This PR rebrands MyTrait<'a> as IntoOwned<'a>, an analogue of ToOwned that is more GAT-friendly. GATs can implement IntoOwned<'_> and support the re-borrowing of the owned type back into the GAT. The existing ToOwned infrastructure relies instead on Rust's Borrow, which can only borrow into a reference to a Borrowed type.
The PR also streamlines MyTrait, removing comparison functionality from it. The inequalities were unused, and the equality method can be locally implemented using borrow_as, as our implementors all support co-variant PartialOrd implementations.
Some other light changes around removing the push functionality for BatchContainer preceded this work (pointed out that the remaining complexity was MyTrait), but I think we agree that these changes are good too.
Only known regression is in rhh.rs, where previously we supported a non-allocating hash evaluation, but now call into_owned(). We can shake that out once things are more stable and rhh is actually expected to work.
frankmcsherry
commented
1 month ago
This PR rebrands
MyTrait<'a>asIntoOwned<'a>, an analogue ofToOwnedthat is more GAT-friendly. GATs can implementIntoOwned<'_>and support the re-borrowing of the owned type back into the GAT. The existingToOwnedinfrastructure relies instead on Rust'sBorrow, which can only borrow into a reference to aBorrowedtype.The PR also streamlines
MyTrait, removing comparison functionality from it. The inequalities were unused, and the equality method can be locally implemented usingborrow_as, as our implementors all support co-variantPartialOrdimplementations.Some other light changes around removing the
pushfunctionality forBatchContainerpreceded this work (pointed out that the remaining complexity wasMyTrait), but I think we agree that these changes are good too.Only known regression is in
rhh.rs, where previously we supported a non-allocating hash evaluation, but now callinto_owned(). We can shake that out once things are more stable andrhhis actually expected to work.