Simplified the lca implementation (now without state)

CMakeLists.txt

@@ -1,7 +1,7 @@
 project(Yannakakis)
 cmake_minimum_required(VERSION 2.8)
 
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11")
+set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++14")
 
 find_package(Boost REQUIRED COMPONENTS iostreams program_options filesystem system serialization)
 include_directories(SYSTEM ${Boost_INCLUDE_DIRS})

lib/separating_family.cpp

@@ -42,29 +42,18 @@ separating_family create_separating_family(const adaptive_distinguishing_sequenc
 				const auto s = p.second;
 				states[s] = true;
 			}
-			const auto root
-			    = lca(separating_sequences, [&states](state z) -> bool { return states[z]; });
+			const auto roots
+			    = multi_lca(separating_sequences, [&states](state z) -> bool { return states[z]; });
 
-			vector<word> stack_of_words;
-			const function<void(splitting_tree const &)> recursor = [&](splitting_tree const & n) {
-				if (n.children.empty()) {
-					for (state s : n.states) {
-						if (states[s]) {
-							for (auto const & w : stack_of_words) {
-								suffixes[s].insert(w);
-							}
-						}
+			// NOTE: this is slightly less efficient than doing the same thing inline in lca(...)
+			// but I was to lazy to write a dfs again
+			for (const splitting_tree & n : roots) {
+				for (state s : n.states) {
+					if (states[s]) {
+						suffixes[s].insert(n.separator);
 					}
-				} else {
-					if (n.mark > 1) stack_of_words.push_back(n.separator);
-					for (auto const & c : n.children) {
-						recursor(c);
-					}
-					if (n.mark > 1) stack_of_words.pop_back();
 				}
-			};
-
-			recursor(root);
+			}
 
 			// Finalize the suffixes
 			for (auto && p : node.CI) {

lib/splitting_tree.cpp

@@ -15,14 +15,6 @@ splitting_tree::splitting_tree(size_t N, size_t d) : states(N), depth(d) {
 	iota(begin(states), end(states), 0);
 }
 
-splitting_tree & lca_impl2(splitting_tree & node) {
-	if (node.mark > 1) return node;
-	for (auto && c : node.children) {
-		if (c.mark > 0) return lca_impl2(c);
-	}
-	return node; // this is a leaf
-}
-
 result create_splitting_tree(const mealy & g, options opt) {
 	const auto N = g.graph_size;
 	const auto P = g.input_size;

lib/splitting_tree.hpp

@@ -12,36 +12,66 @@ struct splitting_tree {
 	std::vector<splitting_tree> children;
 	word separator;
 	size_t depth = 0;
-	mutable int mark = 0; // used for some algorithms...
 };
 
-template <typename Fun> void lca_impl1(splitting_tree const & node, Fun && f) {
-	node.mark = 0;
-	if (!node.children.empty()) {
-		for (auto && c : node.children) {
-			lca_impl1(c, f);
-			if (c.mark) node.mark++;
+/// \brief the generic lca implementation.
+/// It uses \p store to store the relevant nodes (in some bottom up order), the last store is the
+/// actual lowest common ancestor (but the other might be relevant as well). The function \p f is
+/// the predicate on the states (returns true for the states we want to compute the lca of).
+template <typename Fun, typename Store>
+size_t lca_impl(splitting_tree const & node, Fun && f, Store && store) {
+	static_assert(std::is_same<decltype(f(state(0))), bool>::value, "f should return a bool");
+	if (node.children.empty()) {
+		// if it is a leaf, we search for the states
+		// if it contains a state, return this leaf
+		for (auto s : node.states) {
+			if (f(s)) {
+				store(node);
+				return 1;
+			}
 		}
+		// did not contain the leaf => nullptr
+		return 0;
 	} else {
-		for (auto && s : node.states) {
-			if (f(s)) node.mark++;
+		// otherwise, check our children. If there is a single one giving a node
+		// we return this (it's the lca), if more children return a non-nil
+		// node, then we are the lca
+		size_t count = 0;
+		for (auto & c : node.children) {
+			auto inner_count = lca_impl(c, f, store);
+			if (inner_count > 0) count++;
 		}
-	}
-}
 
-splitting_tree & lca_impl2(splitting_tree & node);
+		if (count >= 2) {
+			store(node);
+		}
+
+		return count;
+	}
+	throw std::logic_error("unreachable code");
+}
 
 /// \brief Find the lowest common ancestor of elements on which \p f returns true.
 template <typename Fun> splitting_tree & lca(splitting_tree & root, Fun && f) {
-	static_assert(std::is_same<decltype(f(0)), bool>::value, "f should return a bool");
-	lca_impl1(root, f);
-	return lca_impl2(root);
+	splitting_tree const * store = nullptr;
+	lca_impl(root, f, [&store](splitting_tree const & node) { store = &node; });
+	return const_cast<splitting_tree &>(*store); // NOTE: this const_cast is safe
 }
 
 template <typename Fun> const splitting_tree & lca(const splitting_tree & root, Fun && f) {
-	static_assert(std::is_same<decltype(f(0)), bool>::value, "f should return a bool");
-	lca_impl1(root, f);
-	return lca_impl2(const_cast<splitting_tree &>(root));
+	splitting_tree const * store = nullptr;
+	lca_impl(root, f, [&store](splitting_tree const & node) { store = &node; });
+	return *store;
+}
+
+/// \brief Find "all" lca's of elements on which \p f returns true.
+/// This can be used to collect all the separating sequences for the subset of states.
+template <typename Fun>
+std::vector<std::reference_wrapper<const splitting_tree>> multi_lca(const splitting_tree & root,
+                                                                    Fun && f) {
+	std::vector<std::reference_wrapper<const splitting_tree>> ret;
+	lca_impl(root, f, [&ret](splitting_tree const & node) { ret.emplace_back(node); });
+	return ret;
 }
 
 

src/CMakeLists.txt

@@ -3,11 +3,9 @@ set (CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
 
 file(GLOB sources "*.cpp")
 
-#foreach(source ${sources})
-#	get_filename_component(exec ${source} NAME_WE)
-#	add_executable(${exec} ${source})
-#	target_link_libraries(${exec} common ${libs})
-#endforeach()
+foreach(source ${sources})
+	get_filename_component(exec ${source} NAME_WE)
+	add_executable(${exec} ${source})
+	target_link_libraries(${exec} common ${libs})
+endforeach()
 
-add_executable(main main.cpp)
-target_link_libraries(main common ${libs})