Suspense boundaries/out of order streaming/anyhow like error handling (#2365)

* create static site generation helpers in the router crate

* work on integrating static site generation into fullstack

* move ssg into a separate crate

* integrate ssg with the launch builder

* simplify ssg example

* fix static_routes for child routes

* move CLI hot reloading websocket code into dioxus-hot-reload

* fix some unused imports

* use the same hot reloading websocket code for fullstack

* fix fullstack hot reloading

* move cli hot reloading logic into the hot reload crate

* ssg example working with dx serve

* add more examples

* fix clippy

* switch to a result for Element

* fix formatting

* fix hot reload doctest imports

* fix axum imports

* add show method to error context

* implement retaining nodes during suspense

* fix unterminated if statements

* communicate between tasks and suspense boundaries

* make suspense placeholders easier to use

* implement IntoDynNode and IntoVNode for more wrappers

* fix clippy examples

* fix rsx tests

* add streaming html utilities to the ssr package

* unify hydration and non-hydration ssr cache

* fix router with Result Element

* don't run server doc tests

* Fix hot reload websocket doc examples

* simple apps working with fullstack streaming

* fix preloading wasm

* Report errors encountered while streaming

* remove async from incremental renderer

* document new VirtualDom suspense methods

* make streaming work with incremental rendering

* fix static site generation

* document suspense structs

* create closure type; allow async event handlers in props; allow shorthand event handlers

* test forwarding event handlers with the shorthand syntax

* fix clippy

* fix imports in spawn async doctest

* fix empty rsx

* fix async result event handlers

* fix mounting router in multiple places

* Fix task dead cancel race condition

* simplify diffing before adding suspense

* fix binary size increase

* fix attribute diffing

* more diffing fixes

* create minimal fullstack feature

* smaller fullstack bundles

* allow mounting nodes that are already created and creating nodes without mounting them

* fix hot reload feature

* fix replacing components

* don't reclaim virtual nodes

* client side suspense working!

* fix CLI

* slightly smaller fullstack builds

* fix multiple suspended scopes

* fix merge errors

* yield back to tokio every few polls to fix suspending on many tasks at once

* remove logs

* document suspense boundary and update suspense example

* fix ssg

* make streaming optional

* fix some router and core tests

* fix suspense example

* fix serialization with out of order server futures

* add incremental streaming hackernews demo

* fix hackernews demo

* fix root hackernews redirect

* fix formatting

* add tests for suspense cases

* slightly smaller binaries

* slightly smaller

* improve error handling docs

* fix errors example link

* fix doc tests

* remove log file

* fix ssr cache type inference

* remove index.html

* fix ssg render template

* fix assigning ids on elements with dynamic attributes

* add desktop feature to the workspace examples

* remove router static generation example; ssg lives in the dioxus-static-generation package

* add a test for effects during suspense

* only run effects on mounted nodes

* fix multiple suspense roots

* fix node iterator

* fix closures without arguments

* fix dioxus-core readme doctest

* remove suspense logs

* fix scope stack

* fix clippy

* remove unused suspense boundary from hackernews

* assert that launch never returns for better compiler errors

* fix static generation launch function

* fix web renderer

* pass context providers into server functions

* add an example for FromContext

* clean up DioxusRouterExt

* fix server function context

* fix fullstack desktop example

* forward CLI serve settings to fullstack

* re-export serve config at the root of fullstack

* forward env directly instead of using a guard

* just set the port in the CLI for fullstack playwright tests

* fix fullstack dioxus-cli-config feature

* fix launch server merge conflicts

* fix fullstack launch context

* Merge branch 'main' into suspense-2.0

* fix fullstack html data

* remove drop virtual dom feature

* add a comment about only_write_templates binary size workaround

* remove explicit dependencies from use_server_future

* make ErrorContext and SuspenseContext more similar

* Tweak: small tweaks to tomls to make diff smaller

* only rerun components under suspense after the initial placeholders are sent to the client

* add module docs for suspense

* keep track of when suspense boundaries are resolved

* start implementing JS out of order streaming

* fix core tests

* implement the server side of suspense with js

* fix streaming ssr with nested suspense

* move streaming ssr code into fullstack

* revert minification changes

* serialize server future data as the html streams

* start loading scripts wasm immediately instead of defering the script

* very basic nested suspense example working with minimal html updates

* clean up some suspense/error docs

* fix hydrating nested pending server futures

* sort resolved boundaries by height

* Fix disconnecting clients while streaming

* fix static generation crate

* don't insert extra divs when hydrating streamed chunks

* wait to swap in the elements until they are hydrated

* remove inline streaming script

* hackernews partially working

* fix spa mode

* banish the open shadow dom

* fix removing placeholder

* set up streaming playwright test

* run web playwright tests on 9999 to avoid port conflicts with other local servers

* remove suspense nodes if the suspense boundary is replaced before the suspense resolves on the server

* ignore hydration of removed suspense boundaries

* use path based indexing to fix hydrating suspense after parent suspense with child is removed

* re-export dioxus error

* remove resolved suspense divs if the suspense boundary has been removed

* Fix client side initialized server futures

* ignore comment nodes while traversing nodes in core to avoid lists getting swapped out with suspense

* Pass initial hydration data to the client

* hide pre nodes

* don't panic if reclaiming an element fails

* fix scope stack when polling tasks

* improve deserialization out of length message

* Ok(VNode::placeholder()) -> VNode::empty()

* fix typo in rsx usage

* restore testing changes from suspense example

* clean up some logs and comments

* fix playwright tests

* clean up more changes in core

* clean up core tests

* remove anymap dependency

* clean up changes to hooks

* clean up changes in the router, rsx, and web

* revert changes to axum-hello-world

* fix use_server_future

* fix clippy in dioxus-core

* check that the next or previous node exist before checking if we should ignore them

* fix formatting

* fix suspense playwright test

* remove unused suspense code

* add more suspense playwright tests

* add more docs for error boundaries

* fix suspense core tests

* fix ErrorBoundary example

* remove a bunch of debug logging in js

* fix router failure_external_navigation

* use absolute paths in the interpreter build.rs

* strip '\r' while hashing ts files

* add a wrapper with a default error boundary and suspense boundary

* restore hot reloading

* ignore non-ts files when hashing

* sort ts files before hashing them

* fix rsx tests

* fix fullstack doc tests

* fix core tests

* fix axum auth example

* update suspense hydration diagram

* longer playwright build limit

* tiny fixes - spelling, formatting

* update diagram link

* remove comment and template nodes for suspense placeholders

* remove comment nodes as we hydrate text

* simplify hackernews example

* clean up hydrating text nodes

* switch to a separate environment variable for the base path for smaller binaries

* clean up file system html trait

* fix form data

* move streaming code into fullstack

* implement serialize and deserialize for CapturedError

* remove waits in the nested suspense playwright spec

* force sequential fullstack builds for CI

* longer nested suspense delay for CI

* fix --force-sequential flag

* wait to launch server until client build is done

---------

Co-authored-by: Jonathan Kelley <jkelleyrtp@gmail.com>

packages/core/src/diff/component.rs

@@ -1,56 +1,125 @@
-use std::ops::{Deref, DerefMut};
+use std::{
+    any::TypeId,
+    ops::{Deref, DerefMut},
+};
 
 use crate::{
     any_props::AnyProps,
-    innerlude::{ElementRef, MountId, ScopeOrder, VComponent, WriteMutations},
-    nodes::RenderReturn,
+    innerlude::{
+        ElementRef, MountId, ScopeOrder, SuspenseBoundaryProps, SuspenseBoundaryPropsWithOwner,
+        VComponent, WriteMutations,
+    },
     nodes::VNode,
     scopes::ScopeId,
     virtual_dom::VirtualDom,
+    RenderReturn,
 };
 
 impl VirtualDom {
-    pub(crate) fn diff_scope(
+    pub(crate) fn run_and_diff_scope<M: WriteMutations>(
         &mut self,
-        to: &mut impl WriteMutations,
+        to: Option<&mut M>,
+        scope_id: ScopeId,
+    ) {
+        let scope = &mut self.scopes[scope_id.0];
+        if SuspenseBoundaryProps::downcast_mut_from_props(&mut *scope.props).is_some() {
+            SuspenseBoundaryProps::diff(scope_id, self, to)
+        } else {
+            let new_nodes = self.run_scope(scope_id);
+            self.diff_scope(to, scope_id, new_nodes);
+        }
+    }
+
+    #[tracing::instrument(skip(self, to), level = "trace", name = "VirtualDom::diff_scope")]
+    fn diff_scope<M: WriteMutations>(
+        &mut self,
+        to: Option<&mut M>,
         scope: ScopeId,
         new_nodes: RenderReturn,
     ) {
-        self.runtime.scope_stack.borrow_mut().push(scope);
+        // We don't diff the nodes if the scope is suspended or has an error
+        let Ok(new_real_nodes) = &new_nodes.node else {
+            return;
+        };
+
+        self.runtime.push_scope(scope);
         let scope_state = &mut self.scopes[scope.0];
-        // Load the old and new bump arenas
-        let new = &new_nodes;
+        // Load the old and new rendered nodes
         let old = scope_state.last_rendered_node.take().unwrap();
 
-        old.diff_node(new, self, to);
+        // If there are suspended scopes, we need to check if the scope is suspended before we diff it
+        // If it is suspended, we need to diff it but write the mutations nothing
+        // Note: It is important that we still diff the scope even if it is suspended, because the scope may render other child components which may change between renders
+        let mut render_to = to.filter(|_| self.runtime.scope_should_render(scope));
+        old.diff_node(new_real_nodes, self, render_to.as_deref_mut());
 
-        let scope_state = &mut self.scopes[scope.0];
-        scope_state.last_rendered_node = Some(new_nodes);
+        self.scopes[scope.0].last_rendered_node = Some(new_nodes);
 
-        self.runtime.scope_stack.borrow_mut().pop();
+        if render_to.is_some() {
+            self.runtime.get_state(scope).unwrap().mount(&self.runtime);
+        }
+
+        self.runtime.pop_scope();
     }
 
-    /// Create a new template [`VNode`] and write it to the [`Mutations`] buffer.
+    /// Create a new [`ScopeState`] for a component that has been created with [`VirtualDom::create_scope`]
     ///
-    /// This method pushes the ScopeID to the internal scopestack and returns the number of nodes created.
-    pub(crate) fn create_scope(
+    /// Returns the number of nodes created on the stack
+    #[tracing::instrument(skip(self, to), level = "trace", name = "VirtualDom::create_scope")]
+    pub(crate) fn create_scope<M: WriteMutations>(
         &mut self,
-        to: &mut impl WriteMutations,
+        to: Option<&mut M>,
         scope: ScopeId,
-        new_node: RenderReturn,
+        new_nodes: RenderReturn,
         parent: Option<ElementRef>,
     ) -> usize {
-        self.runtime.scope_stack.borrow_mut().push(scope);
+        self.runtime.push_scope(scope);
+
+        // If there are suspended scopes, we need to check if the scope is suspended before we diff it
+        // If it is suspended, we need to diff it but write the mutations nothing
+        // Note: It is important that we still diff the scope even if it is suspended, because the scope may render other child components which may change between renders
+        let mut render_to = to.filter(|_| self.runtime.scope_should_render(scope));
 
         // Create the node
-        let nodes = new_node.create(self, to, parent);
+        let nodes = new_nodes.create(self, parent, render_to.as_deref_mut());
 
         // Then set the new node as the last rendered node
-        self.scopes[scope.0].last_rendered_node = Some(new_node);
+        self.scopes[scope.0].last_rendered_node = Some(new_nodes);
 
-        self.runtime.scope_stack.borrow_mut().pop();
+        if render_to.is_some() {
+            self.runtime.get_state(scope).unwrap().mount(&self.runtime);
+        }
+
+        self.runtime.pop_scope();
         nodes
     }
+
+    pub(crate) fn remove_component_node<M: WriteMutations>(
+        &mut self,
+        to: Option<&mut M>,
+        destroy_component_state: bool,
+        scope_id: ScopeId,
+        replace_with: Option<usize>,
+    ) {
+        // If this is a suspense boundary, remove the suspended nodes as well
+        if let Some(mut suspense) =
+            SuspenseBoundaryProps::downcast_mut_from_props(&mut *self.scopes[scope_id.0].props)
+                .cloned()
+        {
+            suspense.remove_suspended_nodes::<M>(self, destroy_component_state);
+        }
+
+        // Remove the component from the dom
+        if let Some(node) = self.scopes[scope_id.0].last_rendered_node.as_ref() {
+            node.clone_mounted()
+                .remove_node_inner(self, to, destroy_component_state, replace_with)
+        };
+
+        if destroy_component_state {
+            // Now drop all the resources
+            self.drop_scope(scope_id);
+        }
+    }
 }
 
 impl VNode {
@@ -63,7 +132,7 @@ impl VNode {
         scope_id: ScopeId,
         parent: Option<ElementRef>,
         dom: &mut VirtualDom,
-        to: &mut impl WriteMutations,
+        to: Option<&mut impl WriteMutations>,
     ) {
         // Replace components that have different render fns
         if old.render_fn != new.render_fn {
@@ -83,9 +152,8 @@ impl VNode {
             return;
         }
 
-        // Now run the component and diff it
-        let new = dom.run_scope(scope_id);
-        dom.diff_scope(to, scope_id, new);
+        // Now diff the scope
+        dom.run_and_diff_scope(to, scope_id);
 
         let height = dom.runtime.get_state(scope_id).unwrap().height;
         dom.dirty_scopes.remove(&ScopeOrder::new(height, scope_id));
@@ -98,16 +166,21 @@ impl VNode {
         new: &VComponent,
         parent: Option<ElementRef>,
         dom: &mut VirtualDom,
-        to: &mut impl WriteMutations,
+        mut to: Option<&mut impl WriteMutations>,
     ) {
         let scope = ScopeId(dom.mounts[mount.0].mounted_dynamic_nodes[idx]);
 
-        let m = self.create_component_node(mount, idx, new, parent, dom, to);
+        // Remove the scope id from the mount
+        dom.mounts[mount.0].mounted_dynamic_nodes[idx] = ScopeId::PLACEHOLDER.0;
+        let m = self.create_component_node(mount, idx, new, parent, dom, to.as_deref_mut());
 
         // Instead of *just* removing it, we can use the replace mutation
-        dom.remove_component_node(to, scope, Some(m), true);
+        dom.remove_component_node(to, true, scope, Some(m));
     }
 
+    /// Create a new component (if it doesn't already exist) node and then mount the [`ScopeState`] for a component
+    ///
+    /// Returns the number of nodes created on the stack
     pub(super) fn create_component_node(
         &self,
         mount: MountId,
@@ -115,19 +188,40 @@ impl VNode {
         component: &VComponent,
         parent: Option<ElementRef>,
         dom: &mut VirtualDom,
-        to: &mut impl WriteMutations,
+        to: Option<&mut impl WriteMutations>,
     ) -> usize {
-        // Load up a ScopeId for this vcomponent. If it's already mounted, then we can just use that
-        let scope = dom
-            .new_scope(component.props.duplicate(), component.name)
-            .state()
-            .id;
+        // If this is a suspense boundary, run our suspense creation logic instead of running the component
+        if component.props.props().type_id() == TypeId::of::<SuspenseBoundaryPropsWithOwner>() {
+            return SuspenseBoundaryProps::create(mount, idx, component, parent, dom, to);
+        }
 
-        // Store the scope id for the next render
-        dom.mounts[mount.0].mounted_dynamic_nodes[idx] = scope.0;
+        let mut scope_id = ScopeId(dom.mounts[mount.0].mounted_dynamic_nodes[idx]);
 
-        let new = dom.run_scope(scope);
+        // If the scopeid is a placeholder, we need to load up a new scope for this vcomponent. If it's already mounted, then we can just use that
+        if scope_id.is_placeholder() {
+            scope_id = dom
+                .new_scope(component.props.duplicate(), component.name)
+                .state()
+                .id;
 
-        dom.create_scope(to, scope, new, parent)
+            // Store the scope id for the next render
+            dom.mounts[mount.0].mounted_dynamic_nodes[idx] = scope_id.0;
+
+            // If this is a new scope, we also need to run it once to get the initial state
+            let new = dom.run_scope(scope_id);
+
+            // Then set the new node as the last rendered node
+            dom.scopes[scope_id.0].last_rendered_node = Some(new);
+        }
+
+        let scope = ScopeId(dom.mounts[mount.0].mounted_dynamic_nodes[idx]);
+
+        let new_node = dom.scopes[scope.0]
+            .last_rendered_node
+            .as_ref()
+            .expect("Component to be mounted")
+            .clone();
+
+        dom.create_scope(to, scope, new_node, parent)
     }
 }