Repair bug in regexp split performance improvements.

RhodiumToad · RhodiumToad · commit dea7fc60a051 · 2018-09-12T19:49:59.000+01:00
Commit c8ea87e introduced a temporary conversion buffer for substrings extracted during regexp splits. Unfortunately the code that sized it was failing to ignore the effects of ignored degenerate regexp matches, so for regexp_split_* calls it could under-size the buffer in such cases. Fix, and add some regression test cases (though those will only catch the bug if run in a multibyte encoding). Backpatch to 9.3 as the faulty code was. Thanks to the PostGIS project, Regina Obe and Paul Ramsey for the report (via IRC) and assistance in analysis. Patch by me.
diff --git a/src/backend/utils/adt/regexp.c b/src/backend/utils/adt/regexp.c
@@ -882,6 +882,7 @@ setup_regexp_matches(text *orig_str, text *pattern, text *flags,
 	int			array_len;
 	int			array_idx;
 	int			prev_match_end;
+	int			prev_valid_match_end;
 	int			start_search;
 	int			maxlen = 0;		/* largest fetch length in characters */
 
@@ -937,6 +938,7 @@ setup_regexp_matches(text *orig_str, text *pattern, text *flags,
 
 	/* search for the pattern, perhaps repeatedly */
 	prev_match_end = 0;
+	prev_valid_match_end = 0;
 	start_search = 0;
 	while (RE_wchar_execute(cpattern, wide_str, wide_len, start_search,
 							pmatch_len, pmatch))
@@ -989,13 +991,15 @@ setup_regexp_matches(text *orig_str, text *pattern, text *flags,
 			matchctx->nmatches++;
 
 			/*
-			 * check length of unmatched portion between end of previous match
-			 * and start of current one
+			 * check length of unmatched portion between end of previous valid
+			 * (nondegenerate, or degenerate but not ignored) match and start
+			 * of current one
 			 */
 			if (fetching_unmatched &&
 				pmatch[0].rm_so >= 0 &&
-				(pmatch[0].rm_so - prev_match_end) > maxlen)
-				maxlen = (pmatch[0].rm_so - prev_match_end);
+				(pmatch[0].rm_so - prev_valid_match_end) > maxlen)
+				maxlen = (pmatch[0].rm_so - prev_valid_match_end);
+			prev_valid_match_end = pmatch[0].rm_eo;
 		}
 		prev_match_end = pmatch[0].rm_eo;
 
@@ -1021,8 +1025,8 @@ setup_regexp_matches(text *orig_str, text *pattern, text *flags,
 	 * input string
 	 */
 	if (fetching_unmatched &&
-		(wide_len - prev_match_end) > maxlen)
-		maxlen = (wide_len - prev_match_end);
+		(wide_len - prev_valid_match_end) > maxlen)
+		maxlen = (wide_len - prev_valid_match_end);
 
 	/*
 	 * Keep a note of the end position of the string for the benefit of
diff --git a/src/test/regress/expected/strings.out b/src/test/regress/expected/strings.out
@@ -674,6 +674,24 @@ SELECT regexp_split_to_array('123456','.');
  {"","","","","","",""}
 (1 row)
 
+SELECT regexp_split_to_array('123456','');
+ regexp_split_to_array 
+-----------------------
+ {1,2,3,4,5,6}
+(1 row)
+
+SELECT regexp_split_to_array('123456','(?:)');
+ regexp_split_to_array 
+-----------------------
+ {1,2,3,4,5,6}
+(1 row)
+
+SELECT regexp_split_to_array('1','');
+ regexp_split_to_array 
+-----------------------
+ {1}
+(1 row)
+
 -- errors
 SELECT foo, length(foo) FROM regexp_split_to_table('thE QUick bROWn FOx jUMPs ovEr The lazy dOG', 'e', 'zippy') AS foo;
 ERROR:  invalid regexp option: "z"
diff --git a/src/test/regress/sql/strings.sql b/src/test/regress/sql/strings.sql
@@ -188,6 +188,9 @@ SELECT regexp_split_to_array('the quick brown fox jumps over the lazy dog', 'nom
 SELECT regexp_split_to_array('123456','1');
 SELECT regexp_split_to_array('123456','6');
 SELECT regexp_split_to_array('123456','.');
+SELECT regexp_split_to_array('123456','');
+SELECT regexp_split_to_array('123456','(?:)');
+SELECT regexp_split_to_array('1','');
 -- errors
 SELECT foo, length(foo) FROM regexp_split_to_table('thE QUick bROWn FOx jUMPs ovEr The lazy dOG', 'e', 'zippy') AS foo;
 SELECT regexp_split_to_array('thE QUick bROWn FOx jUMPs ovEr The lazy dOG', 'e', 'iz');
