Why doesn't that work for mutable collections (are you referring to the need to be invariant here?) or immutable collections? More specifically, why can't you always have `Collection<A>`, upcast to `Collection<Object>`, downcast to `Collection<B>` as an escape hatch? You break type safety but erased generics already break type safety (this is especially clear with C#'s choice to follow Java's convention on array variance and the pain that this causes when it's inconsistent with everything else in C#).
For passing back to a typed language, you add a downcast of `Collection<Object>` to `Collection<String>` and have no runtime conversion cost. Why doesn't that work?
EDIT: Ah I see, is the basic problem you don't get free (more accurately cheap) casts, especially with generics? I'm not familiar with how casting works in the CLR since the vast majority of my time is on the JVM.
Yes, or contravariant.
> erased generics already break type safety
They don't, they just don't enforce type safety across languages. BTW, most languages erase types (including Haskell).
> Why doesn't that work?
Because platforms with safe casts do a runtime instanceof test, and if the platform reifies generics, then it is not true that the runtime type of Collection<Object> can be cast to Collection<String>. But if the generic parameter is erased, they both have the same runtime type.
RE type safety, it doesn't really make sense to talk about type safety across languages since type safety is tied to a particular language's semantics. I would argue that every ClassCastException is a violation of type safety in Java the language even if it's reasonable at the JVM level but whatever that's just definitions diving.
I see. Basically what you're saying is that you don't really see examples of reified types with unsafe runtime casts? Presumably because that kind of negates the whole point of reifying the types in the first place?
I don't know if it's "weird" given that there are only two examples, and they're different from one another. It's not like there's a norm here.
> I would argue that every ClassCastException is a violation of type safety in Java the language even if it's reasonable at the JVM level but whatever that's just definitions diving.
But you can't get a CCE in Java without explicit casts (or a compiler bug), just as in C#.
> Basically what you're saying is that you don't really see examples of reified types with unsafe runtime casts?
I'm not sure what you're asking.
List<String> myStuff0 = new LinkedList<>();
List eraseMyStuff = myStuff0;
// Warning here
eraseMyStuff.add(1);
// CCE here
String myString = myStuff0.get(0);
List<String> myStrings = new LinkedList<>();
List<Integer> myIntegers = new LinkedList<>();
myIntegers.add(1);
Stream.of(myStrings, myIntegers)
// Wildcard capture
.map(x -> x)
.reduce((x1, x2) -> {
// Would expect this line to blow up either at compile time or at runtime with reified generics
list1.addAll(x2);
return x1;
});
// Blows up with a CCE
String result = myStrings.get(0);
I thought, however, there was an easier trick to get a CCE using type erasure of generics without warnings and without having to involve capture conversion, but it looks like I may have just been wrong about that as my efforts all ended up getting warnings from javac.
Of course whether any of this actually matters is a separate issue altogether. Java is still a workhorse language that powers a ton of software.
As for weirdness, I can't think of any other language with generics that preserves runtime type information without a way of preserving that for generics as well. Most as you point out either do full erasure (most stuff in the ML-inspired family) or either fully preserve all the information (of course all the .NET languages)/let you opt-in to preserving generics (Scala to some extent with TypeTags) or just forgo generics (Go).
For example in OCaml if you try to do an unsafe and incorrect cast you segfault. IIRC in GHC you end up with a crazy incorrect value (e.g. unsafeCoerce [1, 2, 3] :: Int results in some messed up integer).
My question was me thinking out loud; I was guessing that CIL bytecode has a instruction analogous to checkcast in JVM bytecode that you can just omit and then you can freely cast at will, potentially wreaking havoc if you cast incorrectly. I don't know if that's true. Basically I was separating the verification of type information from the storage of it at runtime. If the CIL separates them then the downcast trick works. If it doesn't the trick doesn't work.
Also in a way this unsafe cast is really just a run around reified types to begin with. I guess the only difference is that you still have storage of the types so you can do runtime reflection. On reflection, I can certainly see how this complicates implementing things like higher-kinded types (because you sort of end up implementing half of type erasure yourself) but I'm not sure it complicates marshalling ordinary data.
Using raw types is cheating. It's essentially using a different language and, as you say, the compiler will tell you that you are.
> However, if you obfuscate it a little javac doesn't know about it anymore.
It's a bug (rather, a few bugs): http://wouter.coekaerts.be/2018/java-type-system-broken (I assume that's where you got the example from)
You shouldn't get a CCE without explicit casts, and if you do, that's a (soundness) bug in the Java compiler and/or spec. Those bugs aren't always fixed quickly when they're not very important in practice (if it's determined that it's hard to hit them by mistake, which is why it's taken 15 years to find them).
> Would expect this line to blow up either at compile-time or at runtime with reified generics
Reified generics have no impact at compile time. Everything a language with reified generics knows at compile-time, a language with erased generics also knows. However, a runtime with reified generics and a similar compiler bug would, indeed, throw at this line at runtime.
> As for weirdness, I can't think of any other language with generics that preserves runtime type information without a way of preserving that for generics as well.
There are good reasons for doing it the Java way: so that you don't bake a language's particular variance strategy into the VM. There are also good reasons for doing it the C#: it's slightly more convenient for that one language. I can't think of any other language with variance and a runtime type system that reifies generics for reference types. When all you have are two instances, it's hard to say which of them is weird.
> If the CIL separates them then the downcast trick works. If it doesn't the trick doesn't work.
I don't know CIL, but looking at how languages behave on .NET to interop, I wouldn't think that can work. It would totally break the safety of the runtime type system if you could cast without a typecheck.
> I guess the only difference is that you still have storage of the types so you can do runtime reflection.
BTW, you can reflect on generic type arguments in Java, as well (https://docs.oracle.com/en/java/javase/13/docs/api/java.base...). When you reflect on a method that returns a List<String>, it will tell you it returns List<String>, but you can't get such a Type from an instance of that list; you can only get a Class.
... because it's baked into the platform.
> It's not apparent to me that e.g. IronPython has a hard time sharing data structures with C#
https://ironpython.net/documentation/dotnet/dotnet.html#acce...
> Why doesn't the usual strategy of upcast everything to `Collection<Object>` within a language where collections are covariant and then downcast as needed in a language that doesn't care about variance (e.g. a dynamic language) work?
Because that does not work for mutable collections, or even for immutable collections created by, say, an untyped language. Clojure can create a list of strings and pass it to a Java method expecting a List<String>.