Thanks for your experience, really. Here's hoping.

Does Go generics have type erasure? If not, OP can expect and even better journey.

There are two meanings of "type erasure". One concerns the user, the other the compiler writer.

When most users people ask about type erasure, they ask whether types are erased when compiling the reflection information for polymorphic values (like Java does). Go doesn't have polymorphic values, so the question doesn't apply.

Type erasure means something more general than that to compiler people. Type erasure refers to whether you can compile code while erasing its type, and it's generally a useful property. In general statically-typed programming languages are implemented with type erasure, while dynamically-typed languages are not. Of course, since Go has reflection, it can't do type erasure. Interfaces also conflict with type erasure somewhat, though as expected, during the code generation types are erased as much as possible.

As a comparison, Rust does type erasure fully.

So Haskell has type erasure only for data types that do not derive Typeable?

No, because all types have a Typeable instance! Haskell has type erasure, and then using Typeable adds the type information back exactly where it is needed.

No, they don't have type erasure.

Go is getting monomorphization, separate machine code for each type specialization of a generic function. I assume reflection will choose the one for the args you want to pass, or make you do that.

> I assume reflection will choose the one for the args you want to pass, or make you do that.

According to the spec:

> It's impossible for non-generic code to refer to generic code without instantiating it, so there is no reflection information for uninstantiated generic types or functions.

You won't be able to reflect a generic List or List[T] at all which… makes sense, I think? `reflect` works on values at runtime, so it necessarily works with instantiated types as you can't have a value of an un-instantiated (generic) type: what would you give to `reflect.TypeOf` which could return an uninstantiated generic type?

I was thinking of generic functions rather than (non-generic) methods on generic types. Unfortunately it looks like reflection can’t look up functions in a module at all, which is surprisingly broken.

> Unfortunately it looks like reflection can’t look up functions in a module at all, which is surprisingly broken.

The more time passes the more I think ubiquitous reflection in statically typed languages is what's broken (and a reflection (heh) of deep limitations of the language).

So while I could understand having issues with generics breaking existing reflection code (which doesn't seem to be the case), I'm not going to be sad over reflection having never supported toplevel functions in the first place.

What good are top-level functions if they’re unreachable from any script interpreter or management RPC server or other languages’ FFI?

I fail to see what any of this has to do with reflection.

> any script interpreter

What are you even talking about.

> or management RPC server

I would very much expect RPC endpoints to be opted in explicitly and statically.

> or other languages’ FFI?

I fail to see why Go-level reflection would be involved in that in any capacity.

> What good are top-level functions

You can call them? From your go code?

Are you saying rust or C can't be because they don't have reflection? That's a strange position.

When I type at a Groovy REPL, an interpreter is using reflection for nearly everything. Same goes for native code that has to use reflection via JNI entry points to call Java. It’s damn useful. At the heart of every debugger is a half-finished reflection API that deserves to be cleaned up and surfaced for language interop and novel kinds of tools.

Have a look at D and .NET 5 for compile time reflection.

Or in Java's case annotation processors.

Yes, not everything gets exposed, but it also does the job.

Functions are useful because you can call them from your program.

With Java, I didn't find it to be either a huge win or a major loss. And I still don't. It's basically window dressing that lets you avoid casts in certain obvious places where they weren't particularly problematic anyhow.

It's more than window dressing. A year ago I added correct generics to about 50 000 lines of Java, an old module in a bigger program. The change was mostly mechanical, removing casts and adding signatures.

I found a few places where refactoring had left behind a wrong type cast. Production logs demonstrated actual crashes corresponding to these cast. There were bugs in the issue tracker, marked unresolvable.

Interesting, I'd never encountered such bugs, but it makes sense