Conversion of boxed slices to Arc and IVec #1108
Conversation
|
Whoops, I goofed up MSRV, will take another look at Layout construction |
|
Has miri thrown any complaints when run against a variety of sizes? |
|
It looks good. I got a regular old panic when using a ZST, but I fixed that above. |
src/arc.rs
Outdated
| ptr::write( | ||
| &mut dst as *mut _ as *mut *mut u8, | ||
| dst_thin as *mut u8, | ||
| ); |
There was a problem hiding this comment.
This is the part that I'm stumbling a bit over. I'm a bit worried that this might break if the internal layout for fat pointers changes. The reason I went with the fatten method from https://users.rust-lang.org/t/construct-fat-pointer-to-struct/29198/9 was because @dtolnay provided it as a way to fatten things without assuming the memory layout. I think I want to avoid this if a later version of Rust changes the internal layout for fat pointers and causes this to break. I'm having trouble knowing if this will also break.
There was a problem hiding this comment.
Ah, good point. I can make use of that to write a narrower impl<T> From<Box<[T]>>.
This implementation does not rely on fat pointer layout
75a7e07 to
fc8f9aa
Compare
|
Looks good! Thanks for fixing this! |
This PR rewrites the (unsafe code!)
From<Box<T>> for Arc<T>implementation to allow dynamically sized types, likestd::sync::Arcdoes. Constructing a fat pointer to a dynamically sizedArcInner<T>from a*mut u8thin pointer and the source fat pointer requires some extra gymnastics, see the comments for details. This implementation is then used to add back theFrom<Box<[u8]>> for IVecimplementation. Also included: rewrite of Arc's Debug implementation to allow dynamically sized types, and a small test for the conversion. The unsafe code is modeled after the standard library's code, so it ought to be okay, plus I'm going to run tests with Miri overnight.Fixes #1107