Fix backtracking

Author: Kenneth Shelton

src/parser.c

@@ -3006,10 +3006,12 @@ PARSER_Parse(Repeat)
 	struct data_Repeat *const data = (struct data_Repeat*) pdata;
 	struct ln_pdag *endNode = NULL;
 	size_t strtoffs = *offs;
+	size_t lastMatch = strtoffs;
 	size_t lastKnownGood = strtoffs;
 	struct json_object *json_arr = NULL;
 	struct json_object *parsed_value = NULL;
 	const size_t parsedTo_save = npb->parsedTo;
+	const size_t longestParsedTo_save = npb->longestParsedTo;
 	int mergeResults = parser_name != NULL && parser_name[0] == '.' && parser_name[1] == '\0';
 
 	do {
@@ -3031,6 +3033,8 @@ PARSER_Parse(Repeat)
 					"parse ptr back to %zd", strtoffs);
 				goto success;
 			} else {
+				// Reset longest match
+				npb->longestParsedTo = lastMatch > longestParsedTo_save ? lastMatch : longestParsedTo_save;
 				goto done;
 			}
 		}
@@ -3070,6 +3074,7 @@ PARSER_Parse(Repeat)
 
 		/* now check if we shall continue */
 		npb->parsedTo = 0;
+		lastMatch = lastKnownGood;
 		lastKnownGood = strtoffs; /* record pos in case of fail in while */
 		r = ln_normalizeRec(npb, data->while_cond, strtoffs, 1, NULL, &endNode, 0, NULL, parser_name);
 		LN_DBGPRINTF(npb->ctx, "repeat while returns %d, parsed %zu",

src/pdag.c

@@ -1363,6 +1363,7 @@ fixJSON(struct ln_pdag *dag,
 			/* Free the unneeded value */
 			json_object_put(*value);
 		}
+		*value = NULL;
 	} else if(prs->name[0] == '.' && prs->name[1] == '\0') {
 		if(json_object_get_type(*value) == json_type_object) {
 			struct json_object_iterator it = json_object_iter_begin(*value);
@@ -1417,6 +1418,7 @@ fixJSON(struct ln_pdag *dag,
 		}
 	}
 	r = 0;
+	*value = NULL;
 	return r;
 }
 
@@ -1588,7 +1590,7 @@ ln_normalizeRec(npb_t *const __restrict__ npb,
 	size_t iprs;
 	size_t parsedTo = npb->parsedTo;
 	size_t parsed = 0;
-	struct json_object *value;
+	struct json_object *value = NULL;
 
 LN_DBGPRINTF(dag->ctx, "%zu: enter parser, dag node %p, json %p", offs, dag, json);
 
@@ -1614,7 +1616,6 @@ LN_DBGPRINTF(dag->ctx, "%zu: enter parser, dag node %p, json %p", offs, dag, jso
 				 	 : "UNKNOWN");
 		}
 		i = offs;
-		value = NULL;
 		localR = tryParser(npb, dag, &i, &parsed, &value, prs, failOnDuplicate, json, prs->name);
 		if(localR == 0) {
 			parsedTo = i + parsed;
@@ -1628,9 +1629,14 @@ LN_DBGPRINTF(dag->ctx, "%zu: enter parser, dag node %p, json %p", offs, dag, jso
 			if(r == 0) {
 				LN_DBGPRINTF(dag->ctx, "%zu: parser matches at %zu", offs, i);
 				CHKR(fixJSON(dag, &value, json, prs));
+				value = NULL;
 				if(npb->ctx->opts & LN_CTXOPT_ADD_RULE) {
 					add_rule_to_mockup(npb, prs);
 				}
+				/* did we have a longer parser --> then update */
+				if(parsedTo > npb->parsedTo)
+					npb->parsedTo = parsedTo;
+
 			} else {
 				++dag->stats.backtracked;
 				#ifdef	ADVANCED_STATS
@@ -1639,14 +1645,16 @@ LN_DBGPRINTF(dag->ctx, "%zu: enter parser, dag node %p, json %p", offs, dag, jso
 				#endif
 				LN_DBGPRINTF(dag->ctx, "%zu nonmatch, backtracking required, parsed to=%zu",
 						offs, parsedTo);
-				if (value != NULL) { /* Free the value if it was created */
-					json_object_put(value);
-				}
 			}
 		}
+		if (value != NULL) { /* Free the value if it was created */
+			json_object_put(value);
+			value = NULL;
+		}
+
 		/* did we have a longer parser --> then update */
-		if(parsedTo > npb->parsedTo)
-			npb->parsedTo = parsedTo;
+		if(parsedTo > npb->longestParsedTo)
+			npb->longestParsedTo = parsedTo;
 		LN_DBGPRINTF(dag->ctx, "parsedTo %zu, *pParsedTo %zu", parsedTo, npb->parsedTo);
 	}
 
@@ -1725,7 +1733,7 @@ ln_normalize(ln_ctx ctx, const char *str, const size_t strLen, struct json_objec
 		addRuleMetadata(&npb, *json_p, endNode);
 		r = 0;
 	} else {
-		addUnparsedField(str, strLen, npb.parsedTo, *json_p);
+		addUnparsedField(str, strLen, npb.longestParsedTo, *json_p);
 	}
 
 	if(ctx->opts & LN_CTXOPT_ADD_RULE) {

src/pdag.h

@@ -159,6 +159,7 @@ struct npb {
 	const char *str;		/**< to-be-normalized message */
 	size_t strLen;			/**< length of it */
 	size_t parsedTo;		/**< up to which byte could this be parsed? */
+	size_t longestParsedTo;	/**< up to which byte could this be parsed? */
 	es_str_t *rule;			/**< a mock-up of the rule used to parse */
 	es_str_t *exec_path;
 #ifdef ADVANCED_STATS