On 07/06/2011 7:40 AM, Marijn Haverbeke wrote:
Sorry if I'm misunderstanding, but I thought I covered that in my previous
email--we have to forbid access to the expression in an "alt" statement
inside the case blocks.
Your proposal seems to assume you'd only ever directly alias locals.
That is a possible solution, but it seems excessively restrictive and
awkward. I think there's at least 500 cases in rustc where we want to
create an alias to a field or dereferenced box.
Sure. But how many of them are potentially *interfering* with other
aliases in the same scope (under type-based and conservative root-based
analysis)?
I'll reiterate my sense of surprise if we wind up with a system that
doesn't have to do pairwise analysis. I had always assumed this bug was
not about finding a local safety property that each referent needs to
obey, but rather a safe *relationship* between a referent and its
environment.
-Graydon
_______________________________________________
Rust-dev mailing list
[email protected]
https://mail.mozilla.org/listinfo/rust-dev
