> This can't be done with just object destructors as ref's and owned ref's are
> created by compiler magic and we can't make custom destructors for them.
How to deal with that problem is part of the spec, see `.nodestroy`.
> If this example and my thinking are correct, we are doing a lot of work on
> something that doesn't seem to achieve the goals, unless we can extend some
> ideas to include this case?
I hope, it will achieve the goals that I claimed, I never claimed to care about
this list-business, it's a dead-end for performance ever since we got hardware
caches in the 90ies.
> Reference counting is not able to be turned off as the source code says that
> would be an "unsafe" option.
Yes, you keep making this claim. It doesn't suddenly become true though. A
dangling ref detector can always be turned off, just like I can decide to
**not** run my code through Valgrind.
template own[T](x: T): owned T = # complement to `unown`
var rslt: owned T
cast[ptr T](rslt.unsafeAddr)[] = x
rslt
Run
Well at this point you're on your own and way beyond any proposal for the new
runtime. You cannot get ownership back via casting.
