/** Remove edges to accepts that can never produce a match long enough to
* satisfy our min_length and max_offset constraints. */
static
void pruneUnmatchable(NGWrapper &g, const ReportManager &rm) {
if (!g.min_length) {
return;
}
vector<DepthMinMax> depths = getDistancesFromSOM(g);
pruneUnmatchable(g, depths, rm, g.accept);
pruneUnmatchable(g, depths, rm, g.acceptEod);
pruneUseless(g);
}
/** \brief Find the (min, max) length of any match for the given holder. */
static
DepthMinMax findMatchLengths(const ReportManager &rm, const NGHolder &g) {
DepthMinMax match_depths;
vector<DepthMinMax> depths = getDistancesFromSOM(g);
pair<s32, s32> adj;
for (auto v : inv_adjacent_vertices_range(g.accept, g)) {
u32 idx = g[v].index;
DepthMinMax d = depths[idx]; // copy
adj = getMinMaxOffsetAdjust(rm, g, v);
DEBUG_PRINTF("vertex %u: depths=%s, adj=[%d,%d]\n", idx,
d.str().c_str(), adj.first, adj.second);
d.min += adj.first;
d.max += adj.second;
match_depths = unionDepthMinMax(match_depths, d);
}
for (auto v : inv_adjacent_vertices_range(g.acceptEod, g)) {
if (v == g.accept) {
continue;
}
u32 idx = g[v].index;
DepthMinMax d = depths[idx]; // copy
adj = getMinMaxOffsetAdjust(rm, g, v);
DEBUG_PRINTF("vertex %u: depths=%s, adj=[%d,%d]\n", idx,
d.str().c_str(), adj.first, adj.second);
d.min += adj.first;
d.max += adj.second;
match_depths = unionDepthMinMax(match_depths, d);
}
DEBUG_PRINTF("match_depths=%s\n", match_depths.str().c_str());
assert(match_depths.min.is_reachable());
assert(match_depths.max.is_reachable());
return match_depths;
}
map<NFAVertex, NFAStateSet> findSquashers(const NGHolder &g, som_type som) {
map<NFAVertex, NFAStateSet> squash;
// Number of bits to use for all our masks. If we're a triggered graph,
// tops have already been assigned, so we don't have to account for them.
const u32 numStates = num_vertices(g);
// Build post-dominator tree.
PostDomTree pdom_tree;
buildPDomTree(g, pdom_tree);
// Build list of vertices by state ID and a set of init states.
vector<NFAVertex> vByIndex(numStates, NFAGraph::null_vertex());
NFAStateSet initStates(numStates);
smgb_cache cache(g);
// Mappings used for SOM mode calculations, otherwise left empty.
unordered_map<NFAVertex, u32> region_map;
vector<DepthMinMax> som_depths;
if (som) {
region_map = assignRegions(g);
som_depths = getDistancesFromSOM(g);
}
for (auto v : vertices_range(g)) {
const u32 vert_id = g[v].index;
DEBUG_PRINTF("vertex %u/%u\n", vert_id, numStates);
assert(vert_id < numStates);
vByIndex[vert_id] = v;
if (is_any_start(v, g) || !in_degree(v, g)) {
initStates.set(vert_id);
}
}
for (u32 i = 0; i < numStates; i++) {
NFAVertex v = vByIndex[i];
assert(v != NFAGraph::null_vertex());
const CharReach &cr = g[v].char_reach;
/* only non-init cyclics can be squashers */
if (!hasSelfLoop(v, g) || initStates.test(i)) {
continue;
}
DEBUG_PRINTF("state %u is cyclic\n", i);
NFAStateSet mask(numStates), succ(numStates), pred(numStates);
buildSquashMask(mask, g, v, cr, initStates, vByIndex, pdom_tree, som,
som_depths, region_map, cache);
buildSucc(succ, g, v);
buildPred(pred, g, v);
const auto &reports = g[v].reports;
for (size_t j = succ.find_first(); j != succ.npos;
j = succ.find_next(j)) {
NFAVertex vj = vByIndex[j];
NFAStateSet pred2(numStates);
buildPred(pred2, g, vj);
if (pred2 == pred) {
DEBUG_PRINTF("adding the sm from %zu to %u's sm\n", j, i);
NFAStateSet tmp(numStates);
buildSquashMask(tmp, g, vj, cr, initStates, vByIndex, pdom_tree,
som, som_depths, region_map, cache);
mask &= tmp;
}
}
for (size_t j = pred.find_first(); j != pred.npos;
j = pred.find_next(j)) {
NFAVertex vj = vByIndex[j];
NFAStateSet succ2(numStates);
buildSucc(succ2, g, vj);
/* we can use j as a basis for squashing if its succs are a subset
* of ours */
if ((succ2 & ~succ).any()) {
continue;
}
if (som) {
/* We cannot use j to add to the squash mask of v if it may
* have an earlier start of match offset. ie for us j as a
* basis for the squash mask of v we require:
* maxSomDist(j) <= minSomDist(v)
*/
/* ** TODO ** */
const depth &max_som_dist_j =
som_depths[g[vj].index].max;
const depth &min_som_dist_v =
som_depths[g[v].index].min;
if (max_som_dist_j > min_som_dist_v ||
max_som_dist_j.is_infinite()) {
/* j can't be used as it may be storing an earlier SOM */
continue;
}
}
const CharReach &crv = g[vj].char_reach;
/* we also require that j's report information be a subset of ours
*/
bool seen_special = false;
for (auto w : adjacent_vertices_range(vj, g)) {
if (is_special(w, g)) {
if (!edge(v, w, g).second) {
goto next_j;
}
seen_special = true;
}
}
// FIXME: should be subset check?
if (seen_special && g[vj].reports != reports) {
continue;
}
/* ok we can use j */
if ((crv & ~cr).none()) {
NFAStateSet tmp(numStates);
buildSquashMask(tmp, g, vj, cr, initStates, vByIndex, pdom_tree,
som, som_depths, region_map, cache);
mask &= tmp;
mask.reset(j);
}
next_j:;
}
mask.set(i); /* never clear ourselves */
if ((~mask).any()) { // i.e. some bits unset in mask
DEBUG_PRINTF("%u squashes %zu other states\n", i, (~mask).count());
squash.emplace(v, mask);
}
}
findDerivedSquashers(g, vByIndex, pdom_tree, initStates, &squash, som,
som_depths, region_map, cache);
clearMutualSquashers(g, vByIndex, squash);
return squash;
}
