Hacker News Reader

Comment by erk__ | Hacker News Reader

The way Rust dels with this issue is by saying that the Key must implement both `Hash` and `Eq` where Eq is a strict equality compared to `PartialEq` which allows the equality relation to be incomplete. While integers implement both PartialEq and Eq, floating point values only implement PartialEq.

https://doc.rust-lang.org/std/collections/struct.HashMap.htm...

https://doc.rust-lang.org/std/cmp/trait.Eq.html

https://doc.rust-lang.org/std/cmp/trait.PartialEq.html

There is also a partial ordering relation `PartialOrd` (and `Ord` for complete relations) Floating point values only implement the partial relation as well because of NaN's. This means that it is a bit harder to sort them, though the floating point standard also has a seperate complete comparison relation with some extra rules which you can use instead.

https://doc.rust-lang.org/std/primitive.f64.html#method.tota...

View on HN

Rust's HashMap and similar containers have clear() and many of them have drain()

Drain is interesting, because it resolves the other problem if things are in the collection which can never be retrieved from it. Drain gives you each of the things in the collection, one at a time, to do with as you wish, the collection is now empty. So this means even if the collection has a dozen SillyNonsenses in it, all of which insist they're not the one you were looking for whenever you go looking for a SillyNonsense in the collection, you get all twelve of them out in Drain, and can examine them as you wish.

Of course if you just wanted to fish one thing out and then throw the rest away, you can do that, once you drop the result of Drain the rest of the things being drained are dropped.

I thought HashMap behaved in an undefined-but-safe manner if an implementation of Hash or Eq is incorrect. Are you saying there's a reasonable way to put misbehaving types in a HashMap?

(I'm pretty sure I recognize your username as someone with way more rust experience than me. For the crowd an implementation of Eq for floats that used the standard comparison operators would be "incorrect", which is why the stdlib doesn't provide one)

If your key type K doesn't obey the rules explained in HashMap ("if two keys are equal, their hashes must be equal") then Rust doesn't promise HashMap<K,V>'s functions will do something useful (although it will still be safe if your type is safe). So no this isn't a license to put silly things in hash maps. Non-useful but safe consequences might include spinning indefinitely.

But if you have put something daft in a hash map, drain() should definitely get it back out, so there's that.

This is one of those rough edges where Rust is more annoying than helpful imo. It's also inconsistent.

I'd rather panic on NaN comparisons than complicate the fundamental comparison traits. Treat it like integer overflow or array bounds access. We don't have different traits for indexing that might fail or overflow that might happen. There are methods on the trait to handle those cases when you care about them.

But that said, this feels like an XY issue. If you're comparing hashes of floats as keys somewhere you don't want IEEE 754 defined equivalence. You almost certainly want bit equivalence, which means the keys should be type punned to u32/u64. Classic example is memoizing function calls with floats. You don't want to compare the semantic values of the arguments, but the literal bits. For floats that's not the same.

> I'd rather panic on NaN comparisons than complicate the fundamental comparison traits.

The fundamental comparisons are complicated period. Pretending they’re not and hoping you don’t hit the branches that crash is a terrible idea for writing reliable software.

For everything where the comparisons aren’t complicated (like integers), Rust is easy. For things where the comparisons are complicated, Rust makes you acknowledge this reality and deal with it. This is correct behavior.

No you don’t, you want to use the hardware defined equality function. And that one is IEEE 754 compliant.

In a hash table, I want my equality function to be based on the actual value, not the super special comparison for a specific type.

If I write a UniqueNumber that just returns false on the == operator, I'd be stupid but the hash map should still work. Java has a separate getHashCode() and equals() method for a reason.

Either provide a hash key override or use the raw bits for deep compares. Custom equality algorithms usually don't make sense for arbitrary key-value stores.

Because of the Pigeonhole Principle what you're describing is nonsense.

In Rust you can make UniqueNumber, indeed misfortunate::OnewayGreater is such a type. misfortunate::Maxwell even more so.

However of course the HashMap can't meaningfully "work" for this type. Rust promises your program doesn't have Undefined Behaviour despite such a prank, but it might spin forever when you try to insert this type into a map for example, a well defined but undesirable behaviour brought on by your poor choices.

You seem to imagine that Java's hash map doesn't need to compare types, that it can just use getHashCode() -- however because of the Pigeonhole Principle that can't actually work.

And NaN != NaN isn't a "custom equality algorithm" it's literally how the floating point numbers are defined in your CPU.

Floating point numbers are nothing more than bits with context. They're not some kind of super special construct that only dedicated hardware can operate on. They have their own modification instructions so they incorporate the float standard, but that standard is only relevant when you're operating on numbers.

In my opinion, language native hash maps should operate on memory, not on types and their weird implementations. Negative zero and positive zero are defined as different values but are mathematically equal; however, math functions may (and in the case of C, do) behave differently depending on which one you use. Even in higher level languages such as Java sorting gets affected by the presence of negative and positive zero using min/max operators.

If two values claim equality but do effectively alter program behaviour, I consider their equality operation in the context of memory operations such as hash maps to be buggy. There are good reasons to rely on the equality operators, but I do not think such behaviour should be the default.

The Java (openjdk) hashmap putValue doesn't seem to compare types, it just uses .equals()

  if (p.hash == hash &&
                  ((k = p.key) == key || (key != null && key.equals(k))))

You can't really rely on checking types in a Hashmap anyways because the type you're hashing can have more possible values than the 32bit hash. For example you can have 2 strings that have a hash collision.

NaN's work in a Java hashmap because Float.hashCode returns floatToIntBits(float) which normalizes all NaN's to a canonical value.

Sorry, I should have said compare values of-this-type. I was focused on the fact that type does matter here, which as you show the equals method is called on key, not some arbitrary system-wide equals method but a method defined on key's type.

Rust has a magic unstable trait StructuralEq which means not only are we promised that values of this type can be compared for equality (that's what Eq does) but that comparison will just be a bitwise memory comparison. Lots of useful things are not StructuralEq even if they are Eq

Apparently Java's Floats also do the same canonicalisation for equals(). So in effect in Java although NaN != NaN, once you box it that ceases to be true. This seems like a spectacularly bad idea to me, but presumably it made somebody's awful hack work at one point.

> This seems like a spectacularly bad idea to me, but presumably it made somebody's awful hack work at one point.

but elsewhere you say, "You are welcome to build a type which has this property and declares that it is Eq". This is exactly the difference between double and Double in Java. double has the `==` operator that isn't reflexive and works as required by IEEE-754 and Double has Double.equals, which is documented[1] to be a reflexive, transitive, symmetric, consistent, and substitutable relation.

[1]: https://docs.oracle.com/en/java/javase/17/docs/api/java.base...

Is it clearer if I explain that you are welcome to do things which are a spectacularly bad idea ?

How are you so sure what other people want? Using the hardware defined equality function for integers is reasonable a lot of the time.

My point is sometimes I want NaN to compare with NaN as true when I'm using a float as a key in a table.

You are welcome to build a type which has this property and declares that it is Eq.

But Rust's built-in floating point types f32 and 64 do not have that property.

Reply on news.ycombinator.com