Cross-memory model data

[rust-highfive]

Issue by larsbergstrom
Wednesday Jul 16, 2014 at 13:31 GMT

For earlier discussion, see rust-lang/rust#15715

This issue was labelled with: A-servo in the Rust repository

---

In our current model, any time that we return a JS object to Rust, we carefully thread in some Rust code that ensures the JS GC adds it to the root set and unregister it when Rust drops the value.

The ultimate goal would be for the Rust compiler to generate stack / register maps for JS-owned values so that the JS GC can directly walk the Rust stack. This would eliminate the need for explicit rooting and would reduce constraints on JS GC implementation.

CC: @zwarich @pcwalton

[Centril]

@larsbergstrom Could you elaborate on this issue? It's hard to see what it entails.

[larsbergstrom]

Let me bring in @jdm & @asajeffrey for any relevant clarification or closure if we've decided better support for integration of GC'd languages with Rust is no longer an issue.

This is a super-old issue, and I know that @pnkfelix also spent a ton of time looking into whether this was feasible or not. IIRC we ultimately abandoned the approach, but that was a couple of years ago.

[jdm]

The presence of reliable stack maps accessible at any point in execution would be a huge boon for integrating a JS GC (and presumably other GCs) in terms of developer ergonomics. It's certainly not something that I'm making any bets on, however, and I don't see a good reason to track the idea in an issue in this repository.

[Centril]

@jdm Can you provide any resources as to what "reliable stack maps" mean?

[jdm]

https://llvm.org/docs/StackMaps.html