Files
pezkuwi-subxt/substrate/utils/fork-tree/src/lib.rs
T
Davide Galassi 678fd0f52f Fork-Tree import requires post-order DFS traversal (#11531)
* Fork-tree insert requires post-order dfs traversal

* Add dedicated test for methods requireing post-order traversal
2022-05-27 13:29:31 +02:00

1533 lines
47 KiB
Rust

// This file is part of Substrate.
// Copyright (C) 2019-2022 Parity Technologies (UK) Ltd.
// SPDX-License-Identifier: Apache-2.0
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//! Utility library for managing tree-like ordered data with logic for pruning
//! the tree while finalizing nodes.
#![warn(missing_docs)]
use codec::{Decode, Encode};
use std::{cmp::Reverse, fmt};
/// Error occurred when iterating with the tree.
#[derive(Clone, Debug, PartialEq)]
pub enum Error<E> {
/// Adding duplicate node to tree.
Duplicate,
/// Finalizing descendent of tree node without finalizing ancestor(s).
UnfinalizedAncestor,
/// Imported or finalized node that is an ancestor of previously finalized node.
Revert,
/// Error throw by client when checking for node ancestry.
Client(E),
}
impl<E: std::error::Error> fmt::Display for Error<E> {
fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
let message = match *self {
Error::Duplicate => "Hash already exists in Tree".into(),
Error::UnfinalizedAncestor => "Finalized descendent of Tree node without finalizing its ancestor(s) first".into(),
Error::Revert => "Tried to import or finalize node that is an ancestor of a previously finalized node".into(),
Error::Client(ref err) => format!("Client error: {}", err),
};
write!(f, "{}", message)
}
}
impl<E: std::error::Error> std::error::Error for Error<E> {
fn cause(&self) -> Option<&dyn std::error::Error> {
None
}
}
impl<E: std::error::Error> From<E> for Error<E> {
fn from(err: E) -> Error<E> {
Error::Client(err)
}
}
/// Result of finalizing a node (that could be a part of the tree or not).
#[derive(Debug, PartialEq)]
pub enum FinalizationResult<V> {
/// The tree has changed, optionally return the value associated with the finalized node.
Changed(Option<V>),
/// The tree has not changed.
Unchanged,
}
/// Filtering action.
#[derive(Debug, PartialEq)]
pub enum FilterAction {
/// Remove the node and its subtree.
Remove,
/// Maintain the node.
KeepNode,
/// Maintain the node and its subtree.
KeepTree,
}
/// A tree data structure that stores several nodes across multiple branches.
///
/// Top-level branches are called roots. The tree has functionality for
/// finalizing nodes, which means that that node is traversed, and all competing
/// branches are pruned. It also guarantees that nodes in the tree are finalized
/// in order. Each node is uniquely identified by its hash but can be ordered by
/// its number. In order to build the tree an external function must be provided
/// when interacting with the tree to establish a node's ancestry.
#[derive(Clone, Debug, Decode, Encode, PartialEq)]
pub struct ForkTree<H, N, V> {
roots: Vec<Node<H, N, V>>,
best_finalized_number: Option<N>,
}
impl<H, N, V> ForkTree<H, N, V>
where
H: PartialEq,
N: Ord,
{
/// Create a new empty tree.
pub fn new() -> ForkTree<H, N, V> {
ForkTree { roots: Vec::new(), best_finalized_number: None }
}
/// Rebalance the tree, i.e. sort child nodes by max branch depth (decreasing).
///
/// Most operations in the tree are performed with depth-first search
/// starting from the leftmost node at every level, since this tree is meant
/// to be used in a blockchain context, a good heuristic is that the node
/// we'll be looking for at any point will likely be in one of the deepest chains
/// (i.e. the longest ones).
pub fn rebalance(&mut self) {
self.roots.sort_by_key(|n| Reverse(n.max_depth()));
let mut stack: Vec<_> = self.roots.iter_mut().collect();
while let Some(node) = stack.pop() {
node.children.sort_by_key(|n| Reverse(n.max_depth()));
stack.extend(node.children.iter_mut());
}
}
/// Import a new node into the tree. The given function `is_descendent_of`
/// should return `true` if the second hash (target) is a descendent of the
/// first hash (base). This method assumes that nodes in the same branch are
/// imported in order.
///
/// Returns `true` if the imported node is a root.
// WARNING: some users of this method (i.e. consensus epoch changes tree) currently silently
// rely on a **post-order DFS** traversal. If we are using instead a top-down traversal method
// then the `is_descendent_of` closure, when used after a warp-sync, may end up querying the
// backend for a block (the one corresponding to the root) that is not present and thus will
// return a wrong result.
pub fn import<F, E>(
&mut self,
hash: H,
number: N,
data: V,
is_descendent_of: &F,
) -> Result<bool, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
{
if let Some(ref best_finalized_number) = self.best_finalized_number {
if number <= *best_finalized_number {
return Err(Error::Revert)
}
}
let (children, is_root) =
match self.find_node_where_mut(&hash, &number, is_descendent_of, &|_| true)? {
Some(parent) => (&mut parent.children, false),
None => (&mut self.roots, true),
};
if children.iter().any(|elem| elem.hash == hash) {
return Err(Error::Duplicate)
}
children.push(Node { data, hash, number, children: Default::default() });
if children.len() == 1 {
// Rebalance may be required only if we've extended the branch depth.
self.rebalance();
}
Ok(is_root)
}
/// Iterates over the existing roots in the tree.
pub fn roots(&self) -> impl Iterator<Item = (&H, &N, &V)> {
self.roots.iter().map(|node| (&node.hash, &node.number, &node.data))
}
fn node_iter(&self) -> impl Iterator<Item = &Node<H, N, V>> {
// we need to reverse the order of roots to maintain the expected
// ordering since the iterator uses a stack to track state.
ForkTreeIterator { stack: self.roots.iter().rev().collect() }
}
/// Iterates the nodes in the tree in pre-order.
pub fn iter(&self) -> impl Iterator<Item = (&H, &N, &V)> {
self.node_iter().map(|node| (&node.hash, &node.number, &node.data))
}
/// Map fork tree into values of new types.
///
/// Tree traversal technique (e.g. BFS vs DFS) is left as not specified and
/// may be subject to change in the future. In other words, your predicates
/// should not rely on the observed traversal technique currently in use.
pub fn map<VT, F>(self, f: &mut F) -> ForkTree<H, N, VT>
where
F: FnMut(&H, &N, V) -> VT,
{
let mut queue: Vec<_> =
self.roots.into_iter().rev().map(|node| (usize::MAX, node)).collect();
let mut next_queue = Vec::new();
let mut output = Vec::new();
while !queue.is_empty() {
for (parent_index, node) in queue.drain(..) {
let new_data = f(&node.hash, &node.number, node.data);
let new_node = Node {
hash: node.hash,
number: node.number,
data: new_data,
children: Vec::with_capacity(node.children.len()),
};
let node_id = output.len();
output.push((parent_index, new_node));
for child in node.children.into_iter().rev() {
next_queue.push((node_id, child));
}
}
std::mem::swap(&mut queue, &mut next_queue);
}
let mut roots = Vec::new();
while let Some((parent_index, new_node)) = output.pop() {
if parent_index == usize::MAX {
roots.push(new_node);
} else {
output[parent_index].1.children.push(new_node);
}
}
ForkTree { roots, best_finalized_number: self.best_finalized_number }
}
/// Find a node in the tree that is the deepest ancestor of the given
/// block hash and which passes the given predicate. The given function
/// `is_descendent_of` should return `true` if the second hash (target)
/// is a descendent of the first hash (base).
pub fn find_node_where<F, E, P>(
&self,
hash: &H,
number: &N,
is_descendent_of: &F,
predicate: &P,
) -> Result<Option<&Node<H, N, V>>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
let maybe_path = self.find_node_index_where(hash, number, is_descendent_of, predicate)?;
Ok(maybe_path.map(|path| {
let children =
path.iter().take(path.len() - 1).fold(&self.roots, |curr, &i| &curr[i].children);
&children[path[path.len() - 1]]
}))
}
/// Same as [`find_node_where`](ForkTree::find_node_where), but returns mutable reference.
pub fn find_node_where_mut<F, E, P>(
&mut self,
hash: &H,
number: &N,
is_descendent_of: &F,
predicate: &P,
) -> Result<Option<&mut Node<H, N, V>>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
let maybe_path = self.find_node_index_where(hash, number, is_descendent_of, predicate)?;
Ok(maybe_path.map(|path| {
let children = path
.iter()
.take(path.len() - 1)
.fold(&mut self.roots, |curr, &i| &mut curr[i].children);
&mut children[path[path.len() - 1]]
}))
}
/// Same as [`find_node_where`](ForkTree::find_node_where), but returns indices.
///
/// The returned indices represent the full path to reach the matching node starting
/// from first to last, i.e. the earliest index in the traverse path goes first, and the final
/// index in the traverse path goes last. If a node is found that matches the predicate
/// the returned path should always contain at least one index, otherwise `None` is
/// returned.
// WARNING: some users of this method (i.e. consensus epoch changes tree) currently silently
// rely on a **post-order DFS** traversal. If we are using instead a top-down traversal method
// then the `is_descendent_of` closure, when used after a warp-sync, will end up querying the
// backend for a block (the one corresponding to the root) that is not present and thus will
// return a wrong result.
pub fn find_node_index_where<F, E, P>(
&self,
hash: &H,
number: &N,
is_descendent_of: &F,
predicate: &P,
) -> Result<Option<Vec<usize>>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
let mut stack = vec![];
let mut root_idx = 0;
let mut found = false;
let mut is_descendent = false;
while root_idx < self.roots.len() {
if *number <= self.roots[root_idx].number {
root_idx += 1;
continue
}
// The second element in the stack tuple tracks what is the **next** children
// index to search into. If we find an ancestor then we stop searching into
// alternative branches and we focus on the current path up to the root.
stack.push((&self.roots[root_idx], 0));
while let Some((node, i)) = stack.pop() {
if i < node.children.len() && !is_descendent {
stack.push((node, i + 1));
if node.children[i].number < *number {
stack.push((&node.children[i], 0));
}
} else if is_descendent || is_descendent_of(&node.hash, hash)? {
is_descendent = true;
if predicate(&node.data) {
found = true;
break
}
}
}
// If the element we are looking for is a descendent of the current root
// then we can stop the search.
if is_descendent {
break
}
root_idx += 1;
}
Ok(if found {
// The path is the root index followed by the indices of all the children
// we were processing when we found the element (remember the stack
// contains the index of the **next** children to process).
let path: Vec<_> =
std::iter::once(root_idx).chain(stack.iter().map(|(_, i)| *i - 1)).collect();
Some(path)
} else {
None
})
}
/// Prune the tree, removing all non-canonical nodes. We find the node in the
/// tree that is the deepest ancestor of the given hash and that passes the
/// given predicate. If such a node exists, we re-root the tree to this
/// node. Otherwise the tree remains unchanged. The given function
/// `is_descendent_of` should return `true` if the second hash (target) is a
/// descendent of the first hash (base).
///
/// Returns all pruned node data.
pub fn prune<F, E, P>(
&mut self,
hash: &H,
number: &N,
is_descendent_of: &F,
predicate: &P,
) -> Result<impl Iterator<Item = (H, N, V)>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
let root_index =
match self.find_node_index_where(hash, number, is_descendent_of, predicate)? {
Some(idx) => idx,
None => return Ok(RemovedIterator { stack: Vec::new() }),
};
let mut old_roots = std::mem::take(&mut self.roots);
let curr_children = root_index
.iter()
.take(root_index.len() - 1)
.fold(&mut old_roots, |curr, idx| &mut curr[*idx].children);
let mut root = curr_children.remove(root_index[root_index.len() - 1]);
let mut removed = old_roots;
// we found the deepest ancestor of the finalized block, so we prune
// out any children that don't include the finalized block.
let root_children = std::mem::take(&mut root.children);
let mut is_first = true;
for child in root_children {
if is_first &&
(child.number == *number && child.hash == *hash ||
child.number < *number && is_descendent_of(&child.hash, hash)?)
{
root.children.push(child);
// assuming that the tree is well formed only one child should pass this
// requirement due to ancestry restrictions (i.e. they must be different forks).
is_first = false;
} else {
removed.push(child);
}
}
self.roots = vec![root];
self.rebalance();
Ok(RemovedIterator { stack: removed })
}
/// Finalize a root in the tree and return it, return `None` in case no root
/// with the given hash exists. All other roots are pruned, and the children
/// of the finalized node become the new roots.
pub fn finalize_root(&mut self, hash: &H) -> Option<V> {
self.roots
.iter()
.position(|node| node.hash == *hash)
.map(|position| self.finalize_root_at(position))
}
/// Finalize root at given position. See `finalize_root` comment for details.
fn finalize_root_at(&mut self, position: usize) -> V {
let node = self.roots.swap_remove(position);
self.roots = node.children;
self.best_finalized_number = Some(node.number);
node.data
}
/// Finalize a node in the tree. This method will make sure that the node
/// being finalized is either an existing root (and return its data), or a
/// node from a competing branch (not in the tree), tree pruning is done
/// accordingly. The given function `is_descendent_of` should return `true`
/// if the second hash (target) is a descendent of the first hash (base).
pub fn finalize<F, E>(
&mut self,
hash: &H,
number: N,
is_descendent_of: &F,
) -> Result<FinalizationResult<V>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
{
if let Some(ref best_finalized_number) = self.best_finalized_number {
if number <= *best_finalized_number {
return Err(Error::Revert)
}
}
// check if one of the current roots is being finalized
if let Some(root) = self.finalize_root(hash) {
return Ok(FinalizationResult::Changed(Some(root)))
}
// make sure we're not finalizing a descendent of any root
for root in self.roots.iter() {
if number > root.number && is_descendent_of(&root.hash, hash)? {
return Err(Error::UnfinalizedAncestor)
}
}
// we finalized a block earlier than any existing root (or possibly
// another fork not part of the tree). make sure to only keep roots that
// are part of the finalized branch
let mut changed = false;
let roots = std::mem::take(&mut self.roots);
for root in roots {
if root.number > number && is_descendent_of(hash, &root.hash)? {
self.roots.push(root);
} else {
changed = true;
}
}
self.best_finalized_number = Some(number);
if changed {
Ok(FinalizationResult::Changed(None))
} else {
Ok(FinalizationResult::Unchanged)
}
}
/// Finalize a node in the tree and all its ancestors. The given function
/// `is_descendent_of` should return `true` if the second hash (target) is
// a descendent of the first hash (base).
pub fn finalize_with_ancestors<F, E>(
&mut self,
hash: &H,
number: N,
is_descendent_of: &F,
) -> Result<FinalizationResult<V>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
{
if let Some(ref best_finalized_number) = self.best_finalized_number {
if number <= *best_finalized_number {
return Err(Error::Revert)
}
}
// check if one of the current roots is being finalized
if let Some(root) = self.finalize_root(hash) {
return Ok(FinalizationResult::Changed(Some(root)))
}
// we need to:
// 1) remove all roots that are not ancestors AND not descendants of finalized block;
// 2) if node is descendant - just leave it;
// 3) if node is ancestor - 'open it'
let mut changed = false;
let mut idx = 0;
while idx != self.roots.len() {
let (is_finalized, is_descendant, is_ancestor) = {
let root = &self.roots[idx];
let is_finalized = root.hash == *hash;
let is_descendant =
!is_finalized && root.number > number && is_descendent_of(hash, &root.hash)?;
let is_ancestor = !is_finalized &&
!is_descendant && root.number < number &&
is_descendent_of(&root.hash, hash)?;
(is_finalized, is_descendant, is_ancestor)
};
// if we have met finalized root - open it and return
if is_finalized {
return Ok(FinalizationResult::Changed(Some(self.finalize_root_at(idx))))
}
// if node is descendant of finalized block - just leave it as is
if is_descendant {
idx += 1;
continue
}
// if node is ancestor of finalized block - remove it and continue with children
if is_ancestor {
let root = self.roots.swap_remove(idx);
self.roots.extend(root.children);
changed = true;
continue
}
// if node is neither ancestor, nor descendant of the finalized block - remove it
self.roots.swap_remove(idx);
changed = true;
}
self.best_finalized_number = Some(number);
if changed {
Ok(FinalizationResult::Changed(None))
} else {
Ok(FinalizationResult::Unchanged)
}
}
/// Checks if any node in the tree is finalized by either finalizing the
/// node itself or a node's descendent that's not in the tree, guaranteeing
/// that the node being finalized isn't a descendent of (or equal to) any of
/// the node's children. Returns `Some(true)` if the node being finalized is
/// a root, `Some(false)` if the node being finalized is not a root, and
/// `None` if no node in the tree is finalized. The given `predicate` is
/// checked on the prospective finalized root and must pass for finalization
/// to occur. The given function `is_descendent_of` should return `true` if
/// the second hash (target) is a descendent of the first hash (base).
pub fn finalizes_any_with_descendent_if<F, P, E>(
&self,
hash: &H,
number: N,
is_descendent_of: &F,
predicate: P,
) -> Result<Option<bool>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
if let Some(ref best_finalized_number) = self.best_finalized_number {
if number <= *best_finalized_number {
return Err(Error::Revert)
}
}
// check if the given hash is equal or a descendent of any node in the
// tree, if we find a valid node that passes the predicate then we must
// ensure that we're not finalizing past any of its child nodes.
for node in self.node_iter() {
if predicate(&node.data) && (node.hash == *hash || is_descendent_of(&node.hash, hash)?)
{
for child in node.children.iter() {
if child.number <= number &&
(child.hash == *hash || is_descendent_of(&child.hash, hash)?)
{
return Err(Error::UnfinalizedAncestor)
}
}
return Ok(Some(self.roots.iter().any(|root| root.hash == node.hash)))
}
}
Ok(None)
}
/// Finalize a root in the tree by either finalizing the node itself or a
/// node's descendent that's not in the tree, guaranteeing that the node
/// being finalized isn't a descendent of (or equal to) any of the root's
/// children. The given `predicate` is checked on the prospective finalized
/// root and must pass for finalization to occur. The given function
/// `is_descendent_of` should return `true` if the second hash (target) is a
/// descendent of the first hash (base).
pub fn finalize_with_descendent_if<F, P, E>(
&mut self,
hash: &H,
number: N,
is_descendent_of: &F,
predicate: P,
) -> Result<FinalizationResult<V>, Error<E>>
where
E: std::error::Error,
F: Fn(&H, &H) -> Result<bool, E>,
P: Fn(&V) -> bool,
{
if let Some(ref best_finalized_number) = self.best_finalized_number {
if number <= *best_finalized_number {
return Err(Error::Revert)
}
}
// check if the given hash is equal or a a descendent of any root, if we
// find a valid root that passes the predicate then we must ensure that
// we're not finalizing past any children node.
let mut position = None;
for (i, root) in self.roots.iter().enumerate() {
if predicate(&root.data) && (root.hash == *hash || is_descendent_of(&root.hash, hash)?)
{
for child in root.children.iter() {
if child.number <= number &&
(child.hash == *hash || is_descendent_of(&child.hash, hash)?)
{
return Err(Error::UnfinalizedAncestor)
}
}
position = Some(i);
break
}
}
let node_data = position.map(|i| {
let node = self.roots.swap_remove(i);
self.roots = node.children;
self.best_finalized_number = Some(node.number);
node.data
});
// Retain only roots that are descendents of the finalized block (this
// happens if the node has been properly finalized) or that are
// ancestors (or equal) to the finalized block (in this case the node
// wasn't finalized earlier presumably because the predicate didn't
// pass).
let mut changed = false;
let roots = std::mem::take(&mut self.roots);
for root in roots {
let retain = root.number > number && is_descendent_of(hash, &root.hash)? ||
root.number == number && root.hash == *hash ||
is_descendent_of(&root.hash, hash)?;
if retain {
self.roots.push(root);
} else {
changed = true;
}
}
self.best_finalized_number = Some(number);
match (node_data, changed) {
(Some(data), _) => Ok(FinalizationResult::Changed(Some(data))),
(None, true) => Ok(FinalizationResult::Changed(None)),
(None, false) => Ok(FinalizationResult::Unchanged),
}
}
/// Remove from the tree some nodes (and their subtrees) using a `filter` predicate.
///
/// The `filter` is called over tree nodes and returns a filter action:
/// - `Remove` if the node and its subtree should be removed;
/// - `KeepNode` if we should maintain the node and keep processing the tree.
/// - `KeepTree` if we should maintain the node and its entire subtree.
///
/// An iterator over all the pruned nodes is returned.
pub fn drain_filter<F>(&mut self, filter: F) -> impl Iterator<Item = (H, N, V)>
where
F: Fn(&H, &N, &V) -> FilterAction,
{
let mut removed = vec![];
let mut retained = Vec::new();
let mut queue: Vec<_> = std::mem::take(&mut self.roots)
.into_iter()
.rev()
.map(|node| (usize::MAX, node))
.collect();
let mut next_queue = Vec::new();
while !queue.is_empty() {
for (parent_idx, mut node) in queue.drain(..) {
match filter(&node.hash, &node.number, &node.data) {
FilterAction::KeepNode => {
let node_idx = retained.len();
let children = std::mem::take(&mut node.children);
retained.push((parent_idx, node));
for child in children.into_iter().rev() {
next_queue.push((node_idx, child));
}
},
FilterAction::KeepTree => {
retained.push((parent_idx, node));
},
FilterAction::Remove => {
removed.push(node);
},
}
}
std::mem::swap(&mut queue, &mut next_queue);
}
while let Some((parent_idx, node)) = retained.pop() {
if parent_idx == usize::MAX {
self.roots.push(node);
} else {
retained[parent_idx].1.children.push(node);
}
}
if !removed.is_empty() {
self.rebalance();
}
RemovedIterator { stack: removed }
}
}
// Workaround for: https://github.com/rust-lang/rust/issues/34537
use node_implementation::Node;
mod node_implementation {
use super::*;
#[derive(Clone, Debug, Decode, Encode, PartialEq)]
pub struct Node<H, N, V> {
pub hash: H,
pub number: N,
pub data: V,
pub children: Vec<Node<H, N, V>>,
}
impl<H: PartialEq, N: Ord, V> Node<H, N, V> {
/// Finds the max depth among all branches descendent from this node.
pub fn max_depth(&self) -> usize {
let mut max: usize = 0;
let mut stack = vec![(self, 0)];
while let Some((node, height)) = stack.pop() {
if height > max {
max = height;
}
node.children.iter().for_each(|n| stack.push((n, height + 1)));
}
max
}
}
}
struct ForkTreeIterator<'a, H, N, V> {
stack: Vec<&'a Node<H, N, V>>,
}
impl<'a, H, N, V> Iterator for ForkTreeIterator<'a, H, N, V> {
type Item = &'a Node<H, N, V>;
fn next(&mut self) -> Option<Self::Item> {
self.stack.pop().map(|node| {
// child nodes are stored ordered by max branch height (decreasing),
// we want to keep this ordering while iterating but since we're
// using a stack for iterator state we need to reverse it.
self.stack.extend(node.children.iter().rev());
node
})
}
}
struct RemovedIterator<H, N, V> {
stack: Vec<Node<H, N, V>>,
}
impl<H, N, V> Iterator for RemovedIterator<H, N, V> {
type Item = (H, N, V);
fn next(&mut self) -> Option<Self::Item> {
self.stack.pop().map(|mut node| {
// child nodes are stored ordered by max branch height (decreasing),
// we want to keep this ordering while iterating but since we're
// using a stack for iterator state we need to reverse it.
let children = std::mem::take(&mut node.children);
self.stack.extend(children.into_iter().rev());
(node.hash, node.number, node.data)
})
}
}
#[cfg(test)]
mod test {
use crate::FilterAction;
use super::{Error, FinalizationResult, ForkTree};
#[derive(Debug, PartialEq)]
struct TestError;
impl std::fmt::Display for TestError {
fn fmt(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
write!(f, "TestError")
}
}
impl std::error::Error for TestError {}
fn test_fork_tree<'a>(
) -> (ForkTree<&'a str, u64, ()>, impl Fn(&&str, &&str) -> Result<bool, TestError>) {
let mut tree = ForkTree::new();
#[rustfmt::skip]
//
// - B - C - D - E
// /
// / - G
// / /
// A - F - H - I
// \ \
// \ - L - M - N
// \ \
// \ - O
// - J - K
//
// (where N is not a part of fork tree)
//
// NOTE: the tree will get automatically rebalance on import and won't be laid out like the
// diagram above. the children will be ordered by subtree depth and the longest branches
// will be on the leftmost side of the tree.
let is_descendent_of = |base: &&str, block: &&str| -> Result<bool, TestError> {
let letters = vec!["B", "C", "D", "E", "F", "G", "H", "I", "J", "K", "L", "M", "N", "O"];
match (*base, *block) {
("A", b) => Ok(letters.into_iter().any(|n| n == b)),
("B", b) => Ok(b == "C" || b == "D" || b == "E"),
("C", b) => Ok(b == "D" || b == "E"),
("D", b) => Ok(b == "E"),
("E", _) => Ok(false),
("F", b) =>
Ok(b == "G" || b == "H" || b == "I" || b == "L" || b == "M" || b == "N" || b == "O"),
("G", _) => Ok(false),
("H", b) => Ok(b == "I" || b == "L" || b == "M" || b == "N" || b == "O"),
("I", _) => Ok(false),
("J", b) => Ok(b == "K"),
("K", _) => Ok(false),
("L", b) => Ok(b == "M" || b == "O" || b == "N"),
("M", b) => Ok(b == "N"),
("O", _) => Ok(false),
("0", _) => Ok(true),
_ => Ok(false),
}
};
tree.import("A", 1, (), &is_descendent_of).unwrap();
tree.import("B", 2, (), &is_descendent_of).unwrap();
tree.import("C", 3, (), &is_descendent_of).unwrap();
tree.import("D", 4, (), &is_descendent_of).unwrap();
tree.import("E", 5, (), &is_descendent_of).unwrap();
tree.import("F", 2, (), &is_descendent_of).unwrap();
tree.import("G", 3, (), &is_descendent_of).unwrap();
tree.import("H", 3, (), &is_descendent_of).unwrap();
tree.import("I", 4, (), &is_descendent_of).unwrap();
tree.import("L", 4, (), &is_descendent_of).unwrap();
tree.import("M", 5, (), &is_descendent_of).unwrap();
tree.import("O", 5, (), &is_descendent_of).unwrap();
tree.import("J", 2, (), &is_descendent_of).unwrap();
tree.import("K", 3, (), &is_descendent_of).unwrap();
(tree, is_descendent_of)
}
#[test]
fn import_doesnt_revert() {
let (mut tree, is_descendent_of) = test_fork_tree();
tree.finalize_root(&"A");
assert_eq!(tree.best_finalized_number, Some(1));
assert_eq!(tree.import("A", 1, (), &is_descendent_of), Err(Error::Revert));
}
#[test]
fn import_doesnt_add_duplicates() {
let (mut tree, is_descendent_of) = test_fork_tree();
assert_eq!(tree.import("A", 1, (), &is_descendent_of), Err(Error::Duplicate));
assert_eq!(tree.import("I", 4, (), &is_descendent_of), Err(Error::Duplicate));
assert_eq!(tree.import("G", 3, (), &is_descendent_of), Err(Error::Duplicate));
assert_eq!(tree.import("K", 3, (), &is_descendent_of), Err(Error::Duplicate));
}
#[test]
fn finalize_root_works() {
let finalize_a = || {
let (mut tree, ..) = test_fork_tree();
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![("A", 1)]);
// finalizing "A" opens up three possible forks
tree.finalize_root(&"A");
assert_eq!(
tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![("B", 2), ("F", 2), ("J", 2)],
);
tree
};
{
let mut tree = finalize_a();
// finalizing "B" will progress on its fork and remove any other competing forks
tree.finalize_root(&"B");
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![("C", 3)],);
// all the other forks have been pruned
assert!(tree.roots.len() == 1);
}
{
let mut tree = finalize_a();
// finalizing "J" will progress on its fork and remove any other competing forks
tree.finalize_root(&"J");
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![("K", 3)],);
// all the other forks have been pruned
assert!(tree.roots.len() == 1);
}
}
#[test]
fn finalize_works() {
let (mut tree, is_descendent_of) = test_fork_tree();
let original_roots = tree.roots.clone();
// finalizing a block prior to any in the node doesn't change the tree
assert_eq!(tree.finalize(&"0", 0, &is_descendent_of), Ok(FinalizationResult::Unchanged));
assert_eq!(tree.roots, original_roots);
// finalizing "A" opens up three possible forks
assert_eq!(
tree.finalize(&"A", 1, &is_descendent_of),
Ok(FinalizationResult::Changed(Some(()))),
);
assert_eq!(
tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![("B", 2), ("F", 2), ("J", 2)],
);
// finalizing anything lower than what we observed will fail
assert_eq!(tree.best_finalized_number, Some(1));
assert_eq!(tree.finalize(&"Z", 1, &is_descendent_of), Err(Error::Revert));
// trying to finalize a node without finalizing its ancestors first will fail
assert_eq!(tree.finalize(&"H", 3, &is_descendent_of), Err(Error::UnfinalizedAncestor));
// after finalizing "F" we can finalize "H"
assert_eq!(
tree.finalize(&"F", 2, &is_descendent_of),
Ok(FinalizationResult::Changed(Some(()))),
);
assert_eq!(
tree.finalize(&"H", 3, &is_descendent_of),
Ok(FinalizationResult::Changed(Some(()))),
);
assert_eq!(
tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![("L", 4), ("I", 4)],
);
// finalizing a node from another fork that isn't part of the tree clears the tree
assert_eq!(
tree.finalize(&"Z", 5, &is_descendent_of),
Ok(FinalizationResult::Changed(None)),
);
assert!(tree.roots.is_empty());
}
#[test]
fn finalize_with_ancestor_works() {
let (mut tree, is_descendent_of) = test_fork_tree();
let original_roots = tree.roots.clone();
// finalizing a block prior to any in the node doesn't change the tree
assert_eq!(
tree.finalize_with_ancestors(&"0", 0, &is_descendent_of),
Ok(FinalizationResult::Unchanged),
);
assert_eq!(tree.roots, original_roots);
// finalizing "A" opens up three possible forks
assert_eq!(
tree.finalize_with_ancestors(&"A", 1, &is_descendent_of),
Ok(FinalizationResult::Changed(Some(()))),
);
assert_eq!(
tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![("B", 2), ("F", 2), ("J", 2)],
);
// finalizing H:
// 1) removes roots that are not ancestors/descendants of H (B, J)
// 2) opens root that is ancestor of H (F -> G+H)
// 3) finalizes the just opened root H (H -> I + L)
assert_eq!(
tree.finalize_with_ancestors(&"H", 3, &is_descendent_of),
Ok(FinalizationResult::Changed(Some(()))),
);
assert_eq!(
tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![("L", 4), ("I", 4)],
);
assert_eq!(tree.best_finalized_number, Some(3));
// finalizing N (which is not a part of the tree):
// 1) removes roots that are not ancestors/descendants of N (I)
// 2) opens root that is ancestor of N (L -> M+O)
// 3) removes roots that are not ancestors/descendants of N (O)
// 4) opens root that is ancestor of N (M -> {})
assert_eq!(
tree.finalize_with_ancestors(&"N", 6, &is_descendent_of),
Ok(FinalizationResult::Changed(None)),
);
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![],);
assert_eq!(tree.best_finalized_number, Some(6));
}
#[test]
fn finalize_with_descendent_works() {
#[derive(Debug, PartialEq)]
struct Change {
effective: u64,
}
let (mut tree, is_descendent_of) = {
let mut tree = ForkTree::new();
let is_descendent_of = |base: &&str, block: &&str| -> Result<bool, TestError> {
// A0 #1 - (B #2) - (C #5) - D #10 - E #15 - (F #100)
// \
// - (G #100)
//
// A1 #1
//
// Nodes B, C, F and G are not part of the tree.
match (*base, *block) {
("A0", b) => Ok(b == "B" || b == "C" || b == "D" || b == "E" || b == "G"),
("A1", _) => Ok(false),
("C", b) => Ok(b == "D"),
("D", b) => Ok(b == "E" || b == "F" || b == "G"),
("E", b) => Ok(b == "F"),
_ => Ok(false),
}
};
let is_root = tree.import("A0", 1, Change { effective: 5 }, &is_descendent_of).unwrap();
assert!(is_root);
let is_root = tree.import("A1", 1, Change { effective: 5 }, &is_descendent_of).unwrap();
assert!(is_root);
let is_root =
tree.import("D", 10, Change { effective: 10 }, &is_descendent_of).unwrap();
assert!(!is_root);
let is_root =
tree.import("E", 15, Change { effective: 50 }, &is_descendent_of).unwrap();
assert!(!is_root);
(tree, is_descendent_of)
};
assert_eq!(
tree.finalizes_any_with_descendent_if(
&"B",
2,
&is_descendent_of,
|c| c.effective <= 2,
),
Ok(None),
);
// finalizing "D" is not allowed since it is not a root.
assert_eq!(
tree.finalize_with_descendent_if(&"D", 10, &is_descendent_of, |c| c.effective <= 10),
Err(Error::UnfinalizedAncestor)
);
// finalizing "D" will finalize a block from the tree, but it can't be applied yet
// since it is not a root change.
assert_eq!(
tree.finalizes_any_with_descendent_if(&"D", 10, &is_descendent_of, |c| c.effective ==
10),
Ok(Some(false)),
);
// finalizing "E" is not allowed since there are not finalized anchestors.
assert_eq!(
tree.finalizes_any_with_descendent_if(&"E", 15, &is_descendent_of, |c| c.effective ==
10),
Err(Error::UnfinalizedAncestor)
);
// finalizing "B" doesn't finalize "A0" since the predicate doesn't pass,
// although it will clear out "A1" from the tree
assert_eq!(
tree.finalize_with_descendent_if(&"B", 2, &is_descendent_of, |c| c.effective <= 2),
Ok(FinalizationResult::Changed(None)),
);
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![("A0", 1)],);
// finalizing "C" will finalize the node "A0" and prune it out of the tree
assert_eq!(
tree.finalizes_any_with_descendent_if(
&"C",
5,
&is_descendent_of,
|c| c.effective <= 5,
),
Ok(Some(true)),
);
assert_eq!(
tree.finalize_with_descendent_if(&"C", 5, &is_descendent_of, |c| c.effective <= 5),
Ok(FinalizationResult::Changed(Some(Change { effective: 5 }))),
);
assert_eq!(tree.roots().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(), vec![("D", 10)],);
// finalizing "F" will fail since it would finalize past "E" without finalizing "D" first
assert_eq!(
tree.finalizes_any_with_descendent_if(&"F", 100, &is_descendent_of, |c| c.effective <=
100,),
Err(Error::UnfinalizedAncestor),
);
// it will work with "G" though since it is not in the same branch as "E"
assert_eq!(
tree.finalizes_any_with_descendent_if(&"G", 100, &is_descendent_of, |c| c.effective <=
100),
Ok(Some(true)),
);
assert_eq!(
tree.finalize_with_descendent_if(&"G", 100, &is_descendent_of, |c| c.effective <= 100),
Ok(FinalizationResult::Changed(Some(Change { effective: 10 }))),
);
// "E" will be pruned out
assert_eq!(tree.roots().count(), 0);
}
#[test]
fn iter_iterates_in_preorder() {
let (tree, ..) = test_fork_tree();
assert_eq!(
tree.iter().map(|(h, n, _)| (*h, *n)).collect::<Vec<_>>(),
vec![
("A", 1),
("B", 2),
("C", 3),
("D", 4),
("E", 5),
("F", 2),
("H", 3),
("L", 4),
("M", 5),
("O", 5),
("I", 4),
("G", 3),
("J", 2),
("K", 3),
],
);
}
#[test]
fn minimizes_calls_to_is_descendent_of() {
use std::sync::atomic::{AtomicUsize, Ordering};
let n_is_descendent_of_calls = AtomicUsize::new(0);
let is_descendent_of = |_: &&str, _: &&str| -> Result<bool, TestError> {
n_is_descendent_of_calls.fetch_add(1, Ordering::SeqCst);
Ok(true)
};
{
// Deep tree where we want to call `finalizes_any_with_descendent_if`. The
// search for the node should first check the predicate (which is cheaper) and
// only then call `is_descendent_of`
let mut tree = ForkTree::new();
let letters = vec!["A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K"];
for (i, letter) in letters.iter().enumerate() {
tree.import::<_, TestError>(*letter, i, i, &|_, _| Ok(true)).unwrap();
}
// "L" is a descendent of "K", but the predicate will only pass for "K",
// therefore only one call to `is_descendent_of` should be made
assert_eq!(
tree.finalizes_any_with_descendent_if(&"L", 11, &is_descendent_of, |i| *i == 10,),
Ok(Some(false)),
);
assert_eq!(n_is_descendent_of_calls.load(Ordering::SeqCst), 1);
}
n_is_descendent_of_calls.store(0, Ordering::SeqCst);
{
// Multiple roots in the tree where we want to call `finalize_with_descendent_if`.
// The search for the root node should first check the predicate (which is cheaper)
// and only then call `is_descendent_of`
let mut tree = ForkTree::new();
let letters = vec!["A", "B", "C", "D", "E", "F", "G", "H", "I", "J", "K"];
for (i, letter) in letters.iter().enumerate() {
tree.import::<_, TestError>(*letter, i, i, &|_, _| Ok(false)).unwrap();
}
// "L" is a descendent of "K", but the predicate will only pass for "K",
// therefore only one call to `is_descendent_of` should be made
assert_eq!(
tree.finalize_with_descendent_if(&"L", 11, &is_descendent_of, |i| *i == 10,),
Ok(FinalizationResult::Changed(Some(10))),
);
assert_eq!(n_is_descendent_of_calls.load(Ordering::SeqCst), 1);
}
}
#[test]
fn map_works() {
let (mut tree, _) = test_fork_tree();
// Extend the single root fork-tree to also excercise the roots order during map.
let is_descendent_of = |_: &&str, _: &&str| -> Result<bool, TestError> { Ok(false) };
let is_root = tree.import("A1", 1, (), &is_descendent_of).unwrap();
assert!(is_root);
let is_root = tree.import("A2", 1, (), &is_descendent_of).unwrap();
assert!(is_root);
let old_tree = tree.clone();
let new_tree = tree.map(&mut |hash, _, _| hash.to_owned());
// Check content and order
assert!(new_tree.iter().all(|(hash, _, data)| hash == data));
assert_eq!(
old_tree.iter().map(|(hash, _, _)| *hash).collect::<Vec<_>>(),
new_tree.iter().map(|(hash, _, _)| *hash).collect::<Vec<_>>(),
);
}
#[test]
fn prune_works() {
let (mut tree, is_descendent_of) = test_fork_tree();
let removed = tree.prune(&"C", &3, &is_descendent_of, &|_| true).unwrap();
assert_eq!(tree.roots.iter().map(|node| node.hash).collect::<Vec<_>>(), vec!["B"]);
assert_eq!(
tree.iter().map(|(hash, _, _)| *hash).collect::<Vec<_>>(),
vec!["B", "C", "D", "E"],
);
assert_eq!(
removed.map(|(hash, _, _)| hash).collect::<Vec<_>>(),
vec!["A", "F", "H", "L", "M", "O", "I", "G", "J", "K"]
);
let removed = tree.prune(&"E", &5, &is_descendent_of, &|_| true).unwrap();
assert_eq!(tree.roots.iter().map(|node| node.hash).collect::<Vec<_>>(), vec!["D"]);
assert_eq!(tree.iter().map(|(hash, _, _)| *hash).collect::<Vec<_>>(), vec!["D", "E"]);
assert_eq!(removed.map(|(hash, _, _)| hash).collect::<Vec<_>>(), vec!["B", "C"]);
}
#[test]
fn find_node_backtracks_after_finding_highest_descending_node() {
let mut tree = ForkTree::new();
// A - B
// \
// — C
//
let is_descendent_of = |base: &&str, block: &&str| -> Result<bool, TestError> {
match (*base, *block) {
("A", b) => Ok(b == "B" || b == "C" || b == "D"),
("B", b) | ("C", b) => Ok(b == "D"),
("0", _) => Ok(true),
_ => Ok(false),
}
};
tree.import("A", 1, 1, &is_descendent_of).unwrap();
tree.import("B", 2, 2, &is_descendent_of).unwrap();
tree.import("C", 2, 4, &is_descendent_of).unwrap();
// when searching the tree we reach node `C`, but the
// predicate doesn't pass. we should backtrack to `B`, but not to `A`,
// since "B" fulfills the predicate.
let node = tree.find_node_where(&"D", &3, &is_descendent_of, &|data| *data < 3).unwrap();
assert_eq!(node.unwrap().hash, "B");
}
#[test]
fn rebalance_works() {
let (mut tree, _) = test_fork_tree();
// the tree is automatically rebalanced on import, therefore we should iterate in preorder
// exploring the longest forks first. check the ascii art above to understand the expected
// output below.
assert_eq!(
tree.iter().map(|(h, _, _)| *h).collect::<Vec<_>>(),
vec!["A", "B", "C", "D", "E", "F", "H", "L", "M", "O", "I", "G", "J", "K"],
);
// let's add a block "P" which is a descendent of block "O"
let is_descendent_of = |base: &&str, block: &&str| -> Result<bool, TestError> {
match (*base, *block) {
(b, "P") => Ok(vec!["A", "F", "H", "L", "O"].into_iter().any(|n| n == b)),
_ => Ok(false),
}
};
tree.import("P", 6, (), &is_descendent_of).unwrap();
// this should re-order the tree, since the branch "A -> B -> C -> D -> E" is no longer tied
// with 5 blocks depth. additionally "O" should be visited before "M" now, since it has one
// descendent "P" which makes that branch 6 blocks long.
assert_eq!(
tree.iter().map(|(h, _, _)| *h).collect::<Vec<_>>(),
["A", "F", "H", "L", "O", "P", "M", "I", "G", "B", "C", "D", "E", "J", "K"]
);
}
#[test]
fn drain_filter_works() {
let (mut tree, _) = test_fork_tree();
let filter = |h: &&str, _: &u64, _: &()| match *h {
"A" | "B" | "F" | "G" => FilterAction::KeepNode,
"C" => FilterAction::KeepTree,
"H" | "J" => FilterAction::Remove,
_ => panic!("Unexpected filtering for node: {}", *h),
};
let removed = tree.drain_filter(filter);
assert_eq!(
tree.iter().map(|(h, _, _)| *h).collect::<Vec<_>>(),
["A", "B", "C", "D", "E", "F", "G"]
);
assert_eq!(
removed.map(|(h, _, _)| h).collect::<Vec<_>>(),
["H", "L", "M", "O", "I", "J", "K"]
);
}
#[test]
fn find_node_index_works() {
let (tree, is_descendent_of) = test_fork_tree();
let path = tree
.find_node_index_where(&"D", &4, &is_descendent_of, &|_| true)
.unwrap()
.unwrap();
assert_eq!(path, [0, 0, 0]);
let path = tree
.find_node_index_where(&"O", &5, &is_descendent_of, &|_| true)
.unwrap()
.unwrap();
assert_eq!(path, [0, 1, 0, 0]);
let path = tree
.find_node_index_where(&"N", &6, &is_descendent_of, &|_| true)
.unwrap()
.unwrap();
assert_eq!(path, [0, 1, 0, 0, 0]);
}
#[test]
fn find_node_index_with_predicate_works() {
let is_descendent_of = |parent: &char, child: &char| match *parent {
'A' => Ok(['B', 'C', 'D', 'E', 'F'].contains(child)),
'B' => Ok(['C', 'D'].contains(child)),
'C' => Ok(['D'].contains(child)),
'E' => Ok(['F'].contains(child)),
'D' | 'F' => Ok(false),
_ => Err(TestError),
};
// A(t) --- B(f) --- C(t) --- D(f)
// \-- E(t) --- F(f)
let mut tree: ForkTree<char, u8, bool> = ForkTree::new();
tree.import('A', 1, true, &is_descendent_of).unwrap();
tree.import('B', 2, false, &is_descendent_of).unwrap();
tree.import('C', 3, true, &is_descendent_of).unwrap();
tree.import('D', 4, false, &is_descendent_of).unwrap();
tree.import('E', 2, true, &is_descendent_of).unwrap();
tree.import('F', 3, false, &is_descendent_of).unwrap();
let path = tree
.find_node_index_where(&'D', &4, &is_descendent_of, &|&value| !value)
.unwrap()
.unwrap();
assert_eq!(path, [0, 0]);
let path = tree
.find_node_index_where(&'D', &4, &is_descendent_of, &|&value| value)
.unwrap()
.unwrap();
assert_eq!(path, [0, 0, 0]);
let path = tree
.find_node_index_where(&'F', &3, &is_descendent_of, &|&value| !value)
.unwrap();
assert_eq!(path, None);
let path = tree
.find_node_index_where(&'F', &3, &is_descendent_of, &|&value| value)
.unwrap()
.unwrap();
assert_eq!(path, [0, 1]);
}
#[test]
fn find_node_works() {
let (tree, is_descendent_of) = test_fork_tree();
let node = tree.find_node_where(&"B", &2, &is_descendent_of, &|_| true).unwrap().unwrap();
assert_eq!((node.hash, node.number), ("A", 1));
let node = tree.find_node_where(&"D", &4, &is_descendent_of, &|_| true).unwrap().unwrap();
assert_eq!((node.hash, node.number), ("C", 3));
let node = tree.find_node_where(&"O", &5, &is_descendent_of, &|_| true).unwrap().unwrap();
assert_eq!((node.hash, node.number), ("L", 4));
let node = tree.find_node_where(&"N", &6, &is_descendent_of, &|_| true).unwrap().unwrap();
assert_eq!((node.hash, node.number), ("M", 5));
}
#[test]
fn post_order_traversal_requirement() {
let (mut tree, is_descendent_of) = test_fork_tree();
// Test for the post-order DFS traversal requirement as specified by the
// `find_node_index_where` and `import` comments.
let is_descendent_of_for_post_order = |parent: &&str, child: &&str| match *parent {
"A" => Err(TestError),
"K" if *child == "Z" => Ok(true),
_ => is_descendent_of(parent, child),
};
// Post order traversal requirement for `find_node_index_where`
let path = tree
.find_node_index_where(&"N", &6, &is_descendent_of_for_post_order, &|_| true)
.unwrap()
.unwrap();
assert_eq!(path, [0, 1, 0, 0, 0]);
// Post order traversal requirement for `import`
let res = tree.import(&"Z", 100, (), &is_descendent_of_for_post_order);
assert_eq!(res, Ok(false));
assert_eq!(
tree.iter().map(|node| *node.0).collect::<Vec<_>>(),
vec!["A", "B", "C", "D", "E", "F", "H", "L", "M", "O", "I", "G", "J", "K", "Z"],
);
}
}