Non-nullable offsets in null objects...

[behdad]

And other objects that have variable length external to the struct itself, cannot be safely placed on the null pool.

[behdad]

fvar InstanceRecord is another example.

Basically, types where even the null instance has it's size external to the bytes themselves. A good start is to NOT define min_size for these objects.

[behdad]

With our new Null() checking that object has null_size, we don't allow OffsetTo to types without Null, even if OffsetTo has is_null=false. Codebase seems happy even though that's a limitation. So, this is not a problem anymore.

[behdad]

Humm. While I've now fixed disallowing taking Null() of objects with unbounded size, the original report in the subject still is valid. Ie. if a struct has a non-nullable offsets to an unbounded struct (currently allowed), the outer struct itself essentially becomes unbounded and should become non-nullable. Unless it guarantees to access inner struct (referred to by offset) properly. I don't think there's a way to enforce that...

[behdad]

I don't think there's a way to enforce that...

Now that we use C++11 enforcing these kinds of things has become easier, I mean, maybe possible.

[behdad]

Humm. While I've now fixed disallowing taking Null() of objects with unbounded size, the original report in the subject still is valid. Ie. if a struct has a non-nullable offsets to an unbounded struct (currently allowed), the outer struct itself essentially becomes unbounded and should become non-nullable. Unless it guarantees to access inner struct (referred to by offset) properly. I don't think there's a way to enforce that...

So a non-nullable offset to an unbounded size should make the offset refuse to be embedded in Null objects by way of metaprogramming magic. Let me try.