. That's pretty much objectively easier to reason about.

That completely depends on how often you want to use nil refs, and how easy they are to use. Like I said in another response, I agree Rust's design may be better for some domains, but I certainly wouldn't call it "objectively" easier to reason about in a general sense.

> Huh? Lifetimes are totally independent.

Well like my post implied, I was only guessing as to the design. And it's interesting to hear that it takes advantages of special compiler optimizations. That said, I still don't see how it's completely decoupled from the life-time system.. you're saying that if I have a Option<> reference to a mutable list in Rust, the compiler can determine weather or not the list is 'frozen' based on the runtime state of that reference?

> Or you could do what Nim does, and make dereferencing null undefined behavior.

I didn't think derefing nil was undefined behavior. I thought only dereferencing a pointer which points to once-valid-but-now-free memory was undefined behavior, and that situation is covered by GCed refs. Can you explain this a bit?

EDIT:

> Not in my experience. They show up in production all the time.

I did say 'rarely', and I drew a comparison to bounds-check crashes, which surely also show up in production.

[pcwalton]

> Point is, nil is still a useful and commonly used tool. So the argument for verbosity and conveniences is relevant, IMO.

The only advantage of having null references is that the pattern "if this reference is null, dereference it; otherwise throw an exception" is shorter. But the question is: how often do you want that pattern? In a robust program, the answer to that is "rarely".

Put another way, it would be trivial to add sugar for the ".unwrap()" pattern to Rust (perhaps with the "!" operator) if it were necessary, gaining back the only verbosity-related advantage of null pointers. But nobody in the Rust community is asking for it. That's because this pattern is rare. If it were a problem, someone would have at least submitted an RFC by now!

> I certainly wouldn't call it "objectively" easier to reason about in a general sense.

If you write down, formally, what the star or dot operators do, there are strictly more steps involved when you have null pointers. That's why a language without null is objectively easier to reason about.

> if I have a Option<> reference to a mutable list in Rust, the compiler can determine weather or not the list is 'frozen' based on the runtime state of that reference?

I don't know what this means. Lifetimes rule out dangling pointers. They don't have anything to do with nullability. The borrow checker only cares about the structure of your data enough to construct loan paths.

> Can you explain this a bit?

Dereference of null is undefined behavior in C, and Nim compiles to C code that blindly dereferences pointers without inserting null checks. So dereference of null is UB in Nim too. In an earlier comment I was able to construct a Nim program that exhibited very different behavior in debug and optimized builds, using nothing but GC'd pointers.

> I did say 'rarely', and I drew a comparison to bounds-check crashes, which surely also show up in production.

Actually, Rust does try to prevent indexing-related issues by preferring iterators to raw array indexing. But, in any case, the comparison isn't relevant for a couple of reasons. First of all, in a general sense if you have big problems A and B, the fact that you can't solve B isn't an excuse to not solve A. More specifically, though, the amount of type system machinery needed to fully eliminate bounds check failures is much higher than that needed to eliminate null pointer exceptions—you basically need dependent types, whereas to eliminate null pointers all you need are bog-standard algebraic data types, which have existed since the 70s.

[filwit]

> there are strictly more steps involved when you have null pointers.

Well yes, and both Nim and Rust have non-nil pointers.. I suppose I misread your original statement as "Rust is objectively better ..." when you actually just said non-nil vars are an objectively better design pattern in general. My mistake.

Our argument seems to stem around the two assertions (one from you, and one from me), those are: "nil vars are be rare (in optimally written code)", and "Rust's way of working with 'nil' vars is verbose". I suppose I'll concede that non-nil vars is a better default (though I will hold reservation until I see more real statistics, I don't find "no RFC yet!" as hugely convincing), but I also feel Rust could do a better job of giving access to "nilable" vars when they're needed.

> I don't know what this means. Lifetimes rule out dangling pointers...

I mean, Rust prevents you (via compile-time mechanisms) from mutating a variable while it's borrowed by another reference.. If that reference is Option<>, it's only known at runtime weather or not a reference has actually borrowed said varaible. Rust must either treat every Option<> reference as a potential 'loan path', which would significantly diminish their usefulness as a references, encouraging indexing for these scenarios, which leads to almost identical potential for out-of-bounds crashes... or it's relying on some kind of more complex mechanism (lifetime vars maybe?).. or additional runtime overhead.

I really don't know enough about Rust to know how far off-base that is. So any clarity is appreciated.

> In an earlier comment I was able to construct a Nim program that exhibited very different behavior in debug and optimized builds, using nothing but GC'd pointers.

I remember this comment, but I didn't remember it achieving UB in debug code.. I'll look through the history and take another look.

[phaylon]

> Rust must either treat every Option<> reference as a potential 'loan path', which would significantly diminish their usefulness as a references, encouraging indexing for these scenarios, which leads to almost identical potential for out-of-bounds crashes... or it's relying on some kind of more complex mechanism (lifetime vars maybe?).. or additional runtime overhead.

Can you give a concrete example of this? I'm a bit confused, but it might just be a terminology thing. In Rust, `Option<T>` does not imply a reference. If you have a `Option<i32>` there are no references involved. An `Option<T>` also owns the `T` if there is one. You can get a reference to it, but you have to check that it indeed holds a `T` (via `match` or `match` using functionality).

Note: I should clarify: What's confusing me is the indexing stuff. I'm not sure if this is referring to something about the `Option<T>` or something else.

[filwit]

> I should clarify: What's confusing me is the indexing stuff. I'm not sure if this is referring to something about the `Option<T>` or something else.

By indexing, I meant as an alternative to references.. For example, if you had a Sprite type which held a 'reference' to a Texture in your game's Texture list.. as soon as you allocate a Sprite it must borrow a reference to a Texture, preventing any future mutation of the Textures list for the lifetime of the Sprite, which obviously is too restricting for most games.. so the alternative is to have the Sprite simply hold an index to an item in the array instead, but this basically comes with the same pitfalls as nilable refs (ie, if you accidentally change it, your program can crash due to bounds-checking errors.. or end up with visual glitches.. not sure which would be more annoying).

The other alternative is to use an Option<&Texture> instead. However, I'm not familiar enough with Rust to know of the restrictions here, or even if that's possible (taking a look at the docs, it looks like it's possible, but life-time vars come into play, which could complicate things).

[haberman]

> I remember this comment, but I didn't remember it achieving UB in debug code.. I'll look through the history and take another look.

It's: https://news.ycombinator.com/item?id=9050999

[filwit]

So I remembered correctly, Nim does not reach UB in non-release code (or rather, code without --boundCheck:on), it throws an exception. I still think this is a reasonable solution. We catch these errors during development iteration or enable it for safety-critical portions of release code (or the entire project).. and we can opt-out of these checks if we need the performance and safety isn't as important (games, simulations, etc).

I remember Rust does not bounds-check it's iterators, so you don't need to really disable bounds-checks (indeed, you cannot) while Nim, currently, does this more niavely and looses some performance for it. That's a nice thing Rust does, but not something Nim can't eventually catch up too. See this comparison for futher reference: http://arthurtw.github.io/2015/01/12/quick-comparison-nim-vs...

[haberman]

> In an earlier comment I was able to construct a Nim program that exhibited very different behavior in debug and optimized builds, using nothing but GC'd pointers.

This intrigued me so I found the comment: https://news.ycombinator.com/item?id=9050999

[polyfractal]

It's not so much about "never nil", but rather "never accidentally null". Rust's compiler prevents you from moving data into a method which then nulls it out, leaving a dangling pointer in the calling code. At best this dangling pointer will look at garbage and cause a crash or undefined behavior. At worst, it will look at other, actively-used memory and cause a security vulnerability. Rust will just refuse to compile until this problem is fixed.

Static analysis of these lifetimes allow a whole class of errors to be avoided (dangling pointers, double-free, iterator invalidation, etc).

Rust's Options are handy for a lot of stuff (async APIs, concurrent code that may/may not succeed, error codes, etc). But they are just icing really, not really the main thrust of Rust's memory model.

> It's certainly not easier to use and reason about, IMO.

To offer a counter-viewpoint, I find that Option<> (and Result<>) are very easy to reason about. They tell you exactly what to expect from a function, and you don't have to guess if you need to catch exceptions or let them throw higher up. Everything is explicit, the only surprises are panics which are cataclysmic anyhow.

> And it seems just as likely you'll end up crashing your program due to a bounds-check error (which may happen more often since Rust encourages indexing over references due to this very design.. at least, so I've read).

If you use iterators in Rust, you never need to worry about out of bounds errors. In fact they skip range checking altogether, since you are guaranteed that the value you are iterating on won't change under your feet (no iterator invalidation, etc), which generates more efficient code[1]

If you use explicit indexing, then yes, you can have a runtime panic if you go OOB. But that's the nature of explicit indexing. It also has to include those safety checks, so will be slower code.

[1] https://doc.rust-lang.org/book/iterators.html

[filwit]

> Rust's compiler prevents you from moving data into a method which then nulls it out

Just for clarity, we have this in Nim too, eg:

  type
    Foo = ref object
    Bar = object
      val: Foo not nil
  let
    f = Bar() # Error, 'val' must be set
  
  proc foobar(f: Foo not nil): Foo not nil =
    return nil # Error, can't return nil
  
  foobar(nil) # Error, can't pass nil

> At best this dangling pointer will look at garbage and cause a crash or undefined behavior. At worst, it will look at other, actively-used memory and cause a security vulnerability.

In C/C++, yes, but this isn't so applicable to Nim where we have GCed references and 'not nil' constraints.

> If you use iterators in Rust, you never need to worry about out of bounds errors

Well I was not talking about iterating through a list, but rather maintaining arbitrary indexes to a mutable list. Eg, a Sprite which contains a index to a Texture array. In that scenario it's just as easy to miscalculate and crash your program via a bounds-checking error as it is to crash by nil-deref.

> To offer a counter-viewpoint, I find that Option<> (and Result<>) are very easy to reason about..

It's good that Rust works for you, truly. And like I said in another post, I agree Rust's design here may be better for some domains. However, Nim's design still feels more elegant and straight-forward to me. Luckily, we both get a powerful language that suits us, regardless of which one we prefer :)

[polyfractal]

Oh, sorry, I should have been clearer: I wasn't trying to disparage Nim at all (it's on my list of languages to play with). I was just clearing up some points about Rust :)

Edit:

> Well I was not talking about iterating through a list, but rather maintaining arbitrary indexes to a mutable list. Eg, a Sprite which contains a index to a Texture array. In that scenario it's just as easy to miscalculate and crash your program via a bounds-checking error as it is to crash by nil-deref.

The solution here is to just use a reference instead of an arbitrary index. If you hand out references you can lean on the compiler to enforce memory safety -- the compiler won't let you access data that is no longer alive, won't let you accidentally share across thread boundaries if you don't explicitly want that, etc. And if that was a shared mutable list, it's doubly important to let the compiler help you reason about it, since shared, mutable state is the main source of data races.

This is one of those cases where leveraging the compiler allows you to write better, safer code.

[filwit]

No worries. I also wasn't implying you where trying to discourage Nim, and I hope my post didn't come off as accusatory. Cheers!

EDIT:

> The solution here is to just use a reference instead of an arbitrary index.

Ah, sorry I should have said "mutable Texture array". In that case Rust's borrow-checking will 'freeze' the array, preventing Textures from ever being changed during the lifetime of your Sprites. So you're left with either Option<> or indexing as a solution, each with it's own merits, but neither as.. practical as nilable GCed references, IMO (again, just my opinion.. others seem to find it easy enough).

[phaylon]

Out of curiosity: How do you convert from a nilable `Foo` to a `Foo not nil`? As in, what do you do if you have a function maybe returning a `Foo` and want to pass its value to a function taking `Foo not nil`?

[filwit]

You prove to the compiler that the nilable var is not-nil via if statement. Eg:

  proc foobar(f:Foo not nil) =
    discard
  
  let f = Foo() # nilable ref
  let b: Foo not nil = f # Error, can't prove 'f' is not nil

  if f != nil:
    let b: Foo not nil = f # can assign non-nil vars to 'f'
    foobar(f) # or can pass 'f' directly

[phaylon]

Ah, good to know. So Nim does static analysis on conditionals where Rust would use an explicit pattern match to get at the contained value.

Might be a good example for http://nim-lang.org/manual.html#not-nil-annotation

[Dewie3]

> Nil is a useful modelling tool, even in Rust where it exists via Option<>/None, correct?

Yes. So? A lot of things are useful modelling tools, but that doesn't mean you necessarily want to include them in every reference-like type. A tuple of two things is useful modelling tool; should that then be infused into every type? A "either value or error" is a useful type; should that be infused into every type?

No? Then what makes "Either something or nothing" so special?

[steveklabnik]

As for null in general, you can hear it from the horse's mouth here: http://www.infoq.com/presentations/Null-References-The-Billi...

There are a number of ways to approach this topic, so I'll just give you one: in languages with more advanced static type systems, you try to encode as much semantic information as possible in the type. As you've said, the idea of null can be useful, so it deserves a place in the type system. You want to separate things that may be null from things that should never be null. This is because not-null is by far the common case. Allowing everything to be nullable by default optimizes for the lesser-used semantic, which is where errors with null come in: you assume that something isn't null, when it actually is.

[filwit]

I agree the concept of 'non-nil' vars is very useful (and we have that in Nim), but I'm not entirely convinced by the rest of that argument. Namely, I don't agree that nil is rare enough to justify the verbosity Rust uses for it. Non-nil vars may be seen more often, but that doesn't mean nil vars aren't also often used, either. In Nim, both nil and non-nil vars are at roughly the same reach.. while in Rust non-nil vars are significantly easier work with. You may see that as a positive argument for Rust's safety (and you may be right for some domains), but I see it as more a negative argument for Rust's practicality.

[pcwalton]

> Namely, I don't agree that nil is rare enough to justify the verbosity Rust uses for it. Non-nil vars may be seen more often, but that doesn't mean nil vars aren't also often used, either.

It's not verbose. "Option" is 6 characters. ".map" is 4.

> In Nim, both nil and non-nil vars are at roughly the same reach.. while in Rust non-nil vars are significantly easier work with.

Option values are really easy to work with. Just use map or unwrap if you don't care about handling the null case. If you do care about handling it (which you should, after all!) the code using "if let" is the same as the equivalent "if foo == null".

[filwit]

> It's not verbose. "Option" is 6 characters. ".map" is 4.

I just want to note that verbosity isn't just about symbol length, but also about operator noise and the number of available or required commands used to achieve a goal. Just counting these characters isn't very relevant, and isn't even the best Rust can do (as someone pointed out you can use 'Some()' to get an Option var, which is only 4 chars).

That said, I agree this is rather subjective, and can't be well compared outside the context of the rest of the language.

[dragonwriter]

> (as someone pointed out you can use 'Some()' to get an Option var, which is only 4 chars)

Some() is 6 chars.

[filwit]

I was measuring with pcwalton's ruler.

[steveklabnik]

Fair enough. My argument here is more general than Rust itself, it's relevant to all languages with an Option type and no null. The verbosity can of course vary by language.

[beagle3]

What's the rust name/syntax for non-nil vars?

[steveklabnik]

    let foo = 5; // cannot be null
    let bar = Some(5); // technically also can't be null, but could be None
                       // instead of Some(val)

`foo` has the type `i32` here, and `bar` has the type `Option<i32>`.

[Dewie3]

> I agree the concept of 'non-nil' vars is very useful (

non-zebra numbers[1] are also very useful. But why have a "non-zebra number" when I can just have plain numbers?

[1] A "number" which is either a number, or a zebra

[filwit]

...because zebra's are not a universally useful modelling tool to programmers like references are. Thus, the absence of a reference, ie nil, also becomes a useful, commonly used modeling tool. If we all wrote software using african wildlife metaphor, 'non-zebra' might then be just as useful.

[Dewie3]

> ...because zebra's are not a universally useful modelling tool to programmers like references are.

References are not the zebras. Nil-references are.

We might as well say that the underlying number that the reference are represented by are useful modelling tools, in some circumstances. That doesn't mean that you want pointer arithmetic for references all the time.