cherry-pick some refactorings

[nikomatsakis]

I had a branch implementing the ability for Rust types to implement Java callbacks. I'm taking a fresh approach but these commits are some small refactorings that I want to land first and which will be used by the new code.

r? @rcoh

[rcoh]

trying to grok the JvmRefOp—my understanding is:

Previously, we only had (by analogy), Future. But we would also like IntoFuture because that's much more convenient to implement. Which on its own wouldn't require the creation of JvmRefOp — but IntoJava only makes sense for references not, scalars?

[nikomatsakis]

I should document the traits -- and we should consider if they need a new name. Here is the motivation, for the record: Before we accepted `IntoJava<J>` which was an alias for a `JvmOp` that produces a Java object of type `J` (technically, something that could be deref'd to `&J`, essentially, so that covers a `Local<J>` and a `Java<J>` etc). But to support callbacks, I want to be able to take a Rust value and then do some preprocessing before invoking the jni method. But this preprocessing should take place with a jvm in scope. So I changed it so that `IntoJava`, as you surmised, has a role more like `IntoFuture` -- it is something that can be made into a `JvmOp`, but not necessarily a `JvmOp` itself. And then I added a trait alias `JvmRefOp<J>` for a `JvmOp` that produces an object of type `J`. What I had planned (and still do plan) is to be able to (1) take a Rust value that implements a Java trait; (2) box it; (3) create a java object that owns that boxed pointer (i.e., we mem::forget the box); (4) when that java object is GC'd, we'll call back into the Rust crate to free the box. I don't 100% remember why this couldn't be done by making the Rust type into a `JvmOp`, maybe it could but it seemed bad for other reasons?

…

On Mon, Jul 15, 2024, at 5:41 PM, Russell Cohen wrote: trying to grok the JvmRefOp—my understanding is: Previously, we only had (by analogy), `Future`. But we would also like `IntoFuture` because that's much more convenient to implement. Which on its own wouldn't require the creation of JvmRefOp — but `IntoJava` only makes sense for references not, scalars? — Reply to this email directly, view it on GitHub <#177 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AABF4ZSBU2ETMJDPLOJA5LTZMQ6XTAVCNFSM6AAAAABK5H3CNCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMRZGQ4DQMBWG4>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[nikomatsakis]

I'm going to land this set of changes, but let's revisit the traits more holistically.