Re: [ID 20010803.016] Possible regex bug in 5.6.1

Message-Id: <200108161527.f7GFRnH09419@crypt.compulink.co.uk>

p4raw-id: //depot/perl@11697

diff --git a/regexec.c b/regexec.c
index e95bd16..3f062ed 100644 (file)
--- a/regexec.c
+++ b/regexec.c
@@ -3033,12 +3033,15 @@ S_regmatch(pTHX_ regnode *prog)
                minmod = 0;
                if (ln && regrepeat_hard(scan, ln, &l) < ln)
                    sayNO;
-               if (ln && l == 0 && n >= ln
-                   /* In fact, this is tricky.  If paren, then the
-                      fact that we did/didnot match may influence
-                      future execution. */
-                   && !(paren && ln == 0))
-                   ln = n;
+               /* if we matched something zero-length we don't need to
+                  backtrack - capturing parens are already defined, so
+                  the caveat in the maximal case doesn't apply
+
+                  XXXX if ln == 0, we can redo this check first time
+                  through the following loop
+               */
+               if (ln && l == 0)
+                   n = ln;     /* don't backtrack */
                locinput = PL_reginput;
                if (PL_regkind[(U8)OP(next)] == EXACT) {
                    c1 = (U8)*STRING(next);
@@ -3060,7 +3063,7 @@ S_regmatch(pTHX_ regnode *prog)
                        UCHARAT(PL_reginput) == c2)
                    {
                        if (paren) {
-                           if (n) {
+                           if (ln) {
                                PL_regstartp[paren] =
                                    HOPc(PL_reginput, -l) - PL_bostr;
                                PL_regendp[paren] = PL_reginput - PL_bostr;
@@ -3084,12 +3087,13 @@ S_regmatch(pTHX_ regnode *prog)
            }
            else {
                n = regrepeat_hard(scan, n, &l);
-               if (n != 0 && l == 0
-                   /* In fact, this is tricky.  If paren, then the
-                      fact that we did/didnot match may influence
-                      future execution. */
-                   && !(paren && ln == 0))
-                   ln = n;
+               /* if we matched something zero-length we don't need to
+                  backtrack, unless the minimum count is zero and we
+                  are capturing the result - in that case the capture
+                  being defined or not may affect later execution
+               */
+               if (n != 0 && l == 0 && !(paren && ln == 0))
+                   ln = n;     /* don't backtrack */
                locinput = PL_reginput;
                DEBUG_r(
                    PerlIO_printf(Perl_debug_log,

diff --git a/t/op/re_tests b/t/op/re_tests
index cda8790..1e80f50 100644 (file)
--- a/t/op/re_tests
+++ b/t/op/re_tests
@@ -792,3 +792,7 @@ tt+$        xxxtt   y       -       -
 ^(a(??{"(?!)"})|(a)(?{1}))b    ab      y       $2      a       # [ID 20010811.006]
 ab(?i)cd       AbCd    n       -       -       # [ID 20010809.023]
 ab(?i)cd       abCd    y       -       -
+(A|B)*(?(1)(CD)|(CD))  CD      y       $2-$3   -CD
+(A|B)*(?(1)(CD)|(CD))  ABCD    y       $2-$3   CD-
+(A|B)*?(?(1)(CD)|(CD)) CD      y       $2-$3   -CD     # [ID 20010803.016]
+(A|B)*?(?(1)(CD)|(CD)) ABCD    y       $2-$3   CD-