TypeContents code is inscrutable

[nikomatsakis]

I can say that because I wrote it. Anyway, I have a refactoring I think is cleaner -- it does a better job of distinguishing owned from reachable properties and so forth. Coming soon to a PR near you.

[thestinger]

@nikomatsakis: I'm curious about what the functional difference will be between owned and reachable properties. There are types like Rc sitting on the boundary between those two, because we could add features like copy-on-write via &mut self, where it copies if the reference count is not one.

[nikomatsakis]

@thestinger I don't quite know what you mean by "functional difference". In any case, I actually drew 3 distinctions: things that are interior to the type, things that are owned by the type, and things that are reachable. For many purposes (e.g. builtin bounds like Send or Freeze), reachable is generally all that matters. But for others, e.g. move-by-default, ownership is what is important. In some cases, like @T, both matters in different contexts: we need to run drop glue if we own the @T, but not if we can merely reach it (e.g., via a borrowed pointer like &@T).

[thestinger]

Ah, that makes sense.

[nikomatsakis]

Ah, that makes sense.

Yes unfortunately there is still some obscure error I've yet to track
down that leads to assertion failures and double frees related to
managed objects. Clearly I'm doing something different.