re2/prefilter_tree.cc

// 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 <memory>
#include <string>
#include <utility>
#include <vector>

#include "absl/strings/str_format.h"
#include "util/logging.h"
#include "re2/prefilter.h"
#include "re2/re2.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_) {
    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_) {
    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:
      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:
        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();
    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;

    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) {
  LOG(ERROR) << DebugNodeString(prefilter_vec_[regexpid]);
}

void PrefilterTree::PrintDebugInfo(NodeSet* nodes) {
  LOG(ERROR) << "#Unique Atoms: " << atom_index_to_id_.size();
  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;
    LOG(ERROR) << "EntryId: " << i
               << " N: " << parents.size() << " R: " << regexps.size();
    for (int parent : parents)
      LOG(ERROR) << parent;
  }
  LOG(ERROR) << "Set:";
  for (NodeSet::const_iterator iter = nodes->begin();
       iter != nodes->end(); ++iter)
    LOG(ERROR) << "NodeId: " << (*iter)->unique_id();
}

std::string PrefilterTree::DebugNodeString(Prefilter* node) const {
  std::string node_string = "";
  if (node->op() == Prefilter::ATOM) {
    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