Fix #641

[ambiguousname]

Fixes #641

[Manishearth]

issue: I still don't think this is the best way to structure this

I don't think this parameter should be necessary. We're manually doing slice.ptr, slice.size above, so what we have is that two codepaths calling gen_js_to_c_for_type, one which uses ...{}.splat() to splat it, and one which manually spreads it, and we have this conversion parameter to deal with that divergence.

gen_js_to_c_for_type already unconditionally returns a ...struct.toFFI() for structs, so all of its call sites already expect spread operators. I think it should unconditionally return ...slice.splat() for slices (with caveats on the arena behavior). The call sites should be adjusted to handle that for slices; that will simplify them since currently this call site special-cases slices and manually spreads it.

[Manishearth]

(all of this is an unfortunate consequence of how slices need to be special snowflakes for borrowing, which makes it easy to accidentally write code tha treats them as more special than they need to be)

[ambiguousname]

Yeah, it's probably for the best that we try to aim for near-universal usage of ...splat() in gen_js_to_c, it's just an absolute pain for freeing. I think I'll be able to make a clean solution though.

[ambiguousname]

@Manishearth ... ___ .splat() is now always returned by gen_js_to_c. The lifetimeEdges code looks a little weird, but I think it still works.

Here's some sample code for inside of methods, for reference, on Foo:

static new_(x) {
        let functionGarbageCollector = new diplomatRuntime.GarbageCollector();

        const xSlice = [...functionGarbageCollector.alloc(diplomatRuntime.DiplomatBuf.str8(wasm, x)).splat()];
        
        // This lifetime edge depends on lifetimes 'a
        let aEdges = [xSlice];
        const result = wasm.Foo_new(...xSlice);
    
        try {
            return new Foo(diplomatRuntime.internalConstructor, result, [], aEdges);
        }
        
        finally {
            functionGarbageCollector.garbageCollect();}
}

[Manishearth]

I think a potentially cleaner way to do the allocation would be to pass in an allocation retainer object into the ctor for these slice objects, but this works too.