tests/type_confusion: add dav1d_{alloc,free} instead of IA2_SHARED_DATA and make Dav1dContext opaque

[kkysen]

donna is down, so I'm testing this in CI.

@fw-immunant, is this what you meant?

[fw-immunant]

I was suggesting individual wrappers for the dav1d initializer-type functions that would return new, allocated-and-initialized objects and be marked as the constructor for the given type. That said, this also achieves the same memory ownership for the allocated objects, which is the main point: the caller can create objects but can only modify them via the dav1d compartment's API, not by poking at their bytes directly.

[kkysen]

I was suggesting individual wrappers for the dav1d initializer-type functions that would return new, allocated-and-initialized objects and be marked as the constructor for the given type.

Correct me if I'm wrong, but I thought we had settled on a more generic dav1d_alloc/dav1d_free being simpler than individual wrappers for each type. Although this was a while ago so I don't remember too much, and I might be remembering wrong.

That said, this also achieves the same memory ownership for the allocated objects, which is the main point: the caller can create objects but can only modify them via the dav1d compartment's API, not by poking at their bytes directly.

Is this way good then? Or what changes did you want me to still make?

[fw-immunant]

Is this way good then? Or what changes did you want me to still make?

Given discussions we've had, I think we should probably just merge this approach when tests pass; what I really wanted was going to need something like #485 but that was mostly a matter of aesthetics.