> FYI, I recently found out that C# can actually do multiple dispatch
But also good to know, that it can be costly and unsafe. `dynamic` is basically using reflection at run-time, which for most cases is fine, but it's something to keep in mind. It also means that type checking is delayed until run-time instead of compile-time.
A dynamic type is probably just implemented as a tagged bit of data, surely. Which is the exact same way an algebraic datatype would be done. So I dunno why it would be any slower.
I think the safety thing is a non issue as well - what arguments could you pass in to the statically typed function that would cause the use of dynamic to bite you in the ass?
> A dynamic type is probably just implemented as a tagged bit of data, surely
A dynamic "type" is really just `object` whose methods are looked up at run time. It's not tagged data at all. See [1] for more info, specifically the example(s) at the bottom, as the reflection code is what gets executed at run (albeit with caching so that methods aren't looked up all the time).
`dynamic` is a complex feature that enables multiple dispatch as a side-effect, but only because it allows a whole lot more.
> what arguments could you pass in to the statically typed function that would cause the use of dynamic to bite you in the ass
public void Output(int value);
public void Output(Person person);
...
dynamic aValue = "Some String";
Output(aValue);
Compiles, because the actual resolving of which overload to call is done at runtime, not at compile time. And at runtime, there is no overload that accepts a string.
I am talking about what value you could pass into the multiple dispatch code in the article I linked to. I know you can cause run time type errors using dynamic in general, but in the code I linked it was very well contained and water tight.
Dynamically typed languages usually represent objects as something like a struct with an int tag to represent the type, and a void pointer for its value. The actual reflection going on in the code I posted for multiple dispatch would surely be nothing more than an int comparison - same as what Ocaml probably does.
> I am talking about what value you could pass into the multiple dispatch code in the article I linked to
Ah, I misunderstood the question. Yes, in that case it's fairly type-safe.
> The actual reflection going on in the code I posted for multiple dispatch would surely be nothing more than an int comparison
Is this based on gut reaction or are you talking about optimizations that `dynamic` performs? I ask because my understanding is that `dynamic` is like a really efficient reflection-emitter, that attempts to do at runtime the same thing that the compiler would do at compile time, but slower because it has to look stuff up via reflection.
From Eric Lippert [2]:
> The magic is: the compiler emits code that starts the C# compiler again at runtime. The runtime version of the compiler analyzes the call as though the compile-time types of all the objects had been their actual runtime types, generates an expression tree representing that call, compiles the expression tree, caches the delegate for next time, and runs the delegate.
So maybe `dynamic` at runtime figures out that it can do a "a struct with an int tag to represent the type, and a void pointer for its value", but I wouldn't assume it does and from what I've read on it, I wouldn't think that it does.
Base classes already have to carry their runtime type with them anyway. As do nodes of an algebraic data type. You may well be right that the compiler might not be "sufficiently smart" to see this very limited use of dynamic and realise no special reflection magic is actually required. But considering dynamic is only used to actually pull out the runtime type in the example code, and not to call arbitrary methods on it that may or not be there, I assumed - maybe wrongly - that the runtime hit would be the same as the branching that occurs in Ocaml or Haskell or F# when it comes across a "match" statement.
But to answer your question - yes, entirely based on gut reaction and what a compiler would ideally be able to do given the code posted. So very probably wrong.
> I think the safety thing is a non issue as well - what arguments could you pass in to the statically typed function that would cause the use of dynamic to bite you in the ass?
None if the function is written correctly, but by that logic you don't need type safety at all. The point is that the dynamic technique gives you no warning if you forget to implement one of the cases.
My point was that the dynamic technique in the posted code was written in a very small, contained point where it would be impossible due to the signature of the surrounding argument for it to go wrong.
void React(Animal me, Animal other)
{
ReactSpecialization(me as dynamic, other as dynamic);
}
Unless you pass in a casted value here, it's fool proof.
The point is that if you forget to write one of the ReactSpecialization methods nothing detects it. Perhaps more importantly, if you add a new type of Animal there's nothing to tell you you need to write a bunch of new ReactSpecializations.
Only if you deliberately add the catch-all case. If you don't and just handle all the cases explicitly, many languages will give you a warning or error when you add a new unhandled variant.
But also good to know, that it can be costly and unsafe. `dynamic` is basically using reflection at run-time, which for most cases is fine, but it's something to keep in mind. It also means that type checking is delayed until run-time instead of compile-time.