This commit adds lifetimes to the diff and realdom methods so consumers may borrow the contents of the DOM for serialization or asynchronous modifications.
This commit solves the memoization , properly memoizing properties that don't have any generic parameters. This is a rough heuristic to prevent non-static lifetimes from creeping into props and breaking our minual lifetime management.
Props that have a generic parameter are opted-out of the `partialeq` requirement and props *without* lifetimes must implement partialeq. We're going to leave manual disabling of memoization for future work.
Currently working through the normalization process. Occasionally, we want to iterate through all the nodes that we know have a real image in the dom. However, fragments and components don't directly have a mirror in the dom. This commit is exploring the concept of a custom iterator that explores every node in an array of nodes, returning only valid nodes which may be mounted to the dom. A big issue we're working through is heavily nested rootless nodes - something not terribly common but important nonetheless.
Inferno, React, and Preact all perform a mutative-form of normalization which alter the children list before comparing to the previous. Mostly, we're concerned about fragments in lists and heavily nested components that do not render real elements.
This change switches back to the original `ctx<props>` syntax for
commponents. This lets lifetime elision to remove the need to match
exactly which lifetime (props or ctx) gets carried to the output. As
such, `Props` is currently required to be static. It *is* possible to
loosen this restriction, and will be done in the future, though only
through adding metadata about the props through the Props derive
macro. Implementing the IS_STATIC trait is unsafe, so the derive macro
will do it through some heuristics.
For now, this unlocks sharing vnodes from parents to children, enabling
pass-thru components, fragments, portals, etc.
This commit reverts to the old style of props + FC. The old style is desirable
because people comfortable with react can automatically be comfortable with
dioxus. It's also nice in that the same props can be used to drive two different
components - something the trait version couldn't do. Now, our trait bound forces
implementations to have the #[derive(Props)] flag. This will need to implement the
Properties trait as well as PartialEq (using ptr::equal for closure fields).
Instead, we're just going stick with unsafely transmuting the static
lifetime when accessing the active frame. This is a bit scarier, but
greatly simplifies containing the VNode.
