v3 simple SoA, not really an improvement bc index hints remove most find iterations

CMakeLists.txt

@@ -38,6 +38,9 @@ endif()
 add_subdirectory(./prototyping EXCLUDE_FROM_ALL)
 
 add_subdirectory(./version0)
+add_subdirectory(./version1)
+add_subdirectory(./version2)
+add_subdirectory(./version3)
 
 add_subdirectory(./bench)
 

bench/CMakeLists.txt

@@ -15,14 +15,34 @@ target_link_libraries(crdt_bench_jpaper_v0 PUBLIC
 
 ########################################
 
-if (FALSE)
 add_executable(crdt_bench_jpaper_v1
-	./test1.cpp
+	./v1_jpaper.cpp
 )
 
 target_link_libraries(crdt_bench_jpaper_v1 PUBLIC
 	crdt_version1
 	nlohmann_json::nlohmann_json
 )
-endif()
+
+########################################
+
+add_executable(crdt_bench_jpaper_v2
+	./v2_jpaper.cpp
+)
+
+target_link_libraries(crdt_bench_jpaper_v2 PUBLIC
+	crdt_version2
+	nlohmann_json::nlohmann_json
+)
+
+########################################
+
+add_executable(crdt_bench_jpaper_v3
+	./v3_jpaper.cpp
+)
+
+target_link_libraries(crdt_bench_jpaper_v3 PUBLIC
+	crdt_version3
+	nlohmann_json::nlohmann_json
+)
 

bench/README.md

@@ -16,12 +16,12 @@ the json contains:
 ## baseline ( just walking through the json, no insertions )
 
 - g++9 -g :
-	- 23.0s
-	- 22.6s
+	- 23.0s		~11294 ops/s
+	- 22.6s		~11494 ops/s
 	- 23.0s
 
 - g++9 -O3 -DNDEBUG :
-	- 9.6s
+	- 9.6s		~27060 ops/s
 	- 9.7s
 	- 9.7s
 
@@ -29,10 +29,32 @@ the json contains:
 ## version0
 
 - g++9 -g -O2 :
-	- 10m35s
+	- 10m35s	~409 ops/s
 
 - g++9 -O3 -DNDEBUG :
-	- 8m7s
+	- 8m7s		~533 ops/s
 
-## version1
+## version1 - actor index
+
+- g++9 -g -O2 :
+	- 4m1s		~1077 ops/s
+
+- g++9 -O3 -DNDEBUG :
+	- 4m5s		~1060 ops/s
+
+## version2 - find with hint, cache last insert and use as hint
+
+- g++9 -g -O2 :
+	- 3m38s		~1191 ops/s
+
+- g++9 -O3 -DNDEBUG :
+	- 3m43s		~1164 ops/s
+
+## version3 - SoA, 1 array only ids, 1 array rest (parents, data)
+
+- g++9 -g -O2 :
+	- 3m36s		~1202 ops/s
+
+- g++9 -O3 -DNDEBUG :
+	- 3m44s		~1159 ops/s
 

bench/v1_jpaper.cpp

@@ -0,0 +1,203 @@
+#define EXTRA_ASSERTS 0
+
+#include <green_crdt/v1/list.hpp>
+#include <nlohmann/json.hpp>
+
+#include <unordered_map>
+#include <string_view>
+#include <fstream>
+#include <iostream>
+#include <cassert>
+
+using ActorID = std::array<uint8_t, 32>;
+//using Doc = GreenCRDT::V0::TextDocument<ActorID>;
+using List = GreenCRDT::V1::List<char, ActorID>;
+
+template<>
+struct std::hash<ActorID> {
+	std::size_t operator()(ActorID const& s) const noexcept {
+		static_assert(sizeof(size_t) == 8);
+		// TODO: maybe shuffle the indices a bit
+		return
+			(static_cast<size_t>(s[0]) << 8*0) |
+			(static_cast<size_t>(s[1]) << 8*1) |
+			(static_cast<size_t>(s[2]) << 8*2) |
+			(static_cast<size_t>(s[3]) << 8*3) |
+			(static_cast<size_t>(s[4]) << 8*4) |
+			(static_cast<size_t>(s[5]) << 8*5) |
+			(static_cast<size_t>(s[6]) << 8*6) |
+			(static_cast<size_t>(s[7]) << 8*7)
+		;
+	}
+};
+
+// for dev, benching in debug is usefull, but only if the ammount of asserts is reasonable
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace detail {
+	uint8_t nib_from_hex(char c) {
+		extra_assert((c >= '0' && c <= '9') || (c >= 'a' && c <= 'f'));
+
+		if (c >= '0' && c <= '9') {
+			return static_cast<uint8_t>(c) - '0';
+		} else if (c >= 'a' && c <= 'f') {
+			return (static_cast<uint8_t>(c) - 'a') + 10u;
+		} else {
+			return 0u;
+		}
+	}
+} // detail
+
+static ActorID ActorIDFromStr(std::string_view str) {
+	extra_assert(str.size() == 32*2);
+	ActorID tmp;
+
+	for (size_t i = 0; i < tmp.size(); i++) {
+		tmp[i] = detail::nib_from_hex(str[i*2]) << 4 | detail::nib_from_hex(str[i*2+1]);
+	}
+
+	return tmp;
+}
+
+// seq@ID type format used in the json
+struct JObj {
+	ActorID id;
+	uint64_t seq {0};
+};
+
+static JObj JObjFromStr(std::string_view str) {
+	extra_assert(str.size() > 32*2 + 1);
+
+	size_t at_pos = str.find_first_of('@');
+	auto seq_sv = str.substr(0, at_pos);
+	auto id_sv = str.substr(at_pos+1);
+
+	assert(seq_sv.size() != 0);
+	assert(id_sv.size() == 32*2);
+
+	uint64_t tmp_seq {0};
+	for (size_t i = 0; i < seq_sv.size(); i++) {
+		assert(seq_sv[i] >= '0' && seq_sv[i] <= '9');
+		tmp_seq *= 10;
+		tmp_seq += seq_sv[i] - '0';
+	}
+
+	return {ActorIDFromStr(id_sv), tmp_seq};
+}
+
+int main(void) {
+	List list;
+
+	std::ifstream file {"../res/paper.json"};
+	std::cout << "start reading...\n";
+
+	uint64_t g_total_inserts {0};
+	uint64_t g_total_deletes {0};
+	//uint64_t g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, uint64_t> g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, std::unordered_map<uint64_t, uint64_t>> map_seq; // maps json op_seq -> lits id seq
+
+	for (std::string line; std::getline(file, line); ) {
+		nlohmann::json j_entry = nlohmann::json::parse(line);
+		const ActorID actor = ActorIDFromStr(static_cast<const std::string&>(j_entry["actor"]));
+		uint64_t op_seq = j_entry["startOp"];
+		for (const auto& j_op : j_entry["ops"]) {
+			if (j_op["action"] == "set") {
+				const auto obj = JObjFromStr(static_cast<const std::string&>(j_op["obj"]));
+				if (obj.seq != 1) {
+					// skip all non text edits (create text doc, curser etc)
+					continue;
+				}
+
+				if (j_op["insert"]) {
+					const auto& j_parent = j_op["key"];
+					extra_assert(!j_parent.is_null());
+					if (j_parent == "_head") {
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							std::nullopt,
+							std::nullopt
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					} else { // we have a parrent
+						extra_assert(static_cast<const std::string&>(j_op["value"]).size() == 1);
+
+						// split parent into seq and actor
+						const auto parent_left = JObjFromStr(static_cast<const std::string&>(j_parent));
+						auto idx_opt = list.findIdx({parent_left.id, map_seq[parent_left.id][parent_left.seq]});
+						assert(idx_opt.has_value());
+
+						std::optional<List::ListID> parent_left_id;
+						{
+							const auto& tmp_parent_left_id = list.list.at(idx_opt.value()).id;
+							parent_left_id = {list._actors[tmp_parent_left_id.actor_idx], tmp_parent_left_id.seq};
+						}
+
+						std::optional<List::ListID> parent_right_id;
+						if (idx_opt.value()+1 < list.list.size()) {
+							const auto& tmp_parent_right_id = list.list.at(idx_opt.value()+1).id;
+							parent_right_id = {list._actors[tmp_parent_right_id.actor_idx], tmp_parent_right_id.seq};
+						}
+
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							parent_left_id,
+							parent_right_id
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					}
+				} else {
+					// i think this is curser movement
+				}
+			} else if (j_op["action"] == "del") {
+				const auto list_id = JObjFromStr(static_cast<const std::string&>(j_op["key"]));
+				bool r = list.del({list_id.id, map_seq[list_id.id][list_id.seq]});
+				assert(r);
+				g_total_deletes++;
+			} else if (j_op["action"] == "makeText") {
+				// doc.clear();
+			} else if (j_op["action"] == "makeMap") {
+				// no idea
+			} else {
+				std::cout << "op: " << j_op << "\n";
+			}
+
+			op_seq++;
+		}
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list.list.size() << "\n";
+	std::cout << "doc size: " << list.doc_size << "\n";
+	std::cout << "total inserts: " << g_total_inserts << "\n";
+	std::cout << "total deletes: " << g_total_deletes << "\n";
+	std::cout << "total ops: " << g_total_inserts + g_total_deletes << "\n";
+
+	// checked, looks correct
+#if 0
+	std::cout << "doc text:\n";
+	// simple print
+	for (const auto& it : list.list) {
+		if (it.value) {
+			std::cout << it.value.value();
+		}
+	}
+	std::cout << "\n";
+#endif
+
+	return 0;
+}
+

bench/v2_jpaper.cpp

@@ -0,0 +1,211 @@
+#define EXTRA_ASSERTS 0
+
+#include <green_crdt/v2/list.hpp>
+#include <nlohmann/json.hpp>
+
+#include <unordered_map>
+#include <string_view>
+#include <fstream>
+#include <iostream>
+#include <cassert>
+
+using ActorID = std::array<uint8_t, 32>;
+using List = GreenCRDT::V2::List<char, ActorID>;
+
+template<>
+struct std::hash<ActorID> {
+	std::size_t operator()(ActorID const& s) const noexcept {
+		static_assert(sizeof(size_t) == 8);
+		// TODO: maybe shuffle the indices a bit
+		return
+			(static_cast<size_t>(s[0]) << 8*0) |
+			(static_cast<size_t>(s[1]) << 8*1) |
+			(static_cast<size_t>(s[2]) << 8*2) |
+			(static_cast<size_t>(s[3]) << 8*3) |
+			(static_cast<size_t>(s[4]) << 8*4) |
+			(static_cast<size_t>(s[5]) << 8*5) |
+			(static_cast<size_t>(s[6]) << 8*6) |
+			(static_cast<size_t>(s[7]) << 8*7)
+		;
+	}
+};
+
+// for dev, benching in debug is usefull, but only if the ammount of asserts is reasonable
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace detail {
+	uint8_t nib_from_hex(char c) {
+		extra_assert((c >= '0' && c <= '9') || (c >= 'a' && c <= 'f'));
+
+		if (c >= '0' && c <= '9') {
+			return static_cast<uint8_t>(c) - '0';
+		} else if (c >= 'a' && c <= 'f') {
+			return (static_cast<uint8_t>(c) - 'a') + 10u;
+		} else {
+			return 0u;
+		}
+	}
+} // detail
+
+static ActorID ActorIDFromStr(std::string_view str) {
+	extra_assert(str.size() == 32*2);
+	ActorID tmp;
+
+	for (size_t i = 0; i < tmp.size(); i++) {
+		tmp[i] = detail::nib_from_hex(str[i*2]) << 4 | detail::nib_from_hex(str[i*2+1]);
+	}
+
+	return tmp;
+}
+
+// seq@ID type format used in the json
+struct JObj {
+	ActorID id;
+	uint64_t seq {0};
+};
+
+static JObj JObjFromStr(std::string_view str) {
+	extra_assert(str.size() > 32*2 + 1);
+
+	size_t at_pos = str.find_first_of('@');
+	auto seq_sv = str.substr(0, at_pos);
+	auto id_sv = str.substr(at_pos+1);
+
+	assert(seq_sv.size() != 0);
+	assert(id_sv.size() == 32*2);
+
+	uint64_t tmp_seq {0};
+	for (size_t i = 0; i < seq_sv.size(); i++) {
+		assert(seq_sv[i] >= '0' && seq_sv[i] <= '9');
+		tmp_seq *= 10;
+		tmp_seq += seq_sv[i] - '0';
+	}
+
+	return {ActorIDFromStr(id_sv), tmp_seq};
+}
+
+int main(void) {
+	List list;
+
+	std::ifstream file {"../res/paper.json"};
+	std::cout << "start reading...\n";
+
+	uint64_t g_total_inserts {0};
+	uint64_t g_total_deletes {0};
+	//uint64_t g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, uint64_t> g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, std::unordered_map<uint64_t, uint64_t>> map_seq; // maps json op_seq -> lits id seq
+
+	for (std::string line; std::getline(file, line); ) {
+		nlohmann::json j_entry = nlohmann::json::parse(line);
+		const ActorID actor = ActorIDFromStr(static_cast<const std::string&>(j_entry["actor"]));
+		const size_t actor_idx = list.findActor(actor).value_or(0u);
+		uint64_t op_seq = j_entry["startOp"];
+		for (const auto& j_op : j_entry["ops"]) {
+			if (j_op["action"] == "set") {
+				const auto obj = JObjFromStr(static_cast<const std::string&>(j_op["obj"]));
+				if (obj.seq != 1) {
+					// skip all non text edits (create text doc, curser etc)
+					continue;
+				}
+
+				if (j_op["insert"]) {
+					const auto& j_parent = j_op["key"];
+					extra_assert(!j_parent.is_null());
+					if (j_parent == "_head") {
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							std::nullopt,
+							std::nullopt
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					} else { // we have a parrent
+						extra_assert(static_cast<const std::string&>(j_op["value"]).size() == 1);
+
+						size_t hint_last_insert {0};
+						if (list.last_inserted_idx.count(actor_idx)) {
+							hint_last_insert = list.last_inserted_idx[actor_idx];
+						}
+
+						// split parent into seq and actor
+						const auto parent_left = JObjFromStr(static_cast<const std::string&>(j_parent));
+						auto idx_opt = list.findIdx({parent_left.id, map_seq[parent_left.id][parent_left.seq]}, hint_last_insert);
+						assert(idx_opt.has_value());
+
+						std::optional<List::ListID> parent_left_id;
+						{
+							const auto& tmp_parent_left_id = list.list.at(idx_opt.value()).id;
+							parent_left_id = {list._actors[tmp_parent_left_id.actor_idx], tmp_parent_left_id.seq};
+						}
+
+						std::optional<List::ListID> parent_right_id;
+						if (idx_opt.value()+1 < list.list.size()) {
+							const auto& tmp_parent_right_id = list.list.at(idx_opt.value()+1).id;
+							parent_right_id = {list._actors[tmp_parent_right_id.actor_idx], tmp_parent_right_id.seq};
+						}
+
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							parent_left_id,
+							parent_right_id
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					}
+				} else {
+					// i think this is curser movement
+				}
+			} else if (j_op["action"] == "del") {
+				const auto list_id = JObjFromStr(static_cast<const std::string&>(j_op["key"]));
+				bool r = list.del({list_id.id, map_seq[list_id.id][list_id.seq]});
+				assert(r);
+				g_total_deletes++;
+			} else if (j_op["action"] == "makeText") {
+				// doc.clear();
+			} else if (j_op["action"] == "makeMap") {
+				// no idea
+			} else {
+				std::cout << "op: " << j_op << "\n";
+			}
+
+			op_seq++;
+		}
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list.list.size() << "\n";
+	std::cout << "doc size: " << list.doc_size << "\n";
+	std::cout << "total inserts: " << g_total_inserts << "\n";
+	std::cout << "total deletes: " << g_total_deletes << "\n";
+	std::cout << "total ops: " << g_total_inserts + g_total_deletes << "\n";
+
+	//std::cout << "find_hint: " << list._stat_find_with_hint << "\n";
+	//std::cout << "find_hint_hit: " << list._stat_find_with_hint_hit << "\n";
+
+	// checked, looks correct
+#if 0
+	std::cout << "doc text:\n";
+	// simple print
+	for (const auto& it : list.list) {
+		if (it.value) {
+			std::cout << it.value.value();
+		}
+	}
+	std::cout << "\n";
+#endif
+
+	return 0;
+}
+

bench/v3_jpaper.cpp

@@ -0,0 +1,211 @@
+#define EXTRA_ASSERTS 0
+
+#include <green_crdt/v3/list.hpp>
+#include <nlohmann/json.hpp>
+
+#include <unordered_map>
+#include <string_view>
+#include <fstream>
+#include <iostream>
+#include <cassert>
+
+using ActorID = std::array<uint8_t, 32>;
+using List = GreenCRDT::V3::List<char, ActorID>;
+
+template<>
+struct std::hash<ActorID> {
+	std::size_t operator()(ActorID const& s) const noexcept {
+		static_assert(sizeof(size_t) == 8);
+		// TODO: maybe shuffle the indices a bit
+		return
+			(static_cast<size_t>(s[0]) << 8*0) |
+			(static_cast<size_t>(s[1]) << 8*1) |
+			(static_cast<size_t>(s[2]) << 8*2) |
+			(static_cast<size_t>(s[3]) << 8*3) |
+			(static_cast<size_t>(s[4]) << 8*4) |
+			(static_cast<size_t>(s[5]) << 8*5) |
+			(static_cast<size_t>(s[6]) << 8*6) |
+			(static_cast<size_t>(s[7]) << 8*7)
+		;
+	}
+};
+
+// for dev, benching in debug is usefull, but only if the ammount of asserts is reasonable
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace detail {
+	uint8_t nib_from_hex(char c) {
+		extra_assert((c >= '0' && c <= '9') || (c >= 'a' && c <= 'f'));
+
+		if (c >= '0' && c <= '9') {
+			return static_cast<uint8_t>(c) - '0';
+		} else if (c >= 'a' && c <= 'f') {
+			return (static_cast<uint8_t>(c) - 'a') + 10u;
+		} else {
+			return 0u;
+		}
+	}
+} // detail
+
+static ActorID ActorIDFromStr(std::string_view str) {
+	extra_assert(str.size() == 32*2);
+	ActorID tmp;
+
+	for (size_t i = 0; i < tmp.size(); i++) {
+		tmp[i] = detail::nib_from_hex(str[i*2]) << 4 | detail::nib_from_hex(str[i*2+1]);
+	}
+
+	return tmp;
+}
+
+// seq@ID type format used in the json
+struct JObj {
+	ActorID id;
+	uint64_t seq {0};
+};
+
+static JObj JObjFromStr(std::string_view str) {
+	extra_assert(str.size() > 32*2 + 1);
+
+	size_t at_pos = str.find_first_of('@');
+	auto seq_sv = str.substr(0, at_pos);
+	auto id_sv = str.substr(at_pos+1);
+
+	assert(seq_sv.size() != 0);
+	assert(id_sv.size() == 32*2);
+
+	uint64_t tmp_seq {0};
+	for (size_t i = 0; i < seq_sv.size(); i++) {
+		assert(seq_sv[i] >= '0' && seq_sv[i] <= '9');
+		tmp_seq *= 10;
+		tmp_seq += seq_sv[i] - '0';
+	}
+
+	return {ActorIDFromStr(id_sv), tmp_seq};
+}
+
+int main(void) {
+	List list;
+
+	std::ifstream file {"../res/paper.json"};
+	std::cout << "start reading...\n";
+
+	uint64_t g_total_inserts {0};
+	uint64_t g_total_deletes {0};
+	//uint64_t g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, uint64_t> g_seq_inserts {0}; // the opsec are not sequentially growing for inserts, so we sidestep
+	std::unordered_map<ActorID, std::unordered_map<uint64_t, uint64_t>> map_seq; // maps json op_seq -> lits id seq
+
+	for (std::string line; std::getline(file, line); ) {
+		nlohmann::json j_entry = nlohmann::json::parse(line);
+		const ActorID actor = ActorIDFromStr(static_cast<const std::string&>(j_entry["actor"]));
+		const size_t actor_idx = list.findActor(actor).value_or(0u);
+		uint64_t op_seq = j_entry["startOp"];
+		for (const auto& j_op : j_entry["ops"]) {
+			if (j_op["action"] == "set") {
+				const auto obj = JObjFromStr(static_cast<const std::string&>(j_op["obj"]));
+				if (obj.seq != 1) {
+					// skip all non text edits (create text doc, curser etc)
+					continue;
+				}
+
+				if (j_op["insert"]) {
+					const auto& j_parent = j_op["key"];
+					extra_assert(!j_parent.is_null());
+					if (j_parent == "_head") {
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							std::nullopt,
+							std::nullopt
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					} else { // we have a parrent
+						extra_assert(static_cast<const std::string&>(j_op["value"]).size() == 1);
+
+						size_t hint_last_insert {0};
+						if (list._last_inserted_idx.count(actor_idx)) {
+							hint_last_insert = list._last_inserted_idx[actor_idx];
+						}
+
+						// split parent into seq and actor
+						const auto parent_left = JObjFromStr(static_cast<const std::string&>(j_parent));
+						auto idx_opt = list.findIdx({parent_left.id, map_seq[parent_left.id][parent_left.seq]}, hint_last_insert);
+						assert(idx_opt.has_value());
+
+						std::optional<List::ListID> parent_left_id;
+						{
+							const auto& tmp_parent_left_id = list._list_ids.at(idx_opt.value());
+							parent_left_id = {list._actors[tmp_parent_left_id.actor_idx], tmp_parent_left_id.seq};
+						}
+
+						std::optional<List::ListID> parent_right_id;
+						if (idx_opt.value()+1 < list._list_ids.size()) {
+							const auto& tmp_parent_right_id = list._list_ids.at(idx_opt.value()+1);
+							parent_right_id = {list._actors[tmp_parent_right_id.actor_idx], tmp_parent_right_id.seq};
+						}
+
+						uint64_t tmp_seq {g_seq_inserts[actor]++};
+						bool r = list.add(
+							{actor, tmp_seq},
+							static_cast<const std::string&>(j_op["value"]).front(),
+							parent_left_id,
+							parent_right_id
+						);
+						assert(r);
+						map_seq[actor][op_seq] = tmp_seq;
+						g_total_inserts++;
+					}
+				} else {
+					// i think this is curser movement
+				}
+			} else if (j_op["action"] == "del") {
+				const auto list_id = JObjFromStr(static_cast<const std::string&>(j_op["key"]));
+				bool r = list.del({list_id.id, map_seq[list_id.id][list_id.seq]});
+				assert(r);
+				g_total_deletes++;
+			} else if (j_op["action"] == "makeText") {
+				// doc.clear();
+			} else if (j_op["action"] == "makeMap") {
+				// no idea
+			} else {
+				std::cout << "op: " << j_op << "\n";
+			}
+
+			op_seq++;
+		}
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list._list_ids.size() << "\n";
+	std::cout << "doc size: " << list.getDocSize() << "\n";
+	std::cout << "total inserts: " << g_total_inserts << "\n";
+	std::cout << "total deletes: " << g_total_deletes << "\n";
+	std::cout << "total ops: " << g_total_inserts + g_total_deletes << "\n";
+
+	//std::cout << "find_hint: " << list._stat_find_with_hint << "\n";
+	//std::cout << "find_hint_hit: " << list._stat_find_with_hint_hit << "\n";
+
+	// checked, looks correct
+#if 0
+	std::cout << "doc text:\n";
+	// simple print
+	for (const auto& it : list.list) {
+		if (it.value) {
+			std::cout << it.value.value();
+		}
+	}
+	std::cout << "\n";
+#endif
+
+	return 0;
+}
+

version1/CMakeLists.txt

@@ -0,0 +1,26 @@
+cmake_minimum_required(VERSION 3.9 FATAL_ERROR)
+
+project(crdt_version1 CXX C)
+
+add_library(crdt_version1 INTERFACE)
+
+target_compile_features(crdt_version1 INTERFACE cxx_std_17)
+
+target_include_directories(crdt_version1 INTERFACE "${PROJECT_SOURCE_DIR}")
+
+########################################
+
+add_executable(v1_test1
+	./test1.cpp
+)
+
+target_link_libraries(v1_test1 PUBLIC crdt_version1)
+
+########################################
+
+#add_executable(v1_test2
+	#./test2.cpp
+#)
+
+#target_link_libraries(v1_test2 PUBLIC crdt_version1)
+

version1/green_crdt/v1/list.hpp

@@ -0,0 +1,328 @@
+#pragma once
+
+#include <cstdint>
+#include <optional>
+#include <vector>
+#include <map>
+#include <string>
+
+#include <cassert>
+
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace GreenCRDT::V1 {
+
+template<typename ValueType, typename ActorType>
+struct List {
+	// for public interface
+	struct ListID {
+		ActorType id;
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator<(const ListID& rhs) const {
+			if (seq < rhs.seq) {
+				return true;
+			} else if (seq > rhs.seq) {
+				return false;
+			} else { // ==
+				return id < rhs.id;
+			}
+		}
+
+		bool operator==(const ListID& rhs) const {
+			return seq == rhs.seq && id == rhs.id;
+		}
+
+		bool operator!=(const ListID& rhs) const {
+			return seq != rhs.seq || id != rhs.id;
+		}
+	};
+
+	struct ListIDInternal {
+		size_t actor_idx{0};
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator==(const ListIDInternal& rhs) const {
+			return seq == rhs.seq && actor_idx == rhs.actor_idx;
+		}
+	};
+
+	// internally the index into this array is used to refer to an actor
+	std::vector<ActorType> _actors;
+
+	// TODO: replace with SoA
+	struct Entry {
+		ListIDInternal id;
+
+		// Yjs
+		std::optional<ListIDInternal> parent_left;
+		std::optional<ListIDInternal> parent_right;
+
+		// might be deleted (yes, *sigh*, crtds need tombstones)
+		std::optional<ValueType> value;
+	};
+
+	// TODO: use something better, edit: this seems fine
+	std::vector<Entry> list;
+
+	// number of not deleted entries
+	size_t doc_size {0};
+
+	std::map<size_t, uint64_t> last_seen_seq;
+
+	std::optional<size_t> findActor(const ActorType& actor) const {
+		for (size_t i = 0; i < _actors.size(); i++) {
+			if (_actors[i] == actor) {
+				return i;
+			}
+		}
+		return std::nullopt;
+	}
+
+	std::optional<size_t> findIdx(const ListIDInternal& list_id) const {
+		extra_assert(verify());
+
+		for (size_t i = 0; i < list.size(); i++) {
+			if (list[i].id == list_id) {
+				return i;
+			}
+		}
+
+		return std::nullopt;
+	}
+
+	std::optional<size_t> findIdx(const ListID& list_id) const {
+		extra_assert(verify());
+
+		const auto actor_idx_opt = findActor(list_id.id);
+		if (!actor_idx_opt.has_value()) {
+			return std::nullopt;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), list_id.seq};
+
+		for (size_t i = 0; i < list.size(); i++) {
+			if (list[i].id == tmp_id) {
+				return i;
+			}
+		}
+
+		return std::nullopt;
+	}
+
+	// returns false if missing OPs
+	// based on YjsMod https://github.com/josephg/reference-crdts/blob/9f4f9c3a97b497e2df8ae4473d1e521d3c3bf2d2/crdts.ts#L293-L348
+	// which is a modified Yjs(YATA) algo
+	bool add(const ListID& list_id, const ValueType& value, const std::optional<ListID>& parent_left, const std::optional<ListID>& parent_right) {
+		extra_assert(verify());
+
+		size_t actor_idx {0};
+		{ // new actor?
+			// add, even if op fails
+			const auto actor_opt = findActor(list_id.id);
+			if (!actor_opt.has_value()) {
+				actor_idx = _actors.size();
+				_actors.push_back(list_id.id);
+			} else {
+				actor_idx = actor_opt.value();
+			}
+		}
+
+		// check actor op order
+		if (!last_seen_seq.count(actor_idx)) {
+			// we dont know this actor yet, first seq needs to be 0
+			if (list_id.seq != 0) {
+				return false;
+			}
+		} else {
+			// making sure we dont skip operations by that actor
+			if (list_id.seq != last_seen_seq.at(actor_idx) + 1) {
+				return false;
+			}
+		}
+
+		size_t insert_idx = 0;
+		if (list.empty()) {
+			if (parent_left.has_value() || parent_right.has_value()) {
+				// empty, missing parents
+				return false;
+			}
+		} else {
+			// find left
+			std::optional<size_t> left_idx = std::nullopt;
+			if (parent_left.has_value()) {
+				left_idx = findIdx(parent_left.value());
+				if (!left_idx.has_value()) {
+					// missing parent left
+					return false;
+				}
+
+				// we insert before the it, so we need to go past the left parent
+				insert_idx = left_idx.value() + 1;
+			} // else insert_idx = 0
+
+			// find right
+			size_t right_idx = list.size();
+			if (parent_right.has_value()) {
+				auto tmp_right = findIdx(parent_right.value());
+				if (!tmp_right.has_value()) {
+					return false;
+				}
+				right_idx = tmp_right.value();
+			}
+
+			bool scanning {false};
+
+			for(size_t i = insert_idx;; i++) {
+				if (!scanning) {
+					insert_idx = i;
+				}
+				// if right parent / end of doc, insert
+				if (insert_idx == right_idx) {
+					break;
+				}
+				// we ran past right o.o ?
+				if (insert_idx == list.size()) {
+					break;
+				}
+
+				const Entry& at_i = list[i];
+				// parents left and right
+				std::optional<size_t> i_left_idx {std::nullopt};
+				if (at_i.parent_left.has_value()) {
+					i_left_idx = findIdx(at_i.parent_left.value());
+					if (!i_left_idx.has_value()) {
+						assert(false && "item in list with unknown parent left!!");
+						return false;
+					}
+				}
+
+				// possibility map
+				//
+				//         | ir < r | ir == r       | ir > r
+				// -------------------------------------
+				// il < l  | insert | insert        | insert
+				// il == l | ?      | agentfallback | ?
+				// il > l  | skip   | skip          | skip
+
+				if (i_left_idx < left_idx) {
+					break;
+				} else if (i_left_idx == left_idx) {
+					// get i parent_right
+					size_t i_right_idx = list.size();
+					if (at_i.parent_right.has_value()) {
+						auto tmp_right = findIdx(at_i.parent_right.value());
+						if (!tmp_right.has_value()) {
+							assert(false && "item in list with unknown parent right!!");
+							return false;
+						}
+						i_right_idx = tmp_right.value();
+					}
+
+					if (i_right_idx < right_idx) {
+						scanning = true;
+					} else if (i_right_idx == right_idx) {
+						// actor id tie breaker
+						if (_actors[actor_idx] < _actors[at_i.id.actor_idx]) {
+							break;
+						} else {
+							scanning = false;
+						}
+					} else { // i_right_idx > right_idx
+						scanning = false;
+					}
+				} else { // il > l
+					// do nothing
+				}
+			}
+		}
+
+		{ // actual insert
+			Entry new_entry;
+
+			new_entry.id.actor_idx = actor_idx;
+			new_entry.id.seq = list_id.seq;
+
+			if (parent_left.has_value()) {
+				new_entry.parent_left = ListIDInternal{findActor(parent_left.value().id).value(), parent_left.value().seq};
+			}
+
+			if (parent_right.has_value()) {
+				new_entry.parent_right = ListIDInternal{findActor(parent_right.value().id).value(), parent_right.value().seq};
+			}
+
+			new_entry.value = value;
+
+			list.emplace(list.begin() + insert_idx, new_entry);
+		}
+
+		doc_size++;
+		last_seen_seq[actor_idx] = list_id.seq;
+
+		extra_assert(verify());
+		return true;
+	}
+
+	// returns false if not found
+	bool del(const ListID& id) {
+		extra_assert(verify());
+
+		auto actor_idx_opt = findActor(id.id);
+		if (!actor_idx_opt.has_value()) {
+			// we dont have anything with that actor
+			return false;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), id.seq};
+
+		for (auto& it : list) {
+			if (it.id == tmp_id) {
+				if (it.value.has_value()) {
+					it.value.reset();
+
+					doc_size--;
+					extra_assert(verify());
+					return true;
+				} else {
+					extra_assert(verify());
+					return false; // TODO: allow double deletes?,,,, need ids
+				}
+			}
+		}
+
+		extra_assert(verify());
+		return false;
+	}
+
+	std::vector<ValueType> getArray(void) const {
+		std::vector<ValueType> array;
+		for (const auto& e : list) {
+			if (e.value.has_value()) {
+				array.push_back(e.value.value());
+			}
+		}
+
+		return array;
+	}
+
+	// TODO: only in debug?
+	bool verify(void) const {
+		size_t actual_size = 0;
+		for (const auto& it : list) {
+			if (it.value.has_value()) {
+				actual_size++;
+			}
+		}
+		//assert(doc_size == actual_size);
+		return doc_size == actual_size;
+	}
+};
+
+} // GreenCRDT::V1
+

version1/test1.cpp

@@ -0,0 +1,214 @@
+#define EXTRA_ASSERTS 1
+#include <green_crdt/v1/list.hpp>
+
+#include <numeric>
+#include <random>
+#include <iostream>
+#include <cassert>
+#include <string_view>
+#include <vector>
+
+// single letter actor, for testing only
+using Actor = char;
+using ListType = GreenCRDT::V1::List<char, Actor>;
+
+namespace std {
+bool operator==(const std::vector<char>& lhs, const std::string_view& rhs) {
+	if (lhs.size() != rhs.size()) {
+		return false;
+	}
+
+	for (size_t i = 0; i < rhs.size(); i++) {
+		if (lhs[i] != rhs[i]) {
+			return false;
+		}
+	}
+
+	return true;
+}
+} // namespace std
+
+void testSingle1(void) {
+	ListType list;
+
+	assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+	assert(list.add({'A', 1}, 'b', ListType::ListID{'A', 0u}, std::nullopt));
+
+	assert(list.getArray() == "ab");
+}
+
+
+void testConcurrent1(void) {
+	// agent_a < agent_b
+
+	// concurrent insert of first element
+	{ // variant 1, a then b
+		ListType list;
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+	{ // variant 2, b then a
+		ListType list;
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+}
+
+struct AddOp {
+	ListType::ListID id;
+	char value;
+	std::optional<ListType::ListID> parent_left;
+	std::optional<ListType::ListID> parent_right;
+};
+
+void randomAddPermutations(const std::vector<AddOp>& ops, const std::string& expected) {
+	// TODO: more then 1k?
+	for (size_t i = 0; i < 1000; i++) {
+		std::minstd_rand rng(1337 + i);
+		std::vector<size_t> ops_todo(ops.size());
+		std::iota(ops_todo.begin(), ops_todo.end(), 0u);
+
+		size_t attempts {0};
+
+		ListType list;
+		do {
+			size_t idx = rng() % ops_todo.size();
+
+			if (list.add(ops[ops_todo[idx]].id, ops[ops_todo[idx]].value, ops[ops_todo[idx]].parent_left, ops[ops_todo[idx]].parent_right)) {
+				// only remove if it was possible -> returned true;
+				ops_todo.erase(ops_todo.begin()+idx);
+			}
+
+			attempts++;
+			assert(attempts < 10'000); // in case we run into an endless loop
+		} while (!ops_todo.empty());
+
+		assert(list.getArray() == expected);
+	}
+}
+
+void testInterleave1(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', ListType::ListID{'A', 0u}, std::nullopt},
+		{{'A', 2u}, 'a', ListType::ListID{'A', 1u}, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', ListType::ListID{'B', 0u}, std::nullopt},
+		{{'B', 2u}, 'b', ListType::ListID{'B', 1u}, std::nullopt},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testInterleave2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', std::nullopt, ListType::ListID{'A', 0u}},
+		{{'A', 2u}, 'a', std::nullopt, ListType::ListID{'A', 1u}},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', std::nullopt, ListType::ListID{'B', 0u}},
+		{{'B', 2u}, 'b', std::nullopt, ListType::ListID{'B', 1u}},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testConcurrent2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'C', 0u}, 'c', std::nullopt, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'D', 0u}, 'd', ListType::ListID{'A', 0u}, ListType::ListID{'C', 0u}},
+	};
+
+	randomAddPermutations(ops, "adbc");
+}
+
+void testMain1(void) {
+	ListType list;
+
+	static_assert('0' < '1');
+
+	const std::vector<AddOp> a0_ops {
+		{{'0', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'0', 1u}, 'b', ListType::ListID{'0', 0u}, std::nullopt},
+		{{'0', 2u}, 'c', ListType::ListID{'0', 1u}, std::nullopt},
+		{{'0', 3u}, 'd', ListType::ListID{'0', 1u}, ListType::ListID{'0', 2u}},
+	};
+
+	const std::vector<AddOp> a1_ops {
+		// knows of a0 up to {a0, 1}
+		{{'1', 0u}, 'z', ListType::ListID{'0', 0u}, ListType::ListID{'0', 1u}},
+		{{'1', 1u}, 'y', ListType::ListID{'0', 1u}, std::nullopt},
+	};
+
+	{ // the ez, in order stuff
+		// a0 insert first char, 'a', since its the first, we dont have any parents
+		assert(list.add(a0_ops[0].id, a0_ops[0].value, a0_ops[0].parent_left, a0_ops[0].parent_right));
+		assert(list.getArray() == "a");
+
+		// a0 insert secound char, 'b' after 'a', no parents to right
+		assert(list.add(a0_ops[1].id, a0_ops[1].value, a0_ops[1].parent_left, a0_ops[1].parent_right));
+		assert(list.getArray() == "ab");
+
+		// a0 insert 'c' after 'b', no parents to right
+		assert(list.add(a0_ops[2].id, a0_ops[2].value, a0_ops[2].parent_left, a0_ops[2].parent_right));
+		assert(list.getArray() == "abc");
+
+		// a0 insert 'd' after 'b', 'c' parent right
+		assert(list.add(a0_ops[3].id, a0_ops[3].value, a0_ops[3].parent_left, a0_ops[3].parent_right));
+		assert(list.getArray() == "abdc");
+
+		// a1 insert 'z' after 'a', 'b' parent right
+		assert(list.add(a1_ops[0].id, a1_ops[0].value, a1_ops[0].parent_left, a1_ops[0].parent_right));
+		assert(list.getArray() == "azbdc");
+	}
+
+	std::cout << "done with ez\n";
+
+	{ // a1 was not uptodate only had 0,1 of a0
+		// a1 insert 'y' after 'b', no parent right
+		assert(list.add(a1_ops[1].id, a1_ops[1].value, a1_ops[1].parent_left, a1_ops[1].parent_right));
+		assert(list.getArray() == "azbdcy");
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list.list.size() << "\n";
+	std::cout << "\ndoc size: " << list.doc_size << "\n";
+	std::cout << "doc text:\n";
+
+	const auto tmp_array = list.getArray();
+	std::cout << std::string_view(tmp_array.data(), tmp_array.size()) << "\n";
+}
+
+int main(void) {
+	std::cout << "testSingle1:\n";
+	testSingle1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent1:\n";
+	testConcurrent1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave1:\n";
+	testInterleave1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave2:\n";
+	testInterleave2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent2:\n";
+	testConcurrent2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testMain1:\n";
+	testMain1();
+	std::cout << std::string(40, '-') << "\n";
+
+	return 0;
+}
+

version2/CMakeLists.txt

@@ -0,0 +1,26 @@
+cmake_minimum_required(VERSION 3.9 FATAL_ERROR)
+
+project(crdt_version2 CXX C)
+
+add_library(crdt_version2 INTERFACE)
+
+target_compile_features(crdt_version2 INTERFACE cxx_std_17)
+
+target_include_directories(crdt_version2 INTERFACE "${PROJECT_SOURCE_DIR}")
+
+########################################
+
+add_executable(v2_test1
+	./test1.cpp
+)
+
+target_link_libraries(v2_test1 PUBLIC crdt_version2)
+
+########################################
+
+#add_executable(v2_test2
+	#./test2.cpp
+#)
+
+#target_link_libraries(v2_test2 PUBLIC crdt_version2)
+

version2/green_crdt/v2/list.hpp

@@ -0,0 +1,385 @@
+#pragma once
+
+#include <cstdint>
+#include <optional>
+#include <unordered_map>
+#include <vector>
+#include <string>
+
+#include <cassert>
+
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace GreenCRDT::V2 {
+
+template<typename ValueType, typename ActorType>
+struct List {
+	// for public interface
+	struct ListID {
+		ActorType id;
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator<(const ListID& rhs) const {
+			if (seq < rhs.seq) {
+				return true;
+			} else if (seq > rhs.seq) {
+				return false;
+			} else { // ==
+				return id < rhs.id;
+			}
+		}
+
+		bool operator==(const ListID& rhs) const {
+			return seq == rhs.seq && id == rhs.id;
+		}
+
+		bool operator!=(const ListID& rhs) const {
+			return seq != rhs.seq || id != rhs.id;
+		}
+	};
+
+	struct ListIDInternal {
+		size_t actor_idx{0};
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator==(const ListIDInternal& rhs) const {
+			return seq == rhs.seq && actor_idx == rhs.actor_idx;
+		}
+	};
+
+	// internally the index into this array is used to refer to an actor
+	std::vector<ActorType> _actors;
+
+	// TODO: replace with SoA
+	struct Entry {
+		ListIDInternal id;
+
+		// Yjs
+		std::optional<ListIDInternal> parent_left;
+		std::optional<ListIDInternal> parent_right;
+
+		// might be deleted (yes, *sigh*, crtds need tombstones)
+		std::optional<ValueType> value;
+	};
+
+	// TODO: use something better, edit: this seems fine
+	std::vector<Entry> list;
+
+	// number of not deleted entries
+	size_t doc_size {0};
+
+	// TODO: actor index instead of map
+	std::unordered_map<size_t, uint64_t> last_seen_seq;
+
+	// caching only, contains the last index an actor inserted at
+	std::unordered_map<size_t, size_t> last_inserted_idx;
+
+	//size_t _stat_find_with_hint{0};
+	//size_t _stat_find_with_hint_hit{0};
+
+	std::optional<size_t> findActor(const ActorType& actor) const {
+		for (size_t i = 0; i < _actors.size(); i++) {
+			if (_actors[i] == actor) {
+				return i;
+			}
+		}
+		return std::nullopt;
+	}
+
+	std::optional<size_t> findIdx(const ListIDInternal& list_id) const {
+		extra_assert(verify());
+
+		for (size_t i = 0; i < list.size(); i++) {
+			if (list[i].id == list_id) {
+				return i;
+			}
+		}
+
+		return std::nullopt;
+	}
+
+	// search close to hint first
+	std::optional<size_t> findIdx(const ListIDInternal& list_id, size_t hint) const {
+		extra_assert(verify());
+
+		//_stat_find_with_hint++;
+
+		// TODO: find some good magic values here
+		// total: 364150
+		// 2-9 hits: 360164 (3m54)
+		// 1-9 hits: 360161 (3m53)
+		// 1-2 hits: 359800 (3m55s)
+		// 0-2 hits: 359763 (3m54s)
+		// changed from loop to single if:
+		// 1-2 hits: 359800 (3m50s)
+		// 1-4 hits: 359928 (3m51s) (after cond reorder: 3m49s)
+		static constexpr size_t c_hint_pre = 1;
+		static constexpr size_t c_hint_post = 4;
+
+		{ // go back 2, so we dont miss // TODO: is this really needed
+			//for (size_t i = 0; hint > 0 && i < c_hint_pre; hint--, i++) {}
+			if (hint >= c_hint_pre) {
+				hint -= c_hint_pre;
+			}
+		}
+
+		const size_t max_at_hint = hint + c_hint_post; // how many positions we check at hint, before falling back to full lookup
+
+		for (size_t i = hint; i <= max_at_hint && i < list.size(); i++) {
+			if (list[i].id == list_id) {
+				//_stat_find_with_hint_hit++;
+				return i;
+			}
+		}
+
+		// fall back to normal search
+		// TODO: in some cases we scan the list twice now!!
+		return findIdx(list_id);
+	}
+
+	std::optional<size_t> findIdx(const ListID& list_id) const {
+		extra_assert(verify());
+
+		const auto actor_idx_opt = findActor(list_id.id);
+		if (!actor_idx_opt.has_value()) {
+			return std::nullopt;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), list_id.seq};
+
+		return findIdx(tmp_id);
+	}
+
+	std::optional<size_t> findIdx(const ListID& list_id, size_t hint) const {
+		extra_assert(verify());
+
+		const auto actor_idx_opt = findActor(list_id.id);
+		if (!actor_idx_opt.has_value()) {
+			return std::nullopt;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), list_id.seq};
+
+		return findIdx(tmp_id, hint);
+	}
+
+	// returns false if missing OPs
+	// based on YjsMod https://github.com/josephg/reference-crdts/blob/9f4f9c3a97b497e2df8ae4473d1e521d3c3bf2d2/crdts.ts#L293-L348
+	// which is a modified Yjs(YATA) algo
+	// TODO: idx_hint
+	bool add(const ListID& list_id, const ValueType& value, const std::optional<ListID>& parent_left, const std::optional<ListID>& parent_right) {
+		extra_assert(verify());
+
+		size_t actor_idx {0};
+		{ // new actor?
+			// add, even if op fails
+			const auto actor_opt = findActor(list_id.id);
+			if (!actor_opt.has_value()) {
+				actor_idx = _actors.size();
+				last_inserted_idx[_actors.size()] = 0; // hack
+				_actors.push_back(list_id.id);
+			} else {
+				actor_idx = actor_opt.value();
+			}
+		}
+
+		// check actor op order
+		if (!last_seen_seq.count(actor_idx)) {
+			// we dont know this actor yet, first seq needs to be 0
+			if (list_id.seq != 0) {
+				return false;
+			}
+		} else {
+			// making sure we dont skip operations by that actor
+			if (list_id.seq != last_seen_seq.at(actor_idx) + 1) {
+				return false;
+			}
+		}
+
+		size_t insert_idx = 0;
+		if (list.empty()) {
+			if (parent_left.has_value() || parent_right.has_value()) {
+				// empty, missing parents
+				return false;
+			}
+		} else {
+			// find left
+			std::optional<size_t> left_idx_opt = std::nullopt;
+			if (parent_left.has_value()) {
+				left_idx_opt = findIdx(parent_left.value(), last_inserted_idx[actor_idx]);
+				if (!left_idx_opt.has_value()) {
+					// missing parent left
+					return false;
+				}
+
+				// we insert before the it, so we need to go past the left parent
+				insert_idx = left_idx_opt.value() + 1;
+			} // else insert_idx = 0
+			const size_t left_idx_hint = insert_idx;
+
+			// find right
+			size_t right_idx = list.size();
+			if (parent_right.has_value()) {
+				auto tmp_right = findIdx(parent_right.value(), left_idx_hint);
+				if (!tmp_right.has_value()) {
+					return false;
+				}
+				right_idx = tmp_right.value();
+			}
+
+			bool scanning {false};
+
+			for(size_t i = insert_idx;; i++) {
+				if (!scanning) {
+					insert_idx = i;
+				}
+				// if right parent / end of doc, insert
+				if (insert_idx == right_idx) {
+					break;
+				}
+				// we ran past right o.o ?
+				if (insert_idx == list.size()) {
+					break;
+				}
+
+				const Entry& at_i = list[i];
+				// parents left and right
+				std::optional<size_t> i_left_idx {std::nullopt};
+				if (at_i.parent_left.has_value()) {
+					i_left_idx = findIdx(at_i.parent_left.value(), left_idx_hint);
+					if (!i_left_idx.has_value()) {
+						assert(false && "item in list with unknown parent left!!");
+						return false;
+					}
+				}
+
+				// possibility map
+				//
+				//         | ir < r | ir == r       | ir > r
+				// -------------------------------------
+				// il < l  | insert | insert        | insert
+				// il == l | ?      | agentfallback | ?
+				// il > l  | skip   | skip          | skip
+
+				if (i_left_idx < left_idx_opt) {
+					break;
+				} else if (i_left_idx == left_idx_opt) {
+					// get i parent_right
+					size_t i_right_idx = list.size();
+					if (at_i.parent_right.has_value()) {
+						auto tmp_right = findIdx(at_i.parent_right.value(), insert_idx);
+						if (!tmp_right.has_value()) {
+							assert(false && "item in list with unknown parent right!!");
+							return false;
+						}
+						i_right_idx = tmp_right.value();
+					}
+
+					if (i_right_idx < right_idx) {
+						scanning = true;
+					} else if (i_right_idx == right_idx) {
+						// actor id tie breaker
+						if (_actors[actor_idx] < _actors[at_i.id.actor_idx]) {
+							break;
+						} else {
+							scanning = false;
+						}
+					} else { // i_right_idx > right_idx
+						scanning = false;
+					}
+				} else { // il > l
+					// do nothing
+				}
+			}
+		}
+
+		{ // actual insert
+			Entry new_entry;
+
+			new_entry.id.actor_idx = actor_idx;
+			new_entry.id.seq = list_id.seq;
+
+			if (parent_left.has_value()) {
+				new_entry.parent_left = ListIDInternal{findActor(parent_left.value().id).value(), parent_left.value().seq};
+			}
+
+			if (parent_right.has_value()) {
+				new_entry.parent_right = ListIDInternal{findActor(parent_right.value().id).value(), parent_right.value().seq};
+			}
+
+			new_entry.value = value;
+
+			list.emplace(list.begin() + insert_idx, new_entry);
+			last_inserted_idx[actor_idx] = insert_idx;
+		}
+
+		doc_size++;
+		last_seen_seq[actor_idx] = list_id.seq;
+
+		extra_assert(verify());
+		return true;
+	}
+
+	// returns false if not found
+	bool del(const ListID& id) {
+		extra_assert(verify());
+
+		auto actor_idx_opt = findActor(id.id);
+		if (!actor_idx_opt.has_value()) {
+			// we dont have anything with that actor
+			return false;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), id.seq};
+
+		for (auto& it : list) {
+			if (it.id == tmp_id) {
+				if (it.value.has_value()) {
+					it.value.reset();
+
+					doc_size--;
+					extra_assert(verify());
+					return true;
+				} else {
+					extra_assert(verify());
+					return false; // TODO: allow double deletes?,,,, need ids
+				}
+			}
+		}
+
+		extra_assert(verify());
+		return false;
+	}
+
+	std::vector<ValueType> getArray(void) const {
+		std::vector<ValueType> array;
+		for (const auto& e : list) {
+			if (e.value.has_value()) {
+				array.push_back(e.value.value());
+			}
+		}
+
+		return array;
+	}
+
+	// TODO: only in debug?
+	bool verify(void) const {
+		size_t actual_size = 0;
+		for (const auto& it : list) {
+			if (it.value.has_value()) {
+				actual_size++;
+			}
+		}
+		//assert(doc_size == actual_size);
+		return doc_size == actual_size;
+	}
+};
+
+} // GreenCRDT::V1
+

version2/test1.cpp

@@ -0,0 +1,214 @@
+#define EXTRA_ASSERTS 1
+#include <green_crdt/v2/list.hpp>
+
+#include <numeric>
+#include <random>
+#include <iostream>
+#include <cassert>
+#include <string_view>
+#include <vector>
+
+// single letter actor, for testing only
+using Actor = char;
+using ListType = GreenCRDT::V2::List<char, Actor>;
+
+namespace std {
+bool operator==(const std::vector<char>& lhs, const std::string_view& rhs) {
+	if (lhs.size() != rhs.size()) {
+		return false;
+	}
+
+	for (size_t i = 0; i < rhs.size(); i++) {
+		if (lhs[i] != rhs[i]) {
+			return false;
+		}
+	}
+
+	return true;
+}
+} // namespace std
+
+void testSingle1(void) {
+	ListType list;
+
+	assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+	assert(list.add({'A', 1}, 'b', ListType::ListID{'A', 0u}, std::nullopt));
+
+	assert(list.getArray() == "ab");
+}
+
+
+void testConcurrent1(void) {
+	// agent_a < agent_b
+
+	// concurrent insert of first element
+	{ // variant 1, a then b
+		ListType list;
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+	{ // variant 2, b then a
+		ListType list;
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+}
+
+struct AddOp {
+	ListType::ListID id;
+	char value;
+	std::optional<ListType::ListID> parent_left;
+	std::optional<ListType::ListID> parent_right;
+};
+
+void randomAddPermutations(const std::vector<AddOp>& ops, const std::string& expected) {
+	// TODO: more then 1k?
+	for (size_t i = 0; i < 1000; i++) {
+		std::minstd_rand rng(1337 + i);
+		std::vector<size_t> ops_todo(ops.size());
+		std::iota(ops_todo.begin(), ops_todo.end(), 0u);
+
+		size_t attempts {0};
+
+		ListType list;
+		do {
+			size_t idx = rng() % ops_todo.size();
+
+			if (list.add(ops[ops_todo[idx]].id, ops[ops_todo[idx]].value, ops[ops_todo[idx]].parent_left, ops[ops_todo[idx]].parent_right)) {
+				// only remove if it was possible -> returned true;
+				ops_todo.erase(ops_todo.begin()+idx);
+			}
+
+			attempts++;
+			assert(attempts < 10'000); // in case we run into an endless loop
+		} while (!ops_todo.empty());
+
+		assert(list.getArray() == expected);
+	}
+}
+
+void testInterleave1(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', ListType::ListID{'A', 0u}, std::nullopt},
+		{{'A', 2u}, 'a', ListType::ListID{'A', 1u}, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', ListType::ListID{'B', 0u}, std::nullopt},
+		{{'B', 2u}, 'b', ListType::ListID{'B', 1u}, std::nullopt},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testInterleave2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', std::nullopt, ListType::ListID{'A', 0u}},
+		{{'A', 2u}, 'a', std::nullopt, ListType::ListID{'A', 1u}},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', std::nullopt, ListType::ListID{'B', 0u}},
+		{{'B', 2u}, 'b', std::nullopt, ListType::ListID{'B', 1u}},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testConcurrent2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'C', 0u}, 'c', std::nullopt, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'D', 0u}, 'd', ListType::ListID{'A', 0u}, ListType::ListID{'C', 0u}},
+	};
+
+	randomAddPermutations(ops, "adbc");
+}
+
+void testMain1(void) {
+	ListType list;
+
+	static_assert('0' < '1');
+
+	const std::vector<AddOp> a0_ops {
+		{{'0', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'0', 1u}, 'b', ListType::ListID{'0', 0u}, std::nullopt},
+		{{'0', 2u}, 'c', ListType::ListID{'0', 1u}, std::nullopt},
+		{{'0', 3u}, 'd', ListType::ListID{'0', 1u}, ListType::ListID{'0', 2u}},
+	};
+
+	const std::vector<AddOp> a1_ops {
+		// knows of a0 up to {a0, 1}
+		{{'1', 0u}, 'z', ListType::ListID{'0', 0u}, ListType::ListID{'0', 1u}},
+		{{'1', 1u}, 'y', ListType::ListID{'0', 1u}, std::nullopt},
+	};
+
+	{ // the ez, in order stuff
+		// a0 insert first char, 'a', since its the first, we dont have any parents
+		assert(list.add(a0_ops[0].id, a0_ops[0].value, a0_ops[0].parent_left, a0_ops[0].parent_right));
+		assert(list.getArray() == "a");
+
+		// a0 insert secound char, 'b' after 'a', no parents to right
+		assert(list.add(a0_ops[1].id, a0_ops[1].value, a0_ops[1].parent_left, a0_ops[1].parent_right));
+		assert(list.getArray() == "ab");
+
+		// a0 insert 'c' after 'b', no parents to right
+		assert(list.add(a0_ops[2].id, a0_ops[2].value, a0_ops[2].parent_left, a0_ops[2].parent_right));
+		assert(list.getArray() == "abc");
+
+		// a0 insert 'd' after 'b', 'c' parent right
+		assert(list.add(a0_ops[3].id, a0_ops[3].value, a0_ops[3].parent_left, a0_ops[3].parent_right));
+		assert(list.getArray() == "abdc");
+
+		// a1 insert 'z' after 'a', 'b' parent right
+		assert(list.add(a1_ops[0].id, a1_ops[0].value, a1_ops[0].parent_left, a1_ops[0].parent_right));
+		assert(list.getArray() == "azbdc");
+	}
+
+	std::cout << "done with ez\n";
+
+	{ // a1 was not uptodate only had 0,1 of a0
+		// a1 insert 'y' after 'b', no parent right
+		assert(list.add(a1_ops[1].id, a1_ops[1].value, a1_ops[1].parent_left, a1_ops[1].parent_right));
+		assert(list.getArray() == "azbdcy");
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list.list.size() << "\n";
+	std::cout << "\ndoc size: " << list.doc_size << "\n";
+	std::cout << "doc text:\n";
+
+	const auto tmp_array = list.getArray();
+	std::cout << std::string_view(tmp_array.data(), tmp_array.size()) << "\n";
+}
+
+int main(void) {
+	std::cout << "testSingle1:\n";
+	testSingle1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent1:\n";
+	testConcurrent1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave1:\n";
+	testInterleave1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave2:\n";
+	testInterleave2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent2:\n";
+	testConcurrent2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testMain1:\n";
+	testMain1();
+	std::cout << std::string(40, '-') << "\n";
+
+	return 0;
+}
+

version3/CMakeLists.txt

@@ -0,0 +1,26 @@
+cmake_minimum_required(VERSION 3.9 FATAL_ERROR)
+
+project(crdt_version3 CXX C)
+
+add_library(crdt_version3 INTERFACE)
+
+target_compile_features(crdt_version3 INTERFACE cxx_std_17)
+
+target_include_directories(crdt_version3 INTERFACE "${PROJECT_SOURCE_DIR}")
+
+########################################
+
+add_executable(v3_test1
+	./test1.cpp
+)
+
+target_link_libraries(v3_test1 PUBLIC crdt_version3)
+
+########################################
+
+#add_executable(v3_test2
+	#./test2.cpp
+#)
+
+#target_link_libraries(v3_test2 PUBLIC crdt_version3)
+

version3/green_crdt/v3/list.hpp

@@ -0,0 +1,406 @@
+#pragma once
+
+#include <cstdint>
+#include <optional>
+#include <unordered_map>
+#include <vector>
+#include <string>
+
+#include <cassert>
+
+#if !defined(extra_assert)
+	#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
+		#define extra_assert(...) assert(__VA_ARGS__)
+	#else
+		#define extra_assert(...) void(0)
+	#endif
+#endif
+
+namespace GreenCRDT::V3 {
+
+template<typename ValueType, typename ActorType>
+struct List {
+	// for public interface
+	struct ListID {
+		ActorType id;
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator<(const ListID& rhs) const {
+			if (seq < rhs.seq) {
+				return true;
+			} else if (seq > rhs.seq) {
+				return false;
+			} else { // ==
+				return id < rhs.id;
+			}
+		}
+
+		bool operator==(const ListID& rhs) const {
+			return seq == rhs.seq && id == rhs.id;
+		}
+
+		bool operator!=(const ListID& rhs) const {
+			return seq != rhs.seq || id != rhs.id;
+		}
+	};
+
+	struct ListIDInternal {
+		size_t actor_idx{0};
+		uint64_t seq{0}; // strictly increasing for that actor
+
+		bool operator==(const ListIDInternal& rhs) const {
+			return seq == rhs.seq && actor_idx == rhs.actor_idx;
+		}
+	};
+
+	// internally the index into this array is used to refer to an actor
+	std::vector<ActorType> _actors;
+
+	struct Entry_Data {
+		// Yjs
+		std::optional<ListIDInternal> parent_left;
+		std::optional<ListIDInternal> parent_right;
+
+		// might be deleted (yes, *sigh*, crtds need tombstones)
+		std::optional<ValueType> value;
+	};
+
+	std::vector<ListIDInternal> _list_ids;
+	std::vector<Entry_Data> _list_data;
+
+	// number of not deleted entries
+	size_t _doc_size {0};
+
+	// TODO: actor index instead of map
+	std::unordered_map<size_t, uint64_t> _last_seen_seq;
+
+	// caching only, contains the last index an actor inserted at
+	std::unordered_map<size_t, size_t> _last_inserted_idx;
+
+	//size_t _stat_find_with_hint{0};
+	//size_t _stat_find_with_hint_hit{0};
+
+	std::optional<size_t> findActor(const ActorType& actor) const {
+		for (size_t i = 0; i < _actors.size(); i++) {
+			if (_actors[i] == actor) {
+				return i;
+			}
+		}
+		return std::nullopt;
+	}
+
+	std::optional<size_t> findIdx(const ListIDInternal& list_id) const {
+		extra_assert(verify());
+
+		for (size_t i = 0; i < _list_ids.size(); i++) {
+			if (_list_ids[i] == list_id) {
+				return i;
+			}
+		}
+
+		return std::nullopt;
+	}
+
+	// search close to hint first
+	std::optional<size_t> findIdx(const ListIDInternal& list_id, size_t hint) const {
+		extra_assert(verify());
+
+		//_stat_find_with_hint++;
+
+		// TODO: find some good magic values here
+		// total: 364150
+		// 2-9 hits: 360164 (3m54)
+		// 1-9 hits: 360161 (3m53)
+		// 1-2 hits: 359800 (3m55s)
+		// 0-2 hits: 359763 (3m54s)
+		// changed from loop to single if:
+		// 1-2 hits: 359800 (3m50s)
+		// 1-4 hits: 359928 (3m51s) (after cond reorder: 3m49s)
+		static constexpr size_t c_hint_pre = 1;
+		static constexpr size_t c_hint_post = 4;
+
+		{ // go back 2, so we dont miss // TODO: is this really needed
+			//for (size_t i = 0; hint > 0 && i < c_hint_pre; hint--, i++) {}
+			if (hint >= c_hint_pre) {
+				hint -= c_hint_pre;
+			}
+		}
+
+		const size_t max_at_hint = hint + c_hint_post; // how many positions we check at hint, before falling back to full lookup
+
+		for (size_t i = hint; i <= max_at_hint && i < _list_ids.size(); i++) {
+			if (_list_ids[i] == list_id) {
+				//_stat_find_with_hint_hit++;
+				return i;
+			}
+		}
+
+		// fall back to normal search
+		// TODO: in some cases we scan the list twice now!!
+		return findIdx(list_id);
+	}
+
+	std::optional<size_t> findIdx(const ListID& list_id) const {
+		extra_assert(verify());
+
+		const auto actor_idx_opt = findActor(list_id.id);
+		if (!actor_idx_opt.has_value()) {
+			return std::nullopt;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), list_id.seq};
+
+		return findIdx(tmp_id);
+	}
+
+	std::optional<size_t> findIdx(const ListID& list_id, size_t hint) const {
+		extra_assert(verify());
+
+		const auto actor_idx_opt = findActor(list_id.id);
+		if (!actor_idx_opt.has_value()) {
+			return std::nullopt;
+		}
+
+		const ListIDInternal tmp_id {actor_idx_opt.value(), list_id.seq};
+
+		return findIdx(tmp_id, hint);
+	}
+
+	// returns false if missing OPs
+	// based on YjsMod https://github.com/josephg/reference-crdts/blob/9f4f9c3a97b497e2df8ae4473d1e521d3c3bf2d2/crdts.ts#L293-L348
+	// which is a modified Yjs(YATA) algo
+	// TODO: idx_hint
+	bool add(const ListID& list_id, const ValueType& value, const std::optional<ListID>& parent_left, const std::optional<ListID>& parent_right) {
+		extra_assert(verify());
+
+		size_t actor_idx {0};
+		{ // new actor?
+			// add, even if op fails
+			const auto actor_opt = findActor(list_id.id);
+			if (!actor_opt.has_value()) {
+				actor_idx = _actors.size();
+				_last_inserted_idx[_actors.size()] = 0; // hack
+				_actors.push_back(list_id.id);
+			} else {
+				actor_idx = actor_opt.value();
+			}
+		}
+
+		// check actor op order
+		if (!_last_seen_seq.count(actor_idx)) {
+			// we dont know this actor yet, first seq needs to be 0
+			if (list_id.seq != 0) {
+				return false;
+			}
+		} else {
+			// making sure we dont skip operations by that actor
+			if (list_id.seq != _last_seen_seq.at(actor_idx) + 1) {
+				return false;
+			}
+		}
+
+		size_t insert_idx = 0;
+		if (_list_ids.empty()) {
+			if (parent_left.has_value() || parent_right.has_value()) {
+				// empty, missing parents
+				return false;
+			}
+		} else {
+			// find left
+			std::optional<size_t> left_idx_opt = std::nullopt;
+			if (parent_left.has_value()) {
+				left_idx_opt = findIdx(parent_left.value(), _last_inserted_idx[actor_idx]);
+				if (!left_idx_opt.has_value()) {
+					// missing parent left
+					return false;
+				}
+
+				// we insert before the it, so we need to go past the left parent
+				insert_idx = left_idx_opt.value() + 1;
+			} // else insert_idx = 0
+			const size_t left_idx_hint = insert_idx;
+
+			// find right
+			size_t right_idx = _list_ids.size();
+			if (parent_right.has_value()) {
+				auto tmp_right = findIdx(parent_right.value(), left_idx_hint);
+				if (!tmp_right.has_value()) {
+					return false;
+				}
+				right_idx = tmp_right.value();
+			}
+
+			bool scanning {false};
+
+			for(size_t i = insert_idx;; i++) {
+				if (!scanning) {
+					insert_idx = i;
+				}
+				// if right parent / end of doc, insert
+				if (insert_idx == right_idx) {
+					break;
+				}
+				// we ran past right o.o ?
+				if (insert_idx == _list_ids.size()) {
+					break;
+				}
+
+				const Entry_Data& at_i = _list_data[i];
+				// parents left and right
+				std::optional<size_t> i_left_idx {std::nullopt};
+				if (at_i.parent_left.has_value()) {
+					i_left_idx = findIdx(at_i.parent_left.value(), left_idx_hint);
+					if (!i_left_idx.has_value()) {
+						assert(false && "item in list with unknown parent left!!");
+						return false;
+					}
+				}
+
+				// possibility map
+				//
+				//         | ir < r | ir == r       | ir > r
+				// -------------------------------------
+				// il < l  | insert | insert        | insert
+				// il == l | ?      | agentfallback | ?
+				// il > l  | skip   | skip          | skip
+
+				if (i_left_idx < left_idx_opt) {
+					break;
+				} else if (i_left_idx == left_idx_opt) {
+					// get i parent_right
+					size_t i_right_idx = _list_ids.size();
+					if (at_i.parent_right.has_value()) {
+						auto tmp_right = findIdx(at_i.parent_right.value(), insert_idx);
+						if (!tmp_right.has_value()) {
+							assert(false && "item in list with unknown parent right!!");
+							return false;
+						}
+						i_right_idx = tmp_right.value();
+					}
+
+					if (i_right_idx < right_idx) {
+						scanning = true;
+					} else if (i_right_idx == right_idx) {
+						// actor id tie breaker
+						if (_actors[actor_idx] < _actors[_list_ids[i].actor_idx]) {
+							break;
+						} else {
+							scanning = false;
+						}
+					} else { // i_right_idx > right_idx
+						scanning = false;
+					}
+				} else { // il > l
+					// do nothing
+				}
+			}
+		}
+
+		{ // actual insert
+			_list_ids.emplace(_list_ids.begin() + insert_idx, ListIDInternal{actor_idx, list_id.seq});
+
+			Entry_Data new_entry;
+			if (parent_left.has_value()) {
+				new_entry.parent_left = ListIDInternal{findActor(parent_left.value().id).value(), parent_left.value().seq};
+			}
+
+			if (parent_right.has_value()) {
+				new_entry.parent_right = ListIDInternal{findActor(parent_right.value().id).value(), parent_right.value().seq};
+			}
+
+			new_entry.value = value;
+
+			_list_data.emplace(_list_data.begin() + insert_idx, new_entry);
+			_last_inserted_idx[actor_idx] = insert_idx;
+		}
+
+		_doc_size++;
+		_last_seen_seq[actor_idx] = list_id.seq;
+
+		extra_assert(verify());
+		return true;
+	}
+
+	// returns false if not found
+	bool del(const ListID& id) {
+		extra_assert(verify());
+
+		auto actor_idx_opt = findActor(id.id);
+		if (!actor_idx_opt.has_value()) {
+			// we dont have anything with that actor
+			return false;
+		}
+
+#if 0
+		const ListIDInternal tmp_id {actor_idx_opt.value(), id.seq};
+		for (auto& it : list) {
+			if (it.id == tmp_id) {
+				if (it.value.has_value()) {
+					it.value.reset();
+
+					doc_size--;
+					extra_assert(verify());
+					return true;
+				} else {
+					extra_assert(verify());
+					return false; // TODO: allow double deletes?,,,, need ids
+				}
+			}
+		}
+#endif
+		// TODO: actually test deletes lol
+		const auto idx_opt = findIdx(id);
+		if (idx_opt.has_value()) {
+			auto& it = _list_data[idx_opt.value()];
+			if (it.value.has_value()) {
+				it.value.reset();
+
+				_doc_size--;
+				extra_assert(verify());
+				return true;
+			} else {
+				extra_assert(verify());
+				return false; // TODO: allow double deletes?,,,, need ids
+			}
+		}
+
+		// not found
+		extra_assert(verify());
+		return false;
+	}
+
+	// returns the size of alive entries
+	size_t getDocSize(void) const {
+		return _doc_size;
+	}
+
+	std::vector<ValueType> getArray(void) const {
+		std::vector<ValueType> array;
+		for (const auto& e : _list_data) {
+			if (e.value.has_value()) {
+				array.push_back(e.value.value());
+			}
+		}
+
+		return array;
+	}
+
+	// TODO: only in debug?
+	bool verify(void) const {
+		if (_list_ids.size() != _list_data.size()) {
+			return false;
+		}
+
+		size_t actual_size = 0;
+		for (const auto& it : _list_data) {
+			if (it.value.has_value()) {
+				actual_size++;
+			}
+		}
+		//assert(doc_size == actual_size);
+		return _doc_size == actual_size;
+	}
+};
+
+} // GreenCRDT::V1
+

version3/test1.cpp

@@ -0,0 +1,214 @@
+#define EXTRA_ASSERTS 1
+#include <green_crdt/v3/list.hpp>
+
+#include <numeric>
+#include <random>
+#include <iostream>
+#include <cassert>
+#include <string_view>
+#include <vector>
+
+// single letter actor, for testing only
+using Actor = char;
+using ListType = GreenCRDT::V3::List<char, Actor>;
+
+namespace std {
+bool operator==(const std::vector<char>& lhs, const std::string_view& rhs) {
+	if (lhs.size() != rhs.size()) {
+		return false;
+	}
+
+	for (size_t i = 0; i < rhs.size(); i++) {
+		if (lhs[i] != rhs[i]) {
+			return false;
+		}
+	}
+
+	return true;
+}
+} // namespace std
+
+void testSingle1(void) {
+	ListType list;
+
+	assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+	assert(list.add({'A', 1}, 'b', ListType::ListID{'A', 0u}, std::nullopt));
+
+	assert(list.getArray() == "ab");
+}
+
+
+void testConcurrent1(void) {
+	// agent_a < agent_b
+
+	// concurrent insert of first element
+	{ // variant 1, a then b
+		ListType list;
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+	{ // variant 2, b then a
+		ListType list;
+		assert(list.add({'B', 0}, 'b', std::nullopt, std::nullopt));
+		assert(list.add({'A', 0}, 'a', std::nullopt, std::nullopt));
+
+		assert(list.getArray() == "ab");
+	}
+}
+
+struct AddOp {
+	ListType::ListID id;
+	char value;
+	std::optional<ListType::ListID> parent_left;
+	std::optional<ListType::ListID> parent_right;
+};
+
+void randomAddPermutations(const std::vector<AddOp>& ops, const std::string& expected) {
+	// TODO: more then 1k?
+	for (size_t i = 0; i < 1000; i++) {
+		std::minstd_rand rng(1337 + i);
+		std::vector<size_t> ops_todo(ops.size());
+		std::iota(ops_todo.begin(), ops_todo.end(), 0u);
+
+		size_t attempts {0};
+
+		ListType list;
+		do {
+			size_t idx = rng() % ops_todo.size();
+
+			if (list.add(ops[ops_todo[idx]].id, ops[ops_todo[idx]].value, ops[ops_todo[idx]].parent_left, ops[ops_todo[idx]].parent_right)) {
+				// only remove if it was possible -> returned true;
+				ops_todo.erase(ops_todo.begin()+idx);
+			}
+
+			attempts++;
+			assert(attempts < 10'000); // in case we run into an endless loop
+		} while (!ops_todo.empty());
+
+		assert(list.getArray() == expected);
+	}
+}
+
+void testInterleave1(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', ListType::ListID{'A', 0u}, std::nullopt},
+		{{'A', 2u}, 'a', ListType::ListID{'A', 1u}, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', ListType::ListID{'B', 0u}, std::nullopt},
+		{{'B', 2u}, 'b', ListType::ListID{'B', 1u}, std::nullopt},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testInterleave2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'A', 1u}, 'a', std::nullopt, ListType::ListID{'A', 0u}},
+		{{'A', 2u}, 'a', std::nullopt, ListType::ListID{'A', 1u}},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'B', 1u}, 'b', std::nullopt, ListType::ListID{'B', 0u}},
+		{{'B', 2u}, 'b', std::nullopt, ListType::ListID{'B', 1u}},
+	};
+
+	randomAddPermutations(ops, "aaabbb");
+}
+
+void testConcurrent2(void) {
+	const std::vector<AddOp> ops {
+		{{'A', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'C', 0u}, 'c', std::nullopt, std::nullopt},
+		{{'B', 0u}, 'b', std::nullopt, std::nullopt},
+		{{'D', 0u}, 'd', ListType::ListID{'A', 0u}, ListType::ListID{'C', 0u}},
+	};
+
+	randomAddPermutations(ops, "adbc");
+}
+
+void testMain1(void) {
+	ListType list;
+
+	static_assert('0' < '1');
+
+	const std::vector<AddOp> a0_ops {
+		{{'0', 0u}, 'a', std::nullopt, std::nullopt},
+		{{'0', 1u}, 'b', ListType::ListID{'0', 0u}, std::nullopt},
+		{{'0', 2u}, 'c', ListType::ListID{'0', 1u}, std::nullopt},
+		{{'0', 3u}, 'd', ListType::ListID{'0', 1u}, ListType::ListID{'0', 2u}},
+	};
+
+	const std::vector<AddOp> a1_ops {
+		// knows of a0 up to {a0, 1}
+		{{'1', 0u}, 'z', ListType::ListID{'0', 0u}, ListType::ListID{'0', 1u}},
+		{{'1', 1u}, 'y', ListType::ListID{'0', 1u}, std::nullopt},
+	};
+
+	{ // the ez, in order stuff
+		// a0 insert first char, 'a', since its the first, we dont have any parents
+		assert(list.add(a0_ops[0].id, a0_ops[0].value, a0_ops[0].parent_left, a0_ops[0].parent_right));
+		assert(list.getArray() == "a");
+
+		// a0 insert secound char, 'b' after 'a', no parents to right
+		assert(list.add(a0_ops[1].id, a0_ops[1].value, a0_ops[1].parent_left, a0_ops[1].parent_right));
+		assert(list.getArray() == "ab");
+
+		// a0 insert 'c' after 'b', no parents to right
+		assert(list.add(a0_ops[2].id, a0_ops[2].value, a0_ops[2].parent_left, a0_ops[2].parent_right));
+		assert(list.getArray() == "abc");
+
+		// a0 insert 'd' after 'b', 'c' parent right
+		assert(list.add(a0_ops[3].id, a0_ops[3].value, a0_ops[3].parent_left, a0_ops[3].parent_right));
+		assert(list.getArray() == "abdc");
+
+		// a1 insert 'z' after 'a', 'b' parent right
+		assert(list.add(a1_ops[0].id, a1_ops[0].value, a1_ops[0].parent_left, a1_ops[0].parent_right));
+		assert(list.getArray() == "azbdc");
+	}
+
+	std::cout << "done with ez\n";
+
+	{ // a1 was not uptodate only had 0,1 of a0
+		// a1 insert 'y' after 'b', no parent right
+		assert(list.add(a1_ops[1].id, a1_ops[1].value, a1_ops[1].parent_left, a1_ops[1].parent_right));
+		assert(list.getArray() == "azbdcy");
+	}
+
+	std::cout << "\ndoc size (with tombstones): " << list._list_ids.size() << "\n";
+	std::cout << "\ndoc size: " << list.getDocSize() << "\n";
+	std::cout << "doc text:\n";
+
+	const auto tmp_array = list.getArray();
+	std::cout << std::string_view(tmp_array.data(), tmp_array.size()) << "\n";
+}
+
+int main(void) {
+	std::cout << "testSingle1:\n";
+	testSingle1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent1:\n";
+	testConcurrent1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave1:\n";
+	testInterleave1();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testInterleave2:\n";
+	testInterleave2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testConcurrent2:\n";
+	testConcurrent2();
+	std::cout << std::string(40, '-') << "\n";
+
+	std::cout << "testMain1:\n";
+	testMain1();
+	std::cout << std::string(40, '-') << "\n";
+
+	return 0;
+}
+