Hacker News Reader

Comment by hinkley | Hacker News Reader

I wonder if I still have the link.

One of the papers I had bookmarked when toying with my own language design was someone that had worked out how to make interfaces as fast or faster than vtables by using perfect hashing and using the vtable as a hash table instead of a list.

You can also, when inlining a polymorphic call, put a conditional block in that bounces back to full dispatch if the call occasionally doesn’t match the common case. The problem with polymorphic inlining though is that it quickly resembles the exact sort of code we delete and replace with polymorphic dispatch:

    if (typeof arg1 == “string”) {
    } else if typeof arg1 === …) {
    } else if {
    } else if {
    } else {
    }

View on HN

> using the vtable as a hash table instead of a list.

Could you explain this a bit more? The word "list" makes me think you might be thinking that virtual method lookup iterates over each element of the vtable, doing comparisons until it finds a match -- but I'm certain that this is not how virtual method invocation works in C++. The vtable is constructed at compile time and is already the simplest possible "perfect hashtable": a short, dense array with each virtual method mapping to a function pointer at a statically known index.

The problem they were trying to solve was multiple inheritance, and by nominal type not by code reuse. So interfaces, basically.

So these guys essentially assigned a hashcode to every function of every interface and then you would do dispatch instead of obj.vtable[12] you would do modular math x = singature.hash % len(obj.vtable) and call that.

I believe this was sometime around 2005-2008 and they found that it was fast enough on hardware of that era to be usable.

Thanks, I think I get it now. The hash value would be a pure function of the method's signature (argument types and return type) and its name, so that two interfaces with a same-name, same-signature method would hash to the same value and thus invoke the same underlying method; the constraints would be that, after modulo, different methods must map to different indices; and the objective function to minimise would be the vtable size (which I think would be common across all classes).

But maybe I don't get it, since this would require knowledge of all interfaces, and as soon as you require that, it's straightforward to build a minimal-size mapping from method name+signature to integer index: e.g., just form the union of all method declarations appearing in any interface, sort them lexicographically, and use a method's position in this sorted list as its index. Lookups in this map are only ever done at compile time so there's no runtime inefficiency to worry about.

At the time that article was published in SIGPLAN, the dominant language was Java, which is officially statically, strongly typed.

Although really it isn’t because the JVM is strongly typed but evaluates some things at load or first invocation so it allows some languages that run in the JVM to be a bit tricky. They first generics implementation on the JVM, called Pizza, leveraged this load time concretization to do its thing.

But if you have a language that can resolve the type system at link time then you can do this trick. Alternatively you could switch to cuckoo hashing and if you next module load starts causing collisions, then so be it.

The problem is that C++ stores the vtable inside the object, and the objects over which you're iterating often weren't allocated contiguously. Even when they are, if each object contains lots of other data, the vtables won't necessarily be close to each other. That means that invoking virtual functions inside a loop means a lot of cache misses, and since the data you're fetching will be a branch target, it's often hard to find other useful work to accomplish during the memory delay cycles. However, in a language where you can store a relatively tight array of object IDs (or even use tag bits in the this pointer), now you have a much higher cache hit rate on the indexes to your equally tight dispatch table, which will also have a high hit rate.

It's a fair amount of extra work, but in a hot loop it's sometimes worth it. "You can often solve correctness problems (tricky corner cases) by adding an extra layer of indirection. You can solve any performance problem by removing a layer of indirection."

"list" here does not refer to a "linked list". In more academic circles, a "list" referes to any linear container. Such as a Python List. In practice, C++ vtables are effectively structs containing function pointers.

Nice one, TIL

One caveat with "hash vtables" is that you only really see a performance win when the interface has a lot of specializations.

As I just mentioned in another reply, the problem they were trying to solve was hierarchies where it makes sense for a group of types to be constructed by the combination of two or three narrowly scoped interfaces.

For instance, if you treat some collections as read only, you can define comprehensions across them with a single implementation. But that means the mutators have to be contained in another type, which a subset will implement, and may have covariant inputs.

That's because that type of code is actually better performing than the dynamic dispatch.

There's absolutely nothing wrong with this code. It's just that it's not as extensible

It's a 'closed world' representation where the code assumes it knows about every possibility. This make extension more difficult

The code itself is extraordinarily good and performant.

As someone who's favorite language is C, I don't see what is wrong with that code? Sure, you need to extend it with a new subtype, but you also need to implement every virtual function anyway. And if you use switch instead of an if-else-chain the compiler will complain when you are missing a subtype.

What's wrong with it is, when I extend with a new subtype, I have to fix up the locations that use the type. Potentially all of the locations that use it - I at least have to look at all of them.

With the polymorphic approach, I just have to create the new subtype, and all the users can do the right thing (if they were written with polymorphism in mind, anyway - if they use virtual functions on the base class).

Why would I change the users at all instead of just modifying the dispatch method in the super type?

I see. Yes, you can do it that way.

Still... doing it the C++ way, I can just declare the sub type as deriving from the super type, and I don't have to fix up the super type.

That’s the OO way. Of which C++ is an instance.

I've been thinking through what features I'd want in a language if I were designing one myself, and one of my desires is to have exhaustive matches on enums (which could be made of any primitive type) and sum types. The ability to generate perfect hashes at compile time was one of the things that falls out nicely from that

Reply on news.ycombinator.com