Re: Intervention needed?

["Randy Brukardt" <randy@rrsoftware.com>]

"Optikos" <optikos@verizon.net> wrote in message 
news:f4eee526-e607-4e1f-843f-9913c9c093f9@googlegroups.com...
...
>Absolutely not.  The cursor cannot dangle after the finalization of the 
>container.
>No dangling after container-finalization means no dangling-of-cursor at 
>all. Q.E.D.

Baloney. The cursor object still exists, and someone can still try to use 
it. Yes, you'll get an exception rather than erroneous execution, but it is 
still dangling (the term in the RM for cursors is "invalid"). (And note that 
there are cheaper ways to get that effect.)

The whole point of the Rust scheme as I understand it is to avoid such 
things. In particular, you can never have a pointer with multiple owners. 
The "borrow" scheme as I understand it allows the owner to be temporarily 
transfered (so you can walk lists, for instance), but long-term storage of 
such pointers is simply not allowed. You can't build a cursor that way (nor 
a doubly-linked list). Perhaps the Rust people have gone beyond simple 
borrow semantics -- I don't know.

Tucker's latest proposal lets one build cursors, but it can't detect changes 
where the owner of a node is changed after the fact (think of an operation 
like a list Splice where nodes are moved from one list to another). To do 
that would require putting a lot of restrictions on how an ADT is used, 
again something that cannot be done with the existing containers (and it's 
unclear that that is a good idea anyway, since it is leaking implementation 
characteristics to usages).

                           Randy.