@nxtedition/rocksdb 15.1.5 → 15.2.1

package/deps/re2/re2/prefilter.h

@@ -0,0 +1,168 @@
+// Copyright 2009 The RE2 Authors.  All Rights Reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#ifndef RE2_PREFILTER_H_
+#define RE2_PREFILTER_H_
+
+// Prefilter is the class used to extract string guards from regexps.
+// Rather than using Prefilter class directly, use FilteredRE2.
+// See filtered_re2.h
+
+#include <set>
+#include <string>
+#include <vector>
+
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
+
+namespace re2 {
+
+class RE2;
+
+class Regexp;
+
+class Prefilter {
+  // Instead of using Prefilter directly, use FilteredRE2; see filtered_re2.h
+ public:
+  enum Op {
+    ALL = 0,  // Everything matches
+    NONE,  // Nothing matches
+    ATOM,  // The string atom() must match
+    AND,   // All in subs() must match
+    OR,   // One of subs() must match
+  };
+
+  explicit Prefilter(Op op);
+  ~Prefilter();
+
+  Op op() { return op_; }
+  const std::string& atom() const { return atom_; }
+  void set_unique_id(int id) { unique_id_ = id; }
+  int unique_id() const { return unique_id_; }
+
+  // The children of the Prefilter node.
+  std::vector<Prefilter*>* subs() {
+    ABSL_DCHECK(op_ == AND || op_ == OR);
+    return subs_;
+  }
+
+  // Set the children vector. Prefilter takes ownership of subs and
+  // subs_ will be deleted when Prefilter is deleted.
+  void set_subs(std::vector<Prefilter*>* subs) { subs_ = subs; }
+
+  // Given a RE2, return a Prefilter. The caller takes ownership of
+  // the Prefilter and should deallocate it. Returns NULL if Prefilter
+  // cannot be formed.
+  static Prefilter* FromRE2(const RE2* re2);
+
+  // Returns a readable debug string of the prefilter.
+  std::string DebugString() const;
+
+ private:
+  template <typename H>
+  friend H AbslHashValue(H h, const Prefilter& a) {
+    h = H::combine(std::move(h), a.op_);
+    if (a.op_ == ATOM) {
+      h = H::combine(std::move(h), a.atom_);
+    } else if (a.op_ == AND || a.op_ == OR) {
+      h = H::combine(std::move(h), a.subs_->size());
+      for (size_t i = 0; i < a.subs_->size(); ++i) {
+        h = H::combine(std::move(h), (*a.subs_)[i]->unique_id_);
+      }
+    }
+    return h;
+  }
+
+  friend bool operator==(const Prefilter& a, const Prefilter& b) {
+    if (&a == &b) {
+      return true;
+    }
+    if (a.op_ != b.op_) {
+      return false;
+    }
+    if (a.op_ == ATOM) {
+      if (a.atom_ != b.atom_) {
+        return false;
+      }
+    } else if (a.op_ == AND || a.op_ == OR) {
+      if (a.subs_->size() != b.subs_->size()) {
+        return false;
+      }
+      for (size_t i = 0; i < a.subs_->size(); ++i) {
+        if ((*a.subs_)[i]->unique_id_ != (*b.subs_)[i]->unique_id_) {
+          return false;
+        }
+      }
+    }
+    return true;
+  }
+
+  // A comparator used to store exact strings. We compare by length,
+  // then lexicographically. This ordering makes it easier to reduce the
+  // set of strings in SimplifyStringSet.
+  struct LengthThenLex {
+    bool operator()(const std::string& a, const std::string& b) const {
+       return (a.size() < b.size()) || (a.size() == b.size() && a < b);
+    }
+  };
+
+  class Info;
+
+  using SSet = std::set<std::string, LengthThenLex>;
+  using SSIter = SSet::iterator;
+  using ConstSSIter = SSet::const_iterator;
+
+  // Combines two prefilters together to create an AND. The passed
+  // Prefilters will be part of the returned Prefilter or deleted.
+  static Prefilter* And(Prefilter* a, Prefilter* b);
+
+  // Combines two prefilters together to create an OR. The passed
+  // Prefilters will be part of the returned Prefilter or deleted.
+  static Prefilter* Or(Prefilter* a, Prefilter* b);
+
+  // Generalized And/Or
+  static Prefilter* AndOr(Op op, Prefilter* a, Prefilter* b);
+
+  static Prefilter* FromRegexp(Regexp* a);
+
+  static Prefilter* FromString(const std::string& str);
+
+  static Prefilter* OrStrings(SSet* ss);
+
+  static Info* BuildInfo(Regexp* re);
+
+  Prefilter* Simplify();
+
+  // Removes redundant strings from the set. A string is redundant if
+  // any of the other strings appear as a substring. The empty string
+  // is a special case, which is ignored.
+  static void SimplifyStringSet(SSet* ss);
+
+  // Adds the cross-product of a and b to dst.
+  // (For each string i in a and j in b, add i+j.)
+  static void CrossProduct(const SSet& a, const SSet& b, SSet* dst);
+
+  // Kind of Prefilter.
+  Op op_;
+
+  // Sub-matches for AND or OR Prefilter.
+  std::vector<Prefilter*>* subs_;
+
+  // Actual string to match in leaf node.
+  std::string atom_;
+
+  // If different prefilters have the same string atom, or if they are
+  // structurally the same (e.g., OR of same atom strings) they are
+  // considered the same unique nodes. This is the id for each unique
+  // node. This field is populated with a unique id for every node,
+  // and -1 for duplicate nodes.
+  int unique_id_;
+
+  Prefilter(const Prefilter&) = delete;
+  Prefilter& operator=(const Prefilter&) = delete;
+};
+
+}  // namespace re2
+
+#endif  // RE2_PREFILTER_H_

package/deps/re2/re2/prefilter_tree.cc

@@ -0,0 +1,376 @@
+// Copyright 2009 The RE2 Authors.  All Rights Reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "re2/prefilter_tree.h"
+
+#include <stddef.h>
+
+#include <algorithm>
+#include <cmath>
+#include <string>
+#include <utility>
+#include <vector>
+
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
+#include "absl/strings/str_format.h"
+#include "re2/prefilter.h"
+
+namespace re2 {
+
+static const bool ExtraDebug = false;
+
+PrefilterTree::PrefilterTree()
+    : compiled_(false),
+      min_atom_len_(3) {
+}
+
+PrefilterTree::PrefilterTree(int min_atom_len)
+    : compiled_(false),
+      min_atom_len_(min_atom_len) {
+}
+
+PrefilterTree::~PrefilterTree() {
+  for (size_t i = 0; i < prefilter_vec_.size(); i++)
+    delete prefilter_vec_[i];
+}
+
+void PrefilterTree::Add(Prefilter* prefilter) {
+  if (compiled_) {
+    ABSL_LOG(DFATAL) << "Add called after Compile.";
+    return;
+  }
+  if (prefilter != NULL && !KeepNode(prefilter)) {
+    delete prefilter;
+    prefilter = NULL;
+  }
+
+  prefilter_vec_.push_back(prefilter);
+}
+
+void PrefilterTree::Compile(std::vector<std::string>* atom_vec) {
+  if (compiled_) {
+    ABSL_LOG(DFATAL) << "Compile called already.";
+    return;
+  }
+
+  // Some legacy users of PrefilterTree call Compile() before
+  // adding any regexps and expect Compile() to have no effect.
+  if (prefilter_vec_.empty()) {
+    return;
+  }
+
+  compiled_ = true;
+
+  NodeSet nodes;
+  AssignUniqueIds(&nodes, atom_vec);
+  if (ExtraDebug)
+    PrintDebugInfo(&nodes);
+}
+
+Prefilter* PrefilterTree::CanonicalNode(NodeSet* nodes, Prefilter* node) {
+  NodeSet::const_iterator iter = nodes->find(node);
+  if (iter != nodes->end()) {
+    return *iter;
+  }
+  return NULL;
+}
+
+bool PrefilterTree::KeepNode(Prefilter* node) const {
+  if (node == NULL)
+    return false;
+
+  switch (node->op()) {
+    default:
+      ABSL_LOG(DFATAL) << "Unexpected op in KeepNode: " << node->op();
+      return false;
+
+    case Prefilter::ALL:
+    case Prefilter::NONE:
+      return false;
+
+    case Prefilter::ATOM:
+      return node->atom().size() >= static_cast<size_t>(min_atom_len_);
+
+    case Prefilter::AND: {
+      int j = 0;
+      std::vector<Prefilter*>* subs = node->subs();
+      for (size_t i = 0; i < subs->size(); i++)
+        if (KeepNode((*subs)[i]))
+          (*subs)[j++] = (*subs)[i];
+        else
+          delete (*subs)[i];
+
+      subs->resize(j);
+      return j > 0;
+    }
+
+    case Prefilter::OR:
+      for (size_t i = 0; i < node->subs()->size(); i++)
+        if (!KeepNode((*node->subs())[i]))
+          return false;
+      return true;
+  }
+}
+
+void PrefilterTree::AssignUniqueIds(NodeSet* nodes,
+                                    std::vector<std::string>* atom_vec) {
+  atom_vec->clear();
+
+  // Build vector of all filter nodes, sorted topologically
+  // from top to bottom in v.
+  std::vector<Prefilter*> v;
+
+  // Add the top level nodes of each regexp prefilter.
+  for (size_t i = 0; i < prefilter_vec_.size(); i++) {
+    Prefilter* f = prefilter_vec_[i];
+    if (f == NULL)
+      unfiltered_.push_back(static_cast<int>(i));
+
+    // We push NULL also on to v, so that we maintain the
+    // mapping of index==regexpid for level=0 prefilter nodes.
+    v.push_back(f);
+  }
+
+  // Now add all the descendant nodes.
+  for (size_t i = 0; i < v.size(); i++) {
+    Prefilter* f = v[i];
+    if (f == NULL)
+      continue;
+    if (f->op() == Prefilter::AND || f->op() == Prefilter::OR) {
+      const std::vector<Prefilter*>& subs = *f->subs();
+      for (size_t j = 0; j < subs.size(); j++)
+        v.push_back(subs[j]);
+    }
+  }
+
+  // Identify unique nodes.
+  int unique_id = 0;
+  for (int i = static_cast<int>(v.size()) - 1; i >= 0; i--) {
+    Prefilter *node = v[i];
+    if (node == NULL)
+      continue;
+    node->set_unique_id(-1);
+    Prefilter* canonical = CanonicalNode(nodes, node);
+    if (canonical == NULL) {
+      // Any further nodes that have the same atom/subs
+      // will find this node as the canonical node.
+      nodes->emplace(node);
+      if (node->op() == Prefilter::ATOM) {
+        atom_vec->push_back(node->atom());
+        atom_index_to_id_.push_back(unique_id);
+      }
+      node->set_unique_id(unique_id++);
+    } else {
+      node->set_unique_id(canonical->unique_id());
+    }
+  }
+  entries_.resize(unique_id);
+
+  // Fill the entries.
+  for (int i = static_cast<int>(v.size()) - 1; i >= 0; i--) {
+    Prefilter* prefilter = v[i];
+    if (prefilter == NULL)
+      continue;
+    if (CanonicalNode(nodes, prefilter) != prefilter)
+      continue;
+    int id = prefilter->unique_id();
+    switch (prefilter->op()) {
+      default:
+        ABSL_LOG(DFATAL) << "Unexpected op: " << prefilter->op();
+        return;
+
+      case Prefilter::ATOM:
+        entries_[id].propagate_up_at_count = 1;
+        break;
+
+      case Prefilter::OR:
+      case Prefilter::AND: {
+        // For each child, we append our id to the child's list of
+        // parent ids... unless we happen to have done so already.
+        // The number of appends is the number of unique children,
+        // which allows correct upward propagation from AND nodes.
+        int up_count = 0;
+        for (size_t j = 0; j < prefilter->subs()->size(); j++) {
+          int child_id = (*prefilter->subs())[j]->unique_id();
+          std::vector<int>& parents = entries_[child_id].parents;
+          if (parents.empty() || parents.back() != id) {
+            parents.push_back(id);
+            up_count++;
+          }
+        }
+        entries_[id].propagate_up_at_count =
+            prefilter->op() == Prefilter::AND ? up_count : 1;
+        break;
+      }
+    }
+  }
+
+  // For top level nodes, populate regexp id.
+  for (size_t i = 0; i < prefilter_vec_.size(); i++) {
+    if (prefilter_vec_[i] == NULL)
+      continue;
+    int id = CanonicalNode(nodes, prefilter_vec_[i])->unique_id();
+    ABSL_DCHECK_LE(0, id);
+    Entry* entry = &entries_[id];
+    entry->regexps.push_back(static_cast<int>(i));
+  }
+
+  // Lastly, using probability-based heuristics, we identify nodes
+  // that trigger too many parents and then we try to prune edges.
+  // We use logarithms below to avoid the likelihood of underflow.
+  double log_num_regexps = std::log(prefilter_vec_.size() - unfiltered_.size());
+  // Hoisted this above the loop so that we don't thrash the heap.
+  std::vector<std::pair<size_t, int>> entries_by_num_edges;
+  for (int i = static_cast<int>(v.size()) - 1; i >= 0; i--) {
+    Prefilter* prefilter = v[i];
+    // Pruning applies only to AND nodes because it "just" reduces
+    // precision; applied to OR nodes, it would break correctness.
+    if (prefilter == NULL || prefilter->op() != Prefilter::AND)
+      continue;
+    if (CanonicalNode(nodes, prefilter) != prefilter)
+      continue;
+    int id = prefilter->unique_id();
+
+    // Sort the current node's children by the numbers of parents.
+    entries_by_num_edges.clear();
+    for (size_t j = 0; j < prefilter->subs()->size(); j++) {
+      int child_id = (*prefilter->subs())[j]->unique_id();
+      const std::vector<int>& parents = entries_[child_id].parents;
+      entries_by_num_edges.emplace_back(parents.size(), child_id);
+    }
+    std::stable_sort(entries_by_num_edges.begin(), entries_by_num_edges.end());
+
+    // A running estimate of how many regexps will be triggered by
+    // pruning the remaining children's edges to the current node.
+    // Our nominal target is one, so the threshold is log(1) == 0;
+    // pruning occurs iff the child has more than nine edges left.
+    double log_num_triggered = log_num_regexps;
+    for (const auto& pair : entries_by_num_edges) {
+      int child_id = pair.second;
+      std::vector<int>& parents = entries_[child_id].parents;
+      if (log_num_triggered > 0.) {
+        log_num_triggered += std::log(parents.size());
+        log_num_triggered -= log_num_regexps;
+      } else if (parents.size() > 9) {
+        auto it = std::find(parents.begin(), parents.end(), id);
+        if (it != parents.end()) {
+          parents.erase(it);
+          entries_[id].propagate_up_at_count--;
+        }
+      }
+    }
+  }
+}
+
+// Functions for triggering during search.
+void PrefilterTree::RegexpsGivenStrings(
+    const std::vector<int>& matched_atoms,
+    std::vector<int>* regexps) const {
+  regexps->clear();
+  if (!compiled_) {
+    // Some legacy users of PrefilterTree call Compile() before
+    // adding any regexps and expect Compile() to have no effect.
+    // This kludge is a counterpart to that kludge.
+    if (prefilter_vec_.empty()) {
+      return;
+    }
+
+    ABSL_LOG(ERROR) << "RegexpsGivenStrings called before Compile.";
+    for (size_t i = 0; i < prefilter_vec_.size(); i++)
+      regexps->push_back(static_cast<int>(i));
+  } else {
+    IntMap regexps_map(static_cast<int>(prefilter_vec_.size()));
+    std::vector<int> matched_atom_ids;
+    for (size_t j = 0; j < matched_atoms.size(); j++)
+      matched_atom_ids.push_back(atom_index_to_id_[matched_atoms[j]]);
+    PropagateMatch(matched_atom_ids, &regexps_map);
+    for (IntMap::const_iterator it = regexps_map.begin();
+         it != regexps_map.end();
+         ++it)
+      regexps->push_back(it->index());
+
+    regexps->insert(regexps->end(), unfiltered_.begin(), unfiltered_.end());
+  }
+  std::sort(regexps->begin(), regexps->end());
+}
+
+void PrefilterTree::PropagateMatch(const std::vector<int>& atom_ids,
+                                   IntMap* regexps) const {
+  IntMap count(static_cast<int>(entries_.size()));
+  IntMap work(static_cast<int>(entries_.size()));
+  for (size_t i = 0; i < atom_ids.size(); i++)
+    work.set(atom_ids[i], 1);
+  for (IntMap::const_iterator it = work.begin(); it != work.end(); ++it) {
+    const Entry& entry = entries_[it->index()];
+    // Record regexps triggered.
+    for (size_t i = 0; i < entry.regexps.size(); i++)
+      regexps->set(entry.regexps[i], 1);
+    int c;
+    // Pass trigger up to parents.
+    for (int j : entry.parents) {
+      const Entry& parent = entries_[j];
+      // Delay until all the children have succeeded.
+      if (parent.propagate_up_at_count > 1) {
+        if (count.has_index(j)) {
+          c = count.get_existing(j) + 1;
+          count.set_existing(j, c);
+        } else {
+          c = 1;
+          count.set_new(j, c);
+        }
+        if (c < parent.propagate_up_at_count)
+          continue;
+      }
+      // Trigger the parent.
+      work.set(j, 1);
+    }
+  }
+}
+
+// Debugging help.
+void PrefilterTree::PrintPrefilter(int regexpid) {
+  ABSL_LOG(ERROR) << DebugNodeString(prefilter_vec_[regexpid]);
+}
+
+void PrefilterTree::PrintDebugInfo(NodeSet* nodes) {
+  ABSL_LOG(ERROR) << "#Unique Atoms: " << atom_index_to_id_.size();
+  ABSL_LOG(ERROR) << "#Unique Nodes: " << entries_.size();
+
+  for (size_t i = 0; i < entries_.size(); i++) {
+    const std::vector<int>& parents = entries_[i].parents;
+    const std::vector<int>& regexps = entries_[i].regexps;
+    ABSL_LOG(ERROR) << "EntryId: " << i
+                    << " N: " << parents.size() << " R: " << regexps.size();
+    for (int parent : parents)
+      ABSL_LOG(ERROR) << parent;
+  }
+  ABSL_LOG(ERROR) << "Set:";
+  for (NodeSet::const_iterator iter = nodes->begin();
+       iter != nodes->end(); ++iter)
+    ABSL_LOG(ERROR) << "NodeId: " << (*iter)->unique_id();
+}
+
+std::string PrefilterTree::DebugNodeString(Prefilter* node) const {
+  std::string node_string = "";
+  if (node->op() == Prefilter::ATOM) {
+    ABSL_DCHECK(!node->atom().empty());
+    node_string += node->atom();
+  } else {
+    // Adding the operation disambiguates AND and OR nodes.
+    node_string +=  node->op() == Prefilter::AND ? "AND" : "OR";
+    node_string += "(";
+    for (size_t i = 0; i < node->subs()->size(); i++) {
+      if (i > 0)
+        node_string += ',';
+      node_string += absl::StrFormat("%d", (*node->subs())[i]->unique_id());
+      node_string += ":";
+      node_string += DebugNodeString((*node->subs())[i]);
+    }
+    node_string += ")";
+  }
+  return node_string;
+}
+
+}  // namespace re2

package/deps/re2/re2/prefilter_tree.h

@@ -0,0 +1,153 @@
+// Copyright 2009 The RE2 Authors.  All Rights Reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#ifndef RE2_PREFILTER_TREE_H_
+#define RE2_PREFILTER_TREE_H_
+
+// The PrefilterTree class is used to form an AND-OR tree of strings
+// that would trigger each regexp. The 'prefilter' of each regexp is
+// added to PrefilterTree, and then PrefilterTree is used to find all
+// the unique strings across the prefilters. During search, by using
+// matches from a string matching engine, PrefilterTree deduces the
+// set of regexps that are to be triggered. The 'string matching
+// engine' itself is outside of this class, and the caller can use any
+// favorite engine. PrefilterTree provides a set of strings (called
+// atoms) that the user of this class should use to do the string
+// matching.
+
+#include <string>
+#include <vector>
+
+#include "absl/container/flat_hash_set.h"
+#include "absl/log/absl_check.h"
+#include "absl/log/absl_log.h"
+#include "re2/prefilter.h"
+#include "re2/sparse_array.h"
+
+namespace re2 {
+
+class PrefilterTree {
+ public:
+  PrefilterTree();
+  explicit PrefilterTree(int min_atom_len);
+  ~PrefilterTree();
+
+  // Adds the prefilter for the next regexp. Note that we assume that
+  // Add called sequentially for all regexps. All Add calls
+  // must precede Compile.
+  void Add(Prefilter* prefilter);
+
+  // The Compile returns a vector of string in atom_vec.
+  // Call this after all the prefilters are added through Add.
+  // No calls to Add after Compile are allowed.
+  // The caller should use the returned set of strings to do string matching.
+  // Each time a string matches, the corresponding index then has to be
+  // and passed to RegexpsGivenStrings below.
+  void Compile(std::vector<std::string>* atom_vec);
+
+  // Given the indices of the atoms that matched, returns the indexes
+  // of regexps that should be searched.  The matched_atoms should
+  // contain all the ids of string atoms that were found to match the
+  // content. The caller can use any string match engine to perform
+  // this function. This function is thread safe.
+  void RegexpsGivenStrings(const std::vector<int>& matched_atoms,
+                           std::vector<int>* regexps) const;
+
+  // Print debug prefilter. Also prints unique ids associated with
+  // nodes of the prefilter of the regexp.
+  void PrintPrefilter(int regexpid);
+
+ private:
+  using IntMap = SparseArray<int>;
+
+  struct PrefilterHash {
+    size_t operator()(const Prefilter* a) const {
+      ABSL_DCHECK(a != NULL);
+      return absl::Hash<Prefilter>()(*a);
+    }
+  };
+
+  struct PrefilterEqual {
+    bool operator()(const Prefilter* a, const Prefilter* b) const {
+      ABSL_DCHECK(a != NULL);
+      ABSL_DCHECK(b != NULL);
+      return *a == *b;
+    }
+  };
+
+  using NodeSet =
+      absl::flat_hash_set<Prefilter*, PrefilterHash, PrefilterEqual>;
+
+  // Each unique node has a corresponding Entry that helps in
+  // passing the matching trigger information along the tree.
+  struct Entry {
+   public:
+    // How many children should match before this node triggers the
+    // parent. For an atom and an OR node, this is 1 and for an AND
+    // node, it is the number of unique children.
+    int propagate_up_at_count;
+
+    // When this node is ready to trigger the parent, what are the indices
+    // of the parent nodes to trigger. The reason there may be more than
+    // one is because of sharing. For example (abc | def) and (xyz | def)
+    // are two different nodes, but they share the atom 'def'. So when
+    // 'def' matches, it triggers two parents, corresponding to the two
+    // different OR nodes.
+    std::vector<int> parents;
+
+    // When this node is ready to trigger the parent, what are the
+    // regexps that are triggered.
+    std::vector<int> regexps;
+  };
+
+  // Returns true if the prefilter node should be kept.
+  bool KeepNode(Prefilter* node) const;
+
+  // This function assigns unique ids to various parts of the
+  // prefilter, by looking at if these nodes are already in the
+  // PrefilterTree.
+  void AssignUniqueIds(NodeSet* nodes, std::vector<std::string>* atom_vec);
+
+  // Given the matching atoms, find the regexps to be triggered.
+  void PropagateMatch(const std::vector<int>& atom_ids,
+                      IntMap* regexps) const;
+
+  // Returns the prefilter node that has the same atom/subs as this
+  // node. For the canonical node, returns node. Assumes that the
+  // children of node have already been assigned unique ids.
+  Prefilter* CanonicalNode(NodeSet* nodes, Prefilter* node);
+
+  // Recursively constructs a readable prefilter string.
+  std::string DebugNodeString(Prefilter* node) const;
+
+  // Used for debugging.
+  void PrintDebugInfo(NodeSet* nodes);
+
+  // These are all the nodes formed by Compile. Essentially, there is
+  // one node for each unique atom and each unique AND/OR node.
+  std::vector<Entry> entries_;
+
+  // indices of regexps that always pass through the filter (since we
+  // found no required literals in these regexps).
+  std::vector<int> unfiltered_;
+
+  // vector of Prefilter for all regexps.
+  std::vector<Prefilter*> prefilter_vec_;
+
+  // Atom index in returned strings to entry id mapping.
+  std::vector<int> atom_index_to_id_;
+
+  // Has the prefilter tree been compiled.
+  bool compiled_;
+
+  // Strings less than this length are not stored as atoms.
+  const int min_atom_len_;
+
+  PrefilterTree(const PrefilterTree&) = delete;
+  PrefilterTree& operator=(const PrefilterTree&) = delete;
+};
+
+}  // namespace
+
+#endif  // RE2_PREFILTER_TREE_H_