vector<DepthMinMax> getDistancesFromSOM(const NGHolder &g_orig) {
// We operate on a temporary copy of the original graph here, so we don't
// have to mutate the original.
NGHolder g;
ue2::unordered_map<NFAVertex, NFAVertex> vmap; // vertex in g_orig to vertex in g
cloneHolder(g, g_orig, &vmap);
vector<NFAVertex> vstarts;
for (auto v : vertices_range(g)) {
if (is_virtual_start(v, g)) {
vstarts.push_back(v);
}
}
vstarts.push_back(g.startDs);
// wire the successors of every virtual start or startDs to g.start.
for (auto v : vstarts) {
wireSuccessorsToStart(g, v);
}
// drop the in-edges of every virtual start so that they don't participate
// in the depth calculation.
for (auto v : vstarts) {
clear_in_edges(v, g);
}
//dumpGraph("som_depth.dot", g.g);
vector<DepthMinMax> temp_depths; // numbered by vertex index in g
calcDepthsFrom(g, g.start, temp_depths);
// Transfer depths, indexed by vertex index in g_orig.
vector<DepthMinMax> depths(num_vertices(g_orig));
for (auto v_orig : vertices_range(g_orig)) {
assert(contains(vmap, v_orig));
NFAVertex v_new = vmap[v_orig];
u32 orig_idx = g_orig[v_orig].index;
DepthMinMax &d = depths.at(orig_idx);
if (v_orig == g_orig.startDs || is_virtual_start(v_orig, g_orig)) {
// StartDs and virtual starts always have zero depth.
d = DepthMinMax(0, 0);
} else {
u32 new_idx = g[v_new].index;
d = temp_depths.at(new_idx);
}
}
return depths;
}
bool somMayGoBackwards(NFAVertex u, const NGHolder &g,
const ue2::unordered_map<NFAVertex, u32> ®ion_map,
smgb_cache &cache) {
/* Need to ensure all matches of the graph g up to u contain no infixes
* which are also matches of the graph to u.
*
* This is basically the same as firstMatchIsFirst except we g is not
* always a dag. As we haven't gotten around to writing an execute_graph
* that operates on general graphs, we take some (hopefully) conservative
* short cuts.
*
* Note: if the u can be jumped we will take jump edges
* into account as a possibility of som going backwards
*
* TODO: write a generalised ng_execute_graph/make this less hacky
*/
assert(&g == &cache.g);
if (contains(cache.smgb, u)) {
return cache.smgb[u];
}
DEBUG_PRINTF("checking if som can go backwards on %u\n",
g[u].index);
set<NFAEdge> be;
BackEdges<set<NFAEdge>> backEdgeVisitor(be);
depth_first_search(
g.g, visitor(backEdgeVisitor)
.root_vertex(g.start)
.vertex_index_map(get(&NFAGraphVertexProps::index, g.g)));
bool rv;
if (0) {
exit:
DEBUG_PRINTF("using cached result\n");
cache.smgb[u] = rv;
return rv;
}
assert(contains(region_map, u));
const u32 u_region = region_map.at(u);
for (const auto &e : be) {
NFAVertex s = source(e, g);
NFAVertex t = target(e, g);
/* only need to worry about big cycles including/before u */
DEBUG_PRINTF("back edge %u %u\n", g[s].index,
g[t].index);
if (s != t && region_map.at(s) <= u_region) {
DEBUG_PRINTF("eek big cycle\n");
rv = true; /* big cycle -> eek */
goto exit;
}
}
ue2::unordered_map<NFAVertex, NFAVertex> orig_to_copy;
NGHolder c_g;
cloneHolder(c_g, g, &orig_to_copy);
for (NFAVertex v : vertices_range(g)) {
if (!is_virtual_start(v, g)) {
continue;
}
NFAVertex c_v = orig_to_copy[v];
orig_to_copy[v] = c_g.startDs;
for (NFAVertex c_w : adjacent_vertices_range(c_v, c_g)) {
add_edge_if_not_present(c_g.startDs, c_w, c_g);
}
clear_vertex(c_v, c_g);
}
NFAVertex c_u = orig_to_copy[u];
clear_in_edges(c_g.acceptEod, c_g);
add_edge(c_g.accept, c_g.acceptEod, c_g);
clear_in_edges(c_g.accept, c_g);
clear_out_edges(c_u, c_g);
if (hasSelfLoop(u, g)) {
add_edge(c_u, c_u, c_g);
}
add_edge(c_u, c_g.accept, c_g);
set<NFAVertex> u_succ;
insert(&u_succ, adjacent_vertices(u, g));
u_succ.erase(u);
for (auto t : inv_adjacent_vertices_range(u, g)) {
if (t == u) {
continue;
}
for (auto v : adjacent_vertices_range(t, g)) {
if (contains(u_succ, v)) {
add_edge(orig_to_copy[t], c_g.accept, c_g);
break;
}
}
}
pruneUseless(c_g);
be.clear();
depth_first_search(c_g.g, visitor(backEdgeVisitor).root_vertex(c_g.start).
vertex_index_map(get(&NFAGraphVertexProps::index, c_g.g)));
for (const auto &e : be) {
NFAVertex s = source(e, c_g);
NFAVertex t = target(e, c_g);
DEBUG_PRINTF("back edge %u %u\n", c_g[s].index, c_g[t].index);
if (s != t) {
assert(0);
DEBUG_PRINTF("eek big cycle\n");
rv = true; /* big cycle -> eek */
goto exit;
}
}
DEBUG_PRINTF("checking acyclic+selfloop graph\n");
rv = !firstMatchIsFirst(c_g);
DEBUG_PRINTF("som may regress? %d\n", (int)rv);
goto exit;
}
void clear_in_edges(NFAVertex v, NGHolder &h) {
h.isValidNumEdges = false;
clear_in_edges(v, h.g);
}
