[llvm] [NFC][Support] Add RadixTree (PR #164524)

[Florian Mayer via llvm-commits]

================
@@ -0,0 +1,345 @@
+//===-- RadixTree.h - Radix Tree implementation -----------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//===----------------------------------------------------------------------===//
+//
+// This file implements a Radix Tree.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_SUPPORT_RADIXTREE_H
+#define LLVM_SUPPORT_RADIXTREE_H
+
+#include "llvm/ADT/ADL.h"
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/iterator.h"
+#include "llvm/ADT/iterator_range.h"
+#include <cassert>
+#include <cstddef>
+#include <iterator>
+#include <limits>
+#include <list>
+#include <utility>
+
+namespace llvm {
+
+/// \brief A Radix Tree implementation.
+///
+/// A Radix Tree (also known as a compact prefix tree or radix trie) is a
+/// data structure that stores a dynamic set or associative array where keys
+/// are strings and values are associated with these keys. Unlike a regular
+/// trie, the edges of a radix tree can be labeled with sequences of characters
+/// as well as single characters. This makes radix trees more efficient for
+/// storing sparse data sets, where many nodes in a regular trie would have
+/// only one child.
+///
+/// This implementation supports arbitrary key types that can be iterated over
+/// (e.g., `std::string`, `std::vector<char>`, `ArrayRef<char>`). The key type
+/// must provide `begin()` and `end()` for iteration.
+///
+/// The tree stores `std::pair<const KeyType, T>` as its value type.
+///
+/// Example usage:
+/// \code
+///   llvm::RadixTree<StringRef, int> Tree;
+///   Tree.emplace("apple", 1);
+///   Tree.emplace("grapefruit", 2);
+///   Tree.emplace("grape", 3);
+///
+///   // Find prefixes
+///   for (const auto &pair : Tree.find_prefixes("grapefruit juice")) {
+///     // pair will be {"grape", 3}
+///     // pair will be {"grapefruit", 2}
+///     llvm::outs() << pair.first << ": " << pair.second << "\n";
+///   }
+///
+///   // Iterate over all elements
+///   for (const auto &pair : Tree) {
+///     llvm::outs() << pair.first << ": " << pair.second << "\n";
+///   }
+/// \endcode
+///
+/// \note
+/// The `RadixTree` takes ownership of the `KeyType` and `T` objects
+/// inserted into it. When an element is removed or the tree is destroyed,
+/// these objects will be destructed.
+/// However, if `KeyType` is a reference-like type, e.g. StringRef or range,
+/// User must guarantee that destination has lifetime longer than the tree.
+template <typename KeyType, typename T> class RadixTree {
+public:
+  using key_type = KeyType;
+  using mapped_type = T;
+  using value_type = std::pair<const KeyType, mapped_type>;
+
+private:
+  using KeyConstIteratorType =
+      decltype(adl_begin(std::declval<const key_type &>()));
+  using KeyConstIteratorRangeType = iterator_range<KeyConstIteratorType>;
+  using KeyValueType =
+      remove_cvref_t<decltype(*adl_begin(std::declval<key_type &>()))>;
+  using ContainerType = std::list<value_type>;
+
+  /// Represents an internal node in the Radix Tree.
+  struct Node {
+    KeyConstIteratorRangeType Key = {KeyConstIteratorType{},
+                                     KeyConstIteratorType{}};
+    std::vector<Node> Children;
+
+    /// An iterator to the value associated with this node.
+    ///
+    /// If this node does not have a value (i.e., it's an internal node that
+    /// only serves as a path to other values), this iterator will be equal
+    /// to default constructed `ContainerType::iterator()`.
+    typename ContainerType::iterator Value;
+
+    /// The first character of the Key. Used for fast child lookup.
+    KeyValueType KeyFront;
+
+    Node() = default;
+    Node(const KeyConstIteratorRangeType &Key)
+        : Key(Key), KeyFront(*Key.begin()) {
+      assert(!Key.empty());
+    }
+
+    Node(Node &&) = default;
+    Node &operator=(Node &&) = default;
+
+    Node(const Node &) = delete;
+    Node &operator=(const Node &) = delete;
+
+    const Node *findChild(const KeyConstIteratorRangeType &Key) const {
+      if (Key.empty())
+        return nullptr;
+      for (const auto &Child : Children) {
+        assert(!Child.Key.empty()); // Only root can be empty.
+        if (Child.KeyFront == *Key.begin())
+          return &Child;
+      }
+      return nullptr;
+    }
+
+    Node *findChild(const KeyConstIteratorRangeType &Query) {
+      const Node *This = this;
+      return const_cast<Node *>(This->findChild(Query));
+    }
+
+    size_t countNodes() const {
+      size_t R = 1;
+      for (const auto &C : Children)
+        R += C.countNodes();
+      return R;
+    }
+
+    ///
+    /// Splits the current node into two.
+    ///
+    /// This function is used when a new key needs to be inserted that shares
+    /// a common prefix with the current node's key, but then diverges.
+    /// The current `Key` is truncated to the common prefix, and a new child
+    /// node is created for the remainder of the original node's `Key`.
+    ///
+    /// \param SplitPoint An iterator pointing to the character in the current
+    ///                   `Key` where the split should occur.
+    void split(KeyConstIteratorType SplitPoint) {
+      Node Child(make_range(SplitPoint, Key.end()));
+      Key = make_range(Key.begin(), SplitPoint);
+
+      Children.swap(Child.Children);
+      std::swap(Value, Child.Value);
+
+      Children.emplace_back(std::move(Child));
+    }
+  };
+
+  Node Root; // Root is always for empty range.
----------------
fmayer wrote:

I am not sure what "Root is always for empty range." is trying to tell me

https://github.com/llvm/llvm-project/pull/164524