Revert 92ec6835548d36 - don't use DFA if need_start is true.

2 files changed, 21 insertions, 3 deletions

diff --git a/ChangeLog b/ChangeLog
index 4c34f193..8c8d91df 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,12 @@
+2017-07-15         Arnold D. Robbins     <arnold@skeeve.com>
+
+	Revert change of 2016-07-24 that always runs the dfa
+	matcher. Based on a bug report from Alexandre Oliva
+	<lxoliva@fsfla.org>, DFA can cause gawk to hang, even
+	in the C locale.
+
+	* re.c (research): Don't use dfa if need_start is true.
+
 2017-07-07         Arnold D. Robbins     <arnold@skeeve.com>
 
 	Remove warnings from GCC 7.1 compilation.
diff --git a/re.c b/re.c
index 878c884e..d1b0ee7f 100644
--- a/re.c
+++ b/re.c
@@ -268,17 +268,26 @@ research(Regexp *rp, char *str, int start,
 		rp->pat.not_bol = 1;
 
 	/*
-	 * Always do dfa search if can; if it fails, we won't bother
-	 * with the regex search.
+	 * Always do dfa search if can; if it fails, then even if
+	 * need_start is true, we won't bother with the regex search.
 	 *
 	 * The dfa matcher doesn't have a no_bol flag, so don't bother
 	 * trying it in that case.
 	 *
+	 * 7/2008: Skip the dfa matcher if need_start. The dfa matcher
+	 * has bugs in certain multibyte cases and it's too difficult
+	 * to try to special case things.
+	 * 7/2017: Apparently there are some cases where DFA gets
+	 * stuck, even in the C locale, so we use dfa only if not need_start.
+	 *
+	 * Should that issue ever get resolved, note this comment:
+	 *
 	 * 7/2016: The dfa matcher can't handle a case where searching
 	 * starts in the middle of a string, so don't bother trying it
 	 * in that case.
+	 *	if (rp->dfa && ! no_bol && start == 0) ...
 	 */
-	if (rp->dfa && ! no_bol && start == 0) {
+	if (rp->dfa && ! no_bol && ! need_start) {
 		char save;
 		size_t count = 0;
 		struct dfa *superset = dfasuperset(rp->dfareg);