Hacker News Reader

Comment by lsy | Hacker News Reader

What's the advantage of specifying it as random over specifying it as sequential in some way? Aren't both just specifying a behavior, where one behavior is potentially more useful than the other? I guess I understand the principled point that a set of hash keys is not an array. But it seems like more complexity than is necessary, and even possible to fall victim to Hyrum's law besides… you can imagine someone using random hash ordering to implicitly randomize something without needing to call an RNG explicitly.

It seems like the most principled approach might be an re-specification of the data structure: why is "iteration" even possible over a hash table? Shouldn't the keys be specified as a "set" on which only set-appropriate operations like map can be performed?

View on HN

One advantage to randomization, and why various languages did it in the first place, is that it prevents miscreants from DoSing your service if they know its implementation.

Say you're operating a service and you share its source on GitHub such that anyone can see it. Your language doesn't randomize hash values. Hashes are O(1), right? Well, not if a clever attacker can send you a set of values where they know each one will hash into the same bucket. The you end up with like 9 empty buckets and 1 with 100,000 items in it. Oops!

Old Python pre-randomization had a dict.items() method that would yield all the keys in order, conceptually kind of like:

  for bucket in self._buckets:
      for item in bucket:
          yield item

The order of those buckets would be repeatable between runs, so bucket foo would always come first, then bucket bar. Then Python added hash randomization so that the resulting hash key was something like hash(salt+key) instead of just hash(key). Now there's no way to tell in advance which bucket an item would get filed into, and the buckets would end up more or less balanced in size.

Newer Pythons (since 3.6? 3.7?) do something altogether different, and I can't explain exactly how their ordered dicts work, except to say I sat through a presentation on them and thought it was freaking genius even if I could re-implement it myself without sitting down with their docs.

> and I can't explain exactly how their ordered dicts work

Traditionally you simply use a doubly linked list approach on the entries (each entry maintains two additional references to the previous and next entry) for that like LinkedHashMap: https://docs.oracle.com/javase//8/docs/api/java/util/LinkedH...

https://github.com/openjdk-mirror/jdk7u-jdk/blob/master/src/...

Which is also what Python seems to be doing: https://stackoverflow.com/a/34496644

It's fairly intuitive.

Do their new default (now also ordered?) dics do this differently?

Note that OrderedDict is an implementation in Python. CPython's dict has a different implementation. There's more about it at https://docs.python.org/3.6/whatsnew/3.6.html#new-dict-imple... and https://mail.python.org/pipermail/python-dev/2012-December/1... .

This implementation was used from 3.6, right?

It's interesting that the idea mail mentions that nothing changes about the implementation (including order) but the memory layout. Which would imply insertion order was already preserved in older versions (not the case IIRC) or the idea underwent a few more changes that did in fact impact order.

EDIT: I couldn't quite find an answer but https://softwaremaniacs.org/blog/2020/02/05/dicts-ordered/ mentions the behavior happens since then because the implementation only tracks indices in the hash table itself and relies on maintaining entries separately in a second array that gets expanded in insertion order.

This would also seem straightforward but it raises a few questions such as how deletion is implemented (efficiently).

EDIT2: Okay, the talk (https://youtu.be/p33CVV29OG8) mentions they just leave holes until the next resize (at around 42:00).

Raymond also mentions there that his original idea didn't preserve ordering but happened due to an additional compacting optimization? Should probably watch the whole thing some time to get the history. Sounds like a fun talk.

Oh! Yeah, that was the talk, but repeated at PyCon. It’s a very clever design that wasn’t at all obvious to me.

Note that this relies on all the object classes having correct hash implementation (which is actually quite difficult); it's not really a property of the container.

That said, the attack is almost always on builtin types, so ...

In addition to the DoS aspect mentioned in a sibling comment, the primary reason you would do this is to avoid constraining the implementation. If you want to change the design to improve performance, for example, being forced to match the implicit ordering of the old implementation may be very difficult.

It certainly may be useful to define an specific ordering. Maps ordered by insertion time and maps ordered by key order are fairly common. But maps with these constraints probably won't be as fast as those with arbitrary ordering, so there is a trade-off decision to be made. A map constrained to the arbitrary ordering of the initial implementation is the worst of both worlds: difficult to make faster despite not having a particularly useful ordering.

As a concrete example, I am currently in the middle of rewriting Go's map implementation to use a more efficient design (https://go.dev/issue/54766). If Go hadn't randomized map iteration order this work would likely be impossible. It is unlikely we could completely change the design while keeping iteration order identical to the old implementation (assuming we want the new version to be faster at least).

Cool! Yes, that’s been our experience in Java as well. The randomized order of unmodifiable Set and Map (from `Set.of` and `Map.of`) have enabled us to make optimizations to the internals mostly with impunity. We’ve done so twice since their introduction and we might do so again.

> why is "iteration" even possible over a hash table?

Going through all items in a container seems like a logical thing to want to do.

> Shouldn't the keys be specified as a "set" on which only set-appropriate operations like map can be performed?

Efficient implementation (in space and time) of these set operations would limit the possible efficient implementations of a hash table (i.e. to using the hash value as a key in an ordered map).

Going one step further, this guarantee would freeze behaviour of the hash function (and how they are combined) for all time; optimisations for different architectures would not be possible, nor would avoiding hash collision attacks by changing seeds.

Sometimes hash value stability is what you want (e.g. when transmitting data in space and time), so "frozen" hash functions give these guarantees e.g. https://github.com/google/highwayhash#versioning-and-stabili... .

Yes it can be annoying when tests are flaky, but the unit testing/matcher library should have some reasonably efficient way of matching unordered containers.

I think removing methods from the Hashmap was not in scope for their work - they were working on the JDK, and presumably didn't have the bandwidth to patch every single Java library used at Google to use set-based methods instead of iteration-based.

Also, I think in practice you'll find you want to extract items from a set in some order, so you need some kind of transformation from set->list. e.g: you want to print them out.

Edit: Forgot to address your main question. If the specification requires a specific ordering, the implementation is forever bound to do that, even if other implementations would be more efficient/secure/other-desirable-properties. By introducing randomness, you reduce the risk of people accidentally relying on unspecified behavior, and are more able to change your implementation later.

One solution could be adding a `sort` argument (which takes a function that compares two `<T>` items and returns `true` or `false` depending on their order) to all functions on unordered collections of `<T>` items in which an order must be chosen, or requiring that the items in such collections implement a `TotalOrder` interface or something similar. This isn't very ergonomic in languages that don't have an equivalent of Traits or typeclasses though. In languages which permit side effects, this would include any functions that iterate over the items in an unordered collection.

Unfortunately, map can (and does) have external effects in most languages, so your code could still be effected by the external order.

I’ve actually worked on some maths code where we wanted to prove algorithms worked correctly in exactly this type of situation, so we wanted to prove our code iterating over a hashed container was order-invariant. You can do it, but involves pulling in theorem provers, and is beyond what most people would probably tolerate.

Reply on news.ycombinator.com