Stream: git-wasmtime

Topic: wasmtime / issue #8433 Inconsistent GC behavior calling a...

Wasmtime GitHub notifications bot (Apr 22 2024 at 16:26):

alexcrichton opened issue #8433:

For this program:

use wasmtime::*;

fn main() {
    let engine = Engine::default();

    let module = Module::new(
        &engine,
        r#"
            (module
                (import "" "a" (func $a (result externref)))

                (func (export "a")
                    (loop $l
                        call $a
                        drop
                        br $l
                    )
                )
            )
        "#,
    )
    .unwrap();

    let mut linker = Linker::new(&engine);
    linker
        .func_wrap("", "a", |mut caller: Caller<'_, ()>| {
            Ok(Some(ExternRef::new(&mut caller, 100)?))
        })
        .unwrap();

    let ty = FuncType::new(&engine, [], [ValType::EXTERNREF]);
    linker
        .func_new("", "b", ty, |mut caller, _, results| {
            results[0] = Some(ExternRef::new(&mut caller, 100)?).into();
            Ok(())
        })
        .unwrap();

    let mut store = Store::new(&engine, ());
    let i = linker.instantiate(&mut store, &module).unwrap();
    let f = i.get_typed_func::<(), ()>(&mut store, "a").unwrap();
    f.call(&mut store, ()).unwrap();
}

it will, as-is, run with:

$ cargo run -q
thread 'main' panicked at src/main.rs:42:28:
called `Result::unwrap()` on an `Err` value: error while executing at wasm backtrace:
    0:   0x2c - <unknown>!<wasm function 1>

Caused by:
    0: failed to allocate `externref`
    1: GC heap out of memory
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

However if the "a" is switched to "b" to use the func_new definition instead of func_wrap, it will run infinitely and have steady memory usage.

The reason is that this block is only present in the func_new path, not the func_wrap path.

cc @fitzgen

Wasmtime GitHub notifications bot (Apr 22 2024 at 16:43):

alexcrichton commented on issue #8433:

In https://github.com/bytecodealliance/wasmtime/pull/8434 I'm making the behavior "consistent" but in the wrong direction. That PR is removing the auto-gc block I outlined above, which was the source of the panic. This is the "wrong direction", however, as we should still retain the gc-before-returning so hosts aren't required necessarily to GC themselves. (or we should perhaps decide that this is indeed the desired semantics).

In that sense I'm going to leave this issue open after that PR lands.

Last updated: Oct 23 2024 at 20:03 UTC