Thought experiment removing ValOwned from traces, batches, cursors, etc. It removes any opinions about the owned forms of the value GAT, other than that implied by MyTrait::Owned (which .. can vary as a function of the lifetime, and probably isn't load bearing at this point).
The pain here is that you do want some relationship between V and Tr::Val<'a>. All we seem to need is a function from the latter to the former, although From and Into do not work because they are not implemented for e.g. &T where T: Clone which we see a lot. Instead, core methods take a from: Fn(Tr::Val<'_>) -> V, which is not the most ergonomic but seemed better than having an ongoing struggle with new traits.
Thought experiment removing
ValOwned
from traces, batches, cursors, etc. It removes any opinions about the owned forms of the value GAT, other than that implied byMyTrait::Owned
(which .. can vary as a function of the lifetime, and probably isn't load bearing at this point).The pain here is that you do want some relationship between
V
andTr::Val<'a>
. All we seem to need is a function from the latter to the former, althoughFrom
andInto
do not work because they are not implemented for e.g.&T
whereT: Clone
which we see a lot. Instead, core methods take afrom: Fn(Tr::Val<'_>) -> V
, which is not the most ergonomic but seemed better than having an ongoing struggle with new traits.