add jpaper bench for v0

CMakeLists.txt

@@ -39,5 +39,7 @@ add_subdirectory(./prototyping EXCLUDE_FROM_ALL)
 
 add_subdirectory(./version0)
 
+add_subdirectory(./bench)
+
 add_subdirectory(./vim_research)
 

bench/CMakeLists.txt

@@ -0,0 +1,28 @@
+cmake_minimum_required(VERSION 3.9 FATAL_ERROR)
+
+project(crdt_bench CXX C)
+
+########################################
+
+add_executable(crdt_bench_jpaper_v0
+	./v0_jpaper.cpp
+)
+
+target_link_libraries(crdt_bench_jpaper_v0 PUBLIC
+	crdt_version0
+	nlohmann_json::nlohmann_json
+)
+
+########################################
+
+if (FALSE)
+add_executable(crdt_bench_jpaper_v1
+	./test1.cpp
+)
+
+target_link_libraries(crdt_bench_jpaper_v1 PUBLIC
+	crdt_version1
+	nlohmann_json::nlohmann_json
+)
+endif()
+

bench/README.md

@@ -0,0 +1,38 @@
+# Timings
+
+all benches use the uncompressed .json from disk
+
+all done with the "cool an breezy copy" preset
+
+note: the .json is not a perfect fit, bc it is not designed for the Yjs algo (missing parent_right and ids are not perfect)
+
+the json contains:
+	doc size (with tombstones): 182315
+	doc size: 104852
+	total inserts: 182315
+	total deletes: 77463
+	total ops: 259778
+
+## baseline ( just walking through the json, no insertions )
+
+- g++9 -g :
+	- 23.0s
+	- 22.6s
+	- 23.0s
+
+- g++9 -O3 -DNDEBUG :
+	- 9.6s
+	- 9.7s
+	- 9.7s
+
+
+## version0
+
+- g++9 -g -O2 :
+	- 10m35s
+
+- g++9 -O3 -DNDEBUG :
+	- 8m7s
+
+## version1
+

bench/v0_jpaper.cpp

@@ -0,0 +1,198 @@
+#define EXTRA_ASSERTS 0
+
+//#include <green_crdt/v0/text_document.hpp>
+#include <green_crdt/v0/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::V0::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());
+
+						const auto parent_left_id = list.list.at(idx_opt.value()).id;
+						std::optional<List::ListID> parent_right_id;
+						if (idx_opt.value()+1 < list.list.size()) {
+							parent_right_id = list.list.at(idx_opt.value()+1).id;
+						}
+
+						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;
+}
+

version0/CMakeLists.txt

@@ -10,17 +10,17 @@ target_include_directories(crdt_version0 INTERFACE "${PROJECT_SOURCE_DIR}")
 
 ########################################
 
-add_executable(test1
+add_executable(v0_test1
 	./test1.cpp
 )
 
-target_link_libraries(test1 PUBLIC crdt_version0)
+target_link_libraries(v0_test1 PUBLIC crdt_version0)
 
 ########################################
 
-add_executable(test2
+add_executable(v0_test2
 	./test2.cpp
 )
 
-target_link_libraries(test2 PUBLIC crdt_version0)
+target_link_libraries(v0_test2 PUBLIC crdt_version0)
 

version0/green_crdt/v0/list.hpp

@@ -8,7 +8,15 @@
 
 #include <cassert>
 
-namespace GreenCRDT {
+#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::V0 {
 
 template<typename ValueType, typename AgentType>
 struct List {
@@ -71,7 +79,9 @@ struct List {
 	std::map<AgentType, uint64_t> last_seen_seq;
 
 	std::optional<size_t> findIdx(const ListID& list_id) const {
+//#if defined(EXTRA_ASSERTS) && EXTRA_ASSERTS == 1
 		//verify(); // too expensive
+//#endif
 		for (size_t i = 0; i < list.size(); i++) {
 			if (list[i].id == list_id) {
 				return i;
@@ -243,5 +253,5 @@ struct List {
 	}
 };
 
-} // GreenCRDT
+} // GreenCRDT::V0
 

version0/green_crdt/v0/text_document.hpp

@@ -6,7 +6,7 @@
 
 //#include <iostream> // debug
 
-namespace GreenCRDT {
+namespace GreenCRDT::V0 {
 
 template<typename AgentType>
 struct TextDocument {
@@ -281,5 +281,5 @@ struct TextDocument {
 	}
 };
 
-} // GreenCRDT
+} // GreenCRDT::V0
 

version0/test1.cpp

@@ -1,5 +1,5 @@
-#include <crdt/list.hpp>
+#include <green_crdt/v0/list.hpp>
-#include <crdt/text_document.hpp>
+#include <green_crdt/v0/text_document.hpp>
 
 #include <numeric>
 #include <random>
@@ -8,7 +8,7 @@
 
 // single letter agent, for testing only
 using Agent = char;
-using DocType = GreenCRDT::TextDocument<Agent>;
+using DocType = GreenCRDT::V0::TextDocument<Agent>;
 using ListType = DocType::ListType;
 
 void testSingle1(void) {

version0/test2.cpp

@@ -1,4 +1,4 @@
-#include <crdt/text_document.hpp>
+#include <green_crdt/v0/text_document.hpp>
 
 #include <numeric>
 #include <optional>
@@ -9,8 +9,8 @@
 
 // single letter agent, for testing only
 using Agent = char;
-using Doc = GreenCRDT::TextDocument<Agent>;
+using Doc = GreenCRDT::V0::TextDocument<Agent>;
-using Op = GreenCRDT::TextDocument<Agent>::Op;
+using Op = Doc::Op;
 using ListType = Doc::ListType;
 
 // maybe switch it up?

vim_research/test1.cpp

@@ -1,4 +1,4 @@
-#include <crdt/text_document.hpp>
+#include <green_crdt/v0/text_document.hpp>
 #include <nlohmann/json.hpp>
 
 extern "C" {
@@ -15,7 +15,7 @@ extern "C" {
 // single letter agent, for testing only
 //using Agent = char;
 using Agent = uint16_t; // tmp local port
-using Doc = GreenCRDT::TextDocument<Agent>;
+using Doc = GreenCRDT::V0::TextDocument<Agent>;
 using ListType = Doc::ListType;
 
 std::ostream& operator<<(std::ostream& out, const std::optional<ListType::ListID>& id) {

vim_research/test2.cpp

@@ -1,4 +1,4 @@
-#include <crdt/text_document.hpp>
+#include <green_crdt/v0/text_document.hpp>
 #include <nlohmann/json.hpp>
 
 extern "C" {
@@ -59,7 +59,7 @@ struct std::hash<ToxPubKey> {
 //using Agent = char;
 //using Agent = uint16_t; // tmp local port
 using Agent = ToxPubKey;
-using Doc = GreenCRDT::TextDocument<Agent>;
+using Doc = GreenCRDT::V0::TextDocument<Agent>;
 using ListType = Doc::ListType;
 
 struct Command {
@@ -89,7 +89,7 @@ namespace std {
 	}
 } // namespace std
 
-namespace GreenCRDT {
+namespace GreenCRDT::V0 {
 
 NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE(ListType::ListID,
 	id,
@@ -107,7 +107,7 @@ NLOHMANN_DEFINE_TYPE_NON_INTRUSIVE(ListType::OpDel,
 	id
 )
 
-} // namespace GreenCRDT
+} // namespace GreenCRDT::V0
 
 // bc variant <.<
 namespace std {
@@ -235,7 +235,7 @@ function! GreenCRDTCheckTimeAndFetch()
 			endif
 		endif
 
-		let b:green_crdt_fetch_timer = timer_start(503, 'GreenCRDTFetchTimerCallback')
+		let b:green_crdt_fetch_timer = timer_start(203, 'GreenCRDTFetchTimerCallback')
 	endif
 endfunction
 )"
@@ -287,7 +287,7 @@ delfunction GreenCRDTSetupEvents
 )"
 
 R"(
-let b:green_crdt_fetch_timer = timer_start(900, 'GreenCRDTFetchTimerCallback')
+let b:green_crdt_fetch_timer = timer_start(300, 'GreenCRDTFetchTimerCallback')
 
 echo 'setup done'
 )");