vectorscan/src/nfagraph/ng_calc_components.cpp
2023-10-03 20:24:39 +03:00

460 lines
15 KiB
C++

/*
* Copyright (c) 2015-2018, Intel Corporation
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Intel Corporation nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
/** \file
* \brief Splits an NFA graph into its connected components.
*
* This pass takes a NGHolder and splits its graph into a set of connected
* components, returning them as individual NGHolder graphs. For example, the
* graph for the regex /foo.*bar|[a-z]{7,13}|hatstand|teakettle$/ will be split
* into four NGHolders, representing these four components:
*
* - /foo.*bar/
* - /[a-z]{7,13}/
* - /hatstand/
* - /teakettle$/
*
* The pass operates by creating an undirected graph from the input graph, and
* then using the BGL's connected_components algorithm to do the work, cloning
* the identified components into their own graphs. A "shell" of vertices
* is identified and removed first from the head and tail of the graph, in
* order to handle cases where there is a common head/tail region.
*
* Trivial cases, such as an alternation of single vertices like /a|b|c|d|e|f/,
* are not split, as later optimisations will handle these cases efficiently.
*/
#include "ng_calc_components.h"
#include "ng_depth.h"
#include "ng_holder.h"
#include "ng_prune.h"
#include "ng_util.h"
#include "grey.h"
#include "ue2common.h"
#include "util/graph_range.h"
#include "util/graph_undirected.h"
#include <map>
#include <vector>
#include <boost/graph/connected_components.hpp>
#include <boost/graph/filtered_graph.hpp>
using namespace std;
namespace ue2 {
static constexpr u32 MAX_HEAD_SHELL_DEPTH = 3;
static constexpr u32 MAX_TAIL_SHELL_DEPTH = 3;
/**
* \brief Returns true if the whole graph is just an alternation of character
* classes.
*/
bool isAlternationOfClasses(const NGHolder &g) {
for (auto v : vertices_range(g)) {
if (is_special(v, g)) {
continue;
}
// Vertex must have in edges from starts only.
for (auto u : inv_adjacent_vertices_range(v, g)) {
if (!is_any_start(u, g)) {
return false;
}
}
// Vertex must have out edges to accepts only.
for (auto w : adjacent_vertices_range(v, g)) {
if (!is_any_accept(w, g)) {
return false;
}
}
}
DEBUG_PRINTF("alternation of single states, treating as one comp\n");
return true;
}
/**
* \brief Compute initial max distance to v from start (i.e. ignoring its own
* self-loop).
*/
static
depth max_dist_from_start(const NGHolder &g,
const vector<NFAVertexBidiDepth> &depths,
NFAVertex v) {
depth max_depth(0);
for (const auto u : inv_adjacent_vertices_range(v, g)) {
if (u == v) {
continue;
}
const auto &d = depths.at(g[u].index);
if (d.fromStart.max.is_reachable()) {
max_depth = max(max_depth, d.fromStart.max);
}
if (d.fromStartDotStar.max.is_reachable()) {
max_depth = max(max_depth, d.fromStartDotStar.max);
}
}
return max_depth + 1;
}
/**
* \brief Compute initial max depth from v from accept (i.e. ignoring its own
* self-loop).
*/
static
depth max_dist_to_accept(const NGHolder &g,
const vector<NFAVertexBidiDepth> &depths,
NFAVertex v) {
depth max_depth(0);
for (const auto w : adjacent_vertices_range(v, g)) {
if (w == v) {
continue;
}
const auto &d = depths.at(g[w].index);
if (d.toAccept.max.is_reachable()) {
max_depth = max(max_depth, d.toAccept.max);
}
if (d.toAcceptEod.max.is_reachable()) {
max_depth = max(max_depth, d.toAcceptEod.max);
}
}
return max_depth + 1;
}
static
flat_set<NFAVertex> findHeadShell(const NGHolder &g,
const vector<NFAVertexBidiDepth> &depths,
const depth &max_dist) {
flat_set<NFAVertex> shell;
for (auto v : vertices_range(g)) {
if (is_special(v, g)) {
continue;
}
if (max_dist_from_start(g, depths, v) <= max_dist) {
shell.insert(v);
}
}
for (UNUSED auto v : shell) {
DEBUG_PRINTF("shell: %zu\n", g[v].index);
}
return shell;
}
static
flat_set<NFAVertex> findTailShell(const NGHolder &g,
const vector<NFAVertexBidiDepth> &depths,
const depth &max_dist) {
flat_set<NFAVertex> shell;
for (auto v : vertices_range(g)) {
if (is_special(v, g)) {
continue;
}
if (max_dist_to_accept(g, depths, v) <= max_dist) {
shell.insert(v);
}
}
for (UNUSED auto v : shell) {
DEBUG_PRINTF("shell: %zu\n", g[v].index);
}
return shell;
}
static
vector<NFAEdge> findShellEdges(const NGHolder &g,
const flat_set<NFAVertex> &head_shell,
const flat_set<NFAVertex> &tail_shell) {
vector<NFAEdge> shell_edges;
for (const auto &e : edges_range(g)) {
auto u = source(e, g);
auto v = target(e, g);
if (v == g.startDs && is_any_start(u, g)) {
continue;
}
if (u == g.accept && v == g.acceptEod) {
continue;
}
if ((is_special(u, g) || contains(head_shell, u)) &&
(is_special(v, g) || contains(tail_shell, v))) {
DEBUG_PRINTF("edge (%zu,%zu) is a shell edge\n", g[u].index,
g[v].index);
shell_edges.emplace_back(e);
}
}
return shell_edges;
}
template<typename GetAdjRange>
bool shellHasOnePath(const NGHolder &g, const flat_set<NFAVertex> &shell,
GetAdjRange adj_range_func) {
if (shell.empty()) {
DEBUG_PRINTF("no shell\n");
return false;
}
NFAVertex exit_vertex = NGHolder::null_vertex();
for (auto u : shell) {
for (auto v : adj_range_func(u, g)) {
if (contains(shell, v)) {
continue;
}
if (!exit_vertex) {
exit_vertex = v;
continue;
}
if (exit_vertex == v) {
continue;
}
return false;
}
}
return true;
}
/**
* True if all edges out of vertices in the head shell lead to at most a single
* outside vertex, or the inverse for the tail shell.
*/
static
bool shellHasOnePath(const NGHolder &g, const flat_set<NFAVertex> &head_shell,
const flat_set<NFAVertex> &tail_shell) {
if (shellHasOnePath(g, head_shell, adjacent_vertices_range<NGHolder>)) {
DEBUG_PRINTF("head shell has only one path through it\n");
return true;
}
if (shellHasOnePath(g, tail_shell, inv_adjacent_vertices_range<NGHolder>)) {
DEBUG_PRINTF("tail shell has only one path into it\n");
return true;
}
return false;
}
/**
* Common code called by calc- and recalc- below. Splits the given holder into
* one or more connected components, adding them to the comps deque.
*/
static
void splitIntoComponents(unique_ptr<NGHolder> g,
deque<unique_ptr<NGHolder>> &comps,
const depth &max_head_depth,
const depth &max_tail_depth, bool *shell_comp) {
DEBUG_PRINTF("graph has %zu vertices\n", num_vertices(*g));
assert(shell_comp);
*shell_comp = false;
// Compute "shell" head and tail subgraphs.
auto depths = calcBidiDepths(*g);
auto head_shell = findHeadShell(*g, depths, max_head_depth);
auto tail_shell = findTailShell(*g, depths, max_tail_depth);
for (auto v : head_shell) {
tail_shell.erase(v);
}
if (head_shell.size() + tail_shell.size() + N_SPECIALS >=
num_vertices(*g)) {
DEBUG_PRINTF("all in shell component\n");
comps.emplace_back(std::move(g));
*shell_comp = true;
return;
}
// Find edges connecting the head and tail shells directly.
vector<NFAEdge> shell_edges = findShellEdges(*g, head_shell, tail_shell);
DEBUG_PRINTF("%zu vertices in head, %zu in tail, %zu shell edges\n",
head_shell.size(), tail_shell.size(), shell_edges.size());
// If there are no shell edges and only one path out of the head shell or
// into the tail shell, we aren't going to find more than one component.
if (shell_edges.empty() && shellHasOnePath(*g, head_shell, tail_shell)) {
DEBUG_PRINTF("single component\n");
comps.emplace_back(std::move(g));
return;
}
auto ug = make_undirected_graph(*g);
// Filter specials and shell vertices from undirected graph.
unordered_set<NFAVertex> bad_vertices(
{g->start, g->startDs, g->accept, g->acceptEod});
bad_vertices.insert(head_shell.begin(), head_shell.end());
bad_vertices.insert(tail_shell.begin(), tail_shell.end());
auto filtered_ug = boost::make_filtered_graph(
ug, boost::keep_all(), make_bad_vertex_filter(&bad_vertices));
// Actually run the connected components algorithm.
map<NFAVertex, u32> split_components;
const u32 num = connected_components(
filtered_ug, boost::make_assoc_property_map(split_components));
assert(num > 0);
if (num == 1 && shell_edges.empty()) {
DEBUG_PRINTF("single component\n");
comps.emplace_back(std::move(g));
return;
}
DEBUG_PRINTF("broke graph into %u components\n", num);
vector<deque<NFAVertex>> verts(num);
// Collect vertex lists per component.
for (const auto &m : split_components) {
NFAVertex v = m.first;
u32 c = m.second;
verts[c].emplace_back(v);
DEBUG_PRINTF("vertex %zu is in comp %u\n", (*g)[v].index, c);
}
unordered_map<NFAVertex, NFAVertex> v_map; // temp map for fillHolder
for (auto &vv : verts) {
// Shells are in every component.
vv.insert(vv.end(), begin(head_shell), end(head_shell));
vv.insert(vv.end(), begin(tail_shell), end(tail_shell));
/* Sort for determinism. Still required as NFAUndirectedVertex have
* no deterministic ordering (split_components map). */
sort(begin(vv), end(vv));
auto gc = std::make_unique<NGHolder>();
v_map.clear();
fillHolder(gc.get(), *g, vv, &v_map);
// Remove shell edges, which will get their own component.
for (const auto &e : shell_edges) {
auto cu = v_map.at(source(e, *g));
auto cv = v_map.at(target(e, *g));
assert(edge(cu, cv, *gc).second);
remove_edge(cu, cv, *gc);
}
pruneUseless(*gc);
DEBUG_PRINTF("component %zu has %zu vertices\n", comps.size(),
num_vertices(*gc));
comps.emplace_back(std::move(gc));
}
// Another component to handle the direct shell-to-shell edges.
if (!shell_edges.empty()) {
deque<NFAVertex> vv;
vv.insert(vv.end(), begin(head_shell), end(head_shell));
vv.insert(vv.end(), begin(tail_shell), end(tail_shell));
auto gc = std::make_unique<NGHolder>();
v_map.clear();
fillHolder(gc.get(), *g, vv, &v_map);
pruneUseless(*gc);
DEBUG_PRINTF("shell edge component %zu has %zu vertices\n",
comps.size(), num_vertices(*gc));
comps.emplace_back(std::move(gc));
*shell_comp = true;
}
// Ensure that only vertices with accept edges have reports.
for (auto &gc : comps) {
assert(gc);
clearReports(*gc);
}
// We should never produce empty component graphs.
assert(all_of(begin(comps), end(comps),
[](const unique_ptr<NGHolder> &g_comp) {
return num_vertices(*g_comp) > N_SPECIALS;
}));
}
deque<unique_ptr<NGHolder>> calcComponents(unique_ptr<NGHolder> g,
const Grey &grey) {
deque<unique_ptr<NGHolder>> comps;
// For trivial cases, we needn't bother running the full
// connected_components algorithm.
if (!grey.calcComponents || isAlternationOfClasses(*g)) {
comps.emplace_back(std::move(g));
return comps;
}
bool shell_comp = false;
splitIntoComponents(std::move(g), comps, depth(MAX_HEAD_SHELL_DEPTH),
depth(MAX_TAIL_SHELL_DEPTH), &shell_comp);
if (shell_comp) {
DEBUG_PRINTF("re-running on shell comp\n");
assert(!comps.empty());
auto sc = std::move(comps.back());
comps.pop_back();
splitIntoComponents(std::move(sc), comps, depth(0), depth(0),
&shell_comp);
}
DEBUG_PRINTF("finished; split into %zu components\n", comps.size());
return comps;
}
void recalcComponents(deque<unique_ptr<NGHolder>> &comps, const Grey &grey) {
if (!grey.calcComponents) {
return;
}
deque<unique_ptr<NGHolder>> out;
for (auto &gc : comps) {
if (!gc) {
continue; // graph has been consumed already.
}
if (isAlternationOfClasses(*gc)) {
out.emplace_back(std::move(gc));
continue;
}
auto gc_comps = calcComponents(std::move(gc), grey);
out.insert(end(out), std::make_move_iterator(begin(gc_comps)),
std::make_move_iterator(end(gc_comps)));
}
// Replace comps with our recalculated list.
comps.swap(out);
}
} // namespace ue2