Optimize transaction command parsing to avoid unnecessary tokenization

Previously, the parser always tokenized the full command, even when we only
needed to check whether it was a transaction command. Now, it first extracts
the first word to determine relevance and performs full tokenization only
when necessary.

include/PgSQL_ExplicitTxnStateMgr.h

@@ -56,6 +56,9 @@ struct TxnCmd {
  */
 class PgSQL_TxnCmdParser {
 public:
+    PgSQL_TxnCmdParser() noexcept { tokens.reserve(16); }
+    ~PgSQL_TxnCmdParser() noexcept = default;
+
     TxnCmd parse(std::string_view input, bool in_transaction_mode) noexcept;
 
 private:
@@ -67,14 +70,20 @@ private:
     TxnCmd parse_start(size_t& pos) noexcept;
 
     // Helpers
-    static std::string to_lower(std::string_view s) noexcept {
-		std::string s_copy(s);
-        std::transform(s_copy.begin(), s_copy.end(), s_copy.begin(), ::tolower);
-        return s_copy;
+    inline static bool iequals(std::string_view a, std::string_view b) noexcept {
+        if (a.size() != b.size()) return false;
+        for (size_t i = 0; i < a.size(); ++i) {
+            char ca = a[i];
+            char cb = b[i];
+            if (ca >= 'A' && ca <= 'Z') ca += 32;
+            if (cb >= 'A' && cb <= 'Z') cb += 32;
+            if (ca != cb) return false;
+        }
+        return true;
     }
 
     inline static bool contains(std::vector<std::string_view>&& list, std::string_view value) noexcept {
-        for (const auto& item : list) if (item == value) return true;
+        for (const auto& item : list) if (iequals(item, value)) return true;
         return false;
     }
 };

lib/PgSQL_ExplicitTxnStateMgr.cpp

@@ -327,16 +327,97 @@ bool PgSQL_ExplicitTxnStateMgr::handle_transaction(std::string_view input) {
 	return true;
 }
 
-
 TxnCmd PgSQL_TxnCmdParser::parse(std::string_view input, bool in_transaction_mode) noexcept {
-    tokens.clear();
     TxnCmd cmd;
-    bool in_quote = false;
+
+    if (input.empty()) return cmd;
+
+    // Extract first word without full tokenization
     size_t start = 0;
+    size_t end = 0;
+
+    while (start < input.size() && fast_isspace(input[start])) {
+        start++;
+    }
+
+    if (start >= input.size()) return cmd;
+
+    // Find end of first word
+    end = start;
+    bool in_quote = false;
     char quote_char = 0;
 
-    // Tokenize with quote handling
-    for (size_t i = 0; i <= input.size(); ++i) {
+    while (end < input.size()) {
+        char c = input[end];
+
+        if (!in_quote && (c == '"' || c == '\'')) {
+            // If we hit a quote at the start, this isn't a transaction command
+            return cmd;
+        }
+
+        if (fast_isspace(c) || c == ';') {
+            break;
+        }
+
+        end++;
+    }
+
+    std::string_view first_word = input.substr(start, end - start);
+
+    // Check if this is a transaction command we care about
+	TxnCmd::Type cmd_type = TxnCmd::UNKNOWN;
+
+    if (in_transaction_mode) {
+        if (iequals(first_word, "begin")) {
+            cmd.type = TxnCmd::BEGIN;
+            return cmd;
+        }
+        
+        if (iequals(first_word, "start")) {
+			cmd_type = TxnCmd::BEGIN;
+		} else if (iequals(first_word, "savepoint")) {
+			cmd_type = TxnCmd::SAVEPOINT;
+		} else if (iequals(first_word, "release")) {
+			cmd_type = TxnCmd::RELEASE;
+		} else if (iequals(first_word, "rollback")) {
+			cmd_type = TxnCmd::ROLLBACK;
+		}
+    } else {
+
+        if (iequals(first_word, "commit") || iequals(first_word, "end")) {
+            cmd.type = TxnCmd::COMMIT;
+			return cmd;
+        }
+
+        if (iequals(first_word, "abort")) {
+            cmd.type = TxnCmd::ROLLBACK;
+			return cmd;
+        }
+
+        if (iequals(first_word, "rollback")) {
+            cmd_type = TxnCmd::ROLLBACK;
+        }
+    }
+
+    // If not a transaction command, return early
+    if (cmd_type == TxnCmd::UNKNOWN) {
+        return cmd;
+    }
+
+    // Continue tokenization from where we left off
+    tokens.clear();
+
+    // Continue tokenizing the rest of the input
+    in_quote = false;
+    quote_char = 0;
+    start = end; // Continue from after the first word
+
+    while (start < input.size() && fast_isspace(input[start])) {
+        start++;
+    }
+
+    // Tokenize the remaining input
+    for (size_t i = start; i <= input.size(); ++i) {
         const bool at_end = i == input.size();
         const char c = at_end ? 0 : input[i];
 
@@ -344,6 +425,7 @@ TxnCmd PgSQL_TxnCmdParser::parse(std::string_view input, bool in_transaction_mod
             if (c == quote_char || at_end) {
                 tokens.emplace_back(input.substr(start + 1, i - start - 1));
                 in_quote = false;
+                start = i + 1;
             }
             continue;
         }
@@ -353,41 +435,50 @@ TxnCmd PgSQL_TxnCmdParser::parse(std::string_view input, bool in_transaction_mod
             quote_char = c;
             start = i;
         }
-        else if (isspace(c) || c == ';' || at_end) {
+        else if (fast_isspace(c) || c == ';' || at_end) {
             if (start < i) tokens.emplace_back(input.substr(start, i - start));
             start = i + 1;
         }
     }
 
-    if (tokens.empty()) return cmd;
-
     size_t pos = 0;
-    const std::string first = to_lower(tokens[pos++]);
-
-    if (in_transaction_mode == true) {
-        if (first == "begin") cmd.type = TxnCmd::BEGIN;
-		else if (first == "start") cmd = parse_start(pos);
-        else if (first == "savepoint") cmd = parse_savepoint(pos);
-        else if (first == "release") cmd = parse_release(pos);
-        else if (first == "rollback") cmd = parse_rollback(pos);
+
+    if (in_transaction_mode) {
+
+		switch (cmd_type) {
+		case TxnCmd::BEGIN:
+			cmd = parse_start(pos);
+            break;
+		case TxnCmd::SAVEPOINT:
+            cmd = parse_savepoint(pos);
+			break;
+        case TxnCmd::RELEASE:
+			cmd = parse_release(pos);
+            break;
+		case TxnCmd::ROLLBACK:
+            cmd = parse_rollback(pos);
+			break;
+		default:
+			break;
+		}
     } else {
-        if (first == "commit" || first == "end") cmd.type = TxnCmd::COMMIT;
-		else if (first == "abort") cmd.type = TxnCmd::ROLLBACK;
-        else if (first == "rollback") cmd = parse_rollback(pos);
+        if (cmd_type == TxnCmd::ROLLBACK)
+            cmd = parse_rollback(pos);
     }
+
     return cmd;
 }
 
 TxnCmd PgSQL_TxnCmdParser::parse_rollback(size_t& pos) noexcept {
     TxnCmd cmd{ TxnCmd::ROLLBACK };
-    while (pos < tokens.size() && contains({ "work", "transaction" }, to_lower(tokens[pos]))) pos++;
+    while (pos < tokens.size() && contains({ "work", "transaction" }, tokens[pos])) pos++;
 
-    if (pos < tokens.size() && to_lower(tokens[pos]) == "to") {
+    if (pos < tokens.size() && iequals(tokens[pos], "to")) {
         cmd.type = TxnCmd::ROLLBACK_TO;
-        if (++pos < tokens.size() && to_lower(tokens[pos]) == "savepoint") pos++;
+        if (++pos < tokens.size() && iequals(tokens[pos], "savepoint")) pos++;
         if (pos < tokens.size()) cmd.savepoint = tokens[pos++];
-    } else if (pos < tokens.size() && to_lower(tokens[pos]) == "and") {
-        if (++pos < tokens.size() && to_lower(tokens[pos]) == "chain") {
+    } else if (pos < tokens.size() && iequals(tokens[pos], "and")) {
+        if (++pos < tokens.size() && iequals(tokens[pos], "chain")) {
             cmd.type = TxnCmd::ROLLBACK_AND_CHAIN;
 			pos++;
         }
@@ -403,14 +494,14 @@ TxnCmd PgSQL_TxnCmdParser::parse_savepoint(size_t& pos) noexcept {
 
 TxnCmd PgSQL_TxnCmdParser::parse_release(size_t& pos) noexcept {
     TxnCmd cmd{ TxnCmd::RELEASE };
-    if (pos < tokens.size() && to_lower(tokens[pos]) == "savepoint") pos++;
+    if (pos < tokens.size() && iequals(tokens[pos], "savepoint")) pos++;
     if (pos < tokens.size()) cmd.savepoint = tokens[pos++];
     return cmd;
 }
 
 TxnCmd PgSQL_TxnCmdParser::parse_start(size_t& pos) noexcept {
     TxnCmd cmd{ TxnCmd::UNKNOWN };
-    if (pos < tokens.size() && to_lower(tokens[pos]) == "transaction") {
+    if (pos < tokens.size() && iequals(tokens[pos], "transaction")) {
         cmd.type = TxnCmd::BEGIN;
 		pos++;
     }

lib/PgSQL_Protocol.cpp

@@ -741,7 +741,7 @@ std::vector<std::pair<std::string, std::string>> PgSQL_Protocol::parse_options(c
 
 	while (pos < input.size()) {
 		// Skip leading spaces
-		while (pos < input.size() && std::isspace(input[pos])) {
+		while (pos < input.size() && fast_isspace(input[pos])) {
 			++pos;
 		}
 
@@ -751,7 +751,7 @@ std::vector<std::pair<std::string, std::string>> PgSQL_Protocol::parse_options(c
 			pos += 2; // Skip "-c", "--"
 		}
 
-		while (pos < input.size() && std::isspace(input[pos])) {
+		while (pos < input.size() && fast_isspace(input[pos])) {
 			++pos;
 		}
 
@@ -772,7 +772,7 @@ std::vector<std::pair<std::string, std::string>> PgSQL_Protocol::parse_options(c
 		bool last_was_escape = false;
 		while (pos < input.size()) {
 			char c = input[pos];
-			if (std::isspace(c) && !last_was_escape) {
+			if (fast_isspace(c) && !last_was_escape) {
 				break;
 			}
 			if (c == '\\' && !last_was_escape) {

lib/PgSQL_Session.cpp

@@ -6557,7 +6557,7 @@ std::vector<std::string> PgSQL_DateStyle_Util::split_datestyle(std::string_view
 			int* lastNonSpace = (currentToken == 1) ? &lastNonSpace1 : &lastNonSpace2;
 
 			// Cache is-space check.
-			bool is_space = std::isspace(static_cast<unsigned char>(c));
+			bool is_space = fast_isspace(static_cast<unsigned char>(c));
 			// Skip leading whitespace for a new token.
 			if (currentStr->empty() && is_space) {
 				continue;

lib/PgSQL_Variables_Validator.cpp

@@ -183,7 +183,7 @@ bool pgsql_variable_validate_maintenance_work_mem(const char* value, const param
 	if (transformed_value) *transformed_value = nullptr;
 
 	// Skip leading whitespace
-	while (isspace((unsigned char)*p)) p++;
+	while (fast_isspace((unsigned char)*p)) p++;
 
 	// Parse numeric part
 	num = strtoll(p, &endptr, 10);
@@ -196,11 +196,11 @@ bool pgsql_variable_validate_maintenance_work_mem(const char* value, const param
 	p = endptr;
 
 	// Skip whitespace after number
-	while (isspace((unsigned char)*p)) p++;
+	while (fast_isspace((unsigned char)*p)) p++;
 
 	// Parse unit
 	if (*p != '\0') {
-		char tmp_unit = tolower(*p);
+		char tmp_unit = ::tolower(*p);
 		switch (tmp_unit) {
 		case 'k': 
 		case 'm': 
@@ -210,7 +210,7 @@ bool pgsql_variable_validate_maintenance_work_mem(const char* value, const param
 				unit = toupper(*p++);
 			has_unit = true;
 			// Check optional 'b'/'B'
-			if (tolower(*p) == 'b') p++;
+			if (::tolower(*p) == 'b') p++;
 			break;
 		default:
 			return false;
@@ -218,7 +218,7 @@ bool pgsql_variable_validate_maintenance_work_mem(const char* value, const param
 	}
 
 	// Skip trailing whitespace
-	while (isspace((unsigned char)*p)) p++;
+	while (fast_isspace((unsigned char)*p)) p++;
 
 	// Validate entire string consumed
 	if (*p != '\0') return false;
@@ -241,7 +241,7 @@ bool pgsql_variable_validate_maintenance_work_mem_v2(const char* value, const pa
 	const char* input = value;
 
 	/* Trim leading whitespace */
-	while (isspace((unsigned char)*input)) input++;
+	while (fast_isspace((unsigned char)*input)) input++;
 
 	/* Parse numeric part */
 	uint64_t number;
@@ -256,7 +256,7 @@ bool pgsql_variable_validate_maintenance_work_mem_v2(const char* value, const pa
 	//num_len = endptr - input;
 
 	// Skip whitespace after number
-	while (isspace((unsigned char)*endptr)) endptr++;
+	while (fast_isspace((unsigned char)*endptr)) endptr++;
 
 	/* Parse unit part */
 	const char* unit_ptr = endptr;
@@ -273,7 +273,7 @@ bool pgsql_variable_validate_maintenance_work_mem_v2(const char* value, const pa
 		/* Convert unit to lowercase for validation */
 		char u[3] = { 0 };
 		for (int i = 0; i < 2 && unit_ptr[i]; i++)
-			u[i] = tolower((unsigned char)unit_ptr[i]);
+			u[i] = ::tolower((unsigned char)unit_ptr[i]);
 
 		/* Validate unit and set multiplier */
 		if (unit_len == 1 && u[0] == 'b') {
@@ -332,7 +332,7 @@ bool pgsql_variable_validate_maintenance_work_mem_v3(const char* value, const pa
 	(void)session;
 
 	// Trim leading whitespace
-	while (isspace((unsigned char)*value)) value++;
+	while (fast_isspace((unsigned char)*value)) value++;
 
 	char* endptr;
 	const char* num_start = value;
@@ -371,7 +371,7 @@ bool pgsql_variable_validate_maintenance_work_mem_v3(const char* value, const pa
 		// Convert unit to lowercase for validation
 		char u[3] = { 0 };
 		for (int i = 0; i < 2 && unit_ptr[i]; i++)
-			u[i] = tolower((unsigned char)unit_ptr[i]);
+			u[i] = ::tolower((unsigned char)unit_ptr[i]);
 
 		// Validate units and set multipliers
 		if (unit_len == 1 && u[0] == 'b') {
@@ -471,7 +471,7 @@ bool pgsql_variable_validate_search_path(const char* value, const params_t* para
 
 	while (*token && result) {
 		/* skip leading whitespace */
-		while (*token && isspace((unsigned char)*token)) token++;
+		while (*token && fast_isspace((unsigned char)*token)) token++;
 		if (*token == '\0') break;
 
 		const char* part_start = token;
@@ -508,7 +508,7 @@ bool pgsql_variable_validate_search_path(const char* value, const params_t* para
 			}
 		} else {
 			// unquoted identifier or $user
-			while (*token && *token != ',' && !isspace(*token)) token++;
+			while (*token && *token != ',' && !fast_isspace(*token)) token++;
 			part_len = (size_t)(token - part_start);
 			if (part_len == 0 || part_len > 63) {
 				result = false;
@@ -543,7 +543,7 @@ bool pgsql_variable_validate_search_path(const char* value, const params_t* para
 		normalized[norm_pos] = '\0';
 
 		// skip whitespace after part
-		while (*token && isspace(*token)) token++;
+		while (*token && fast_isspace(*token)) token++;
 
 		// expect comma or end
 		if (*token == ',') {

lib/ProxySQL_Config.cpp

@@ -92,7 +92,7 @@ int ProxySQL_Config::Read_Global_Variables_from_configfile(const char *prefix) {
 		char *query=(char *)malloc(strlen(q)+strlen(prefix)+strlen(n)+strlen(value_string.c_str()));
 		sprintf(query,q, prefix, n, value_string.c_str());
 		//fprintf(stderr, "%s\n", query);
-  	admindb->execute(query);
+  		admindb->execute(query);
 		free(query);
 	}
 	admindb->execute("PRAGMA foreign_keys = ON");