On 07/06/2011 8:45 AM, Andreas Gal wrote:
So, I'm about ready to try an 'everything is garbage collected'
approach. There is a reason almost all recent languages have moved to
that...
Patrick and I were chatting about this yesterday. He feels that if we had
unique pointers (which is really reference counting up to a maximum of 1
reference), a lot of places would use them (maybe even most applications where
we use reference counting right now). The rest a GC could handle. If this is
really the case, giving up generalized ref counting doesn't sound too bad.
Ref counting is mostly orthogonal to the alias-analysis problem; there's
a relationship to one of the two induction hypotheses to prove, but even
then, only if you use a very conservative collector. Please don't
conflate the two. "Just use GC everywhere" does not solve the second
induction hypothesis (non-interference) so doesn't save us from having
to dig into this: you can still mutate a multiply-referenced interior
variant underfoot and cause someone to point into garbage memory.
IOW: please, everyone, stop saying "let's just use GC". It's no more an
answer than "let's just do what C++ does" for scheduling. The problem is
not actually solved by that answer. It's a partial and unsafe answer
that paints us into a much harder to escape corner.
(This is not to say we're not having a GC. We've been committed to
having a GC for the GC-allowed kinds of memory from ... at least a year
and a half. And we may lean on it more or less heavily depending on the
refcount costs we observe *once everything else is straightened out*.
But again, mostly orthogonal to alias analysis.)
-Graydon
_______________________________________________
Rust-dev mailing list
[email protected]
https://mail.mozilla.org/listinfo/rust-dev
