On 24/09/2011 1:02 PM, [email protected] wrote:
This test tries to swap unique boxes containing resources, which is not
allowed.
Hi. I know I'm officially on break now but I did see this go past while
checking on other things, and wanted to point out that (though it's
counter-intuitive) I think this is not quite right. In particular, if
you look at the comment in kind.rs you'll see that it says:
MOVE + NOSEND = "Shared": structures containing @, fixed to the local
task heap/pool; or ~ structures pointing to
pinned values.
That is, a ~ box containing pinned-kind lowers to shared-kind, not to
pinned itself. This is to prevent it moving over channels (localizing it
to the current heap) but not prevent swapping pointer-to-resource, which
seems to be the only useful way to make a resource have indefinite
extent (live as long as a data structure).
I realize that "~resource" doesn't "look shared" in an intuitive sense,
but it's a bit of a corner case and I think it's the right rule; other
rules involve having move/swap/copy discriminate by type constructor
first ("is this a box? if so, look at content-kind, else look at
total-kind") which feels clunkier.
I know It's even probably counter-intuitive overall that ~ is the "most
capable" kind, and that @ is lower than ~, but that's how it seems to
work out if you aim for minimal rules :)
It might make more sense if we renamed the kinds to their heaps:
"exchange", "local", "pinned", or something like that? I'm pretty easy
on that point.
-Graydon
_______________________________________________
Rust-dev mailing list
[email protected]
https://mail.mozilla.org/listinfo/rust-dev
