Resolve

http://www.nntp.perl.org/group/perl.perl5.porters/2007/06/msg125667.html
by reverting part of change #29354. 

Unfortunately match vars after a /g match in scalar context will be
unsafe (again) after this, but such matches on long strings won't be
as diabolically slow.

Question: why does the new test in t/op/pat.t pass, but the same test
in t/op/reg_unsafe.t fail? (Latter is TODO for now)
p4raw-link: @29354 on //depot/perl: 58e23c8d7d24dd08c87b5d56819ad45527176c15

p4raw-id: //depot/perl@31451

1 files changed, 6 insertions, 3 deletions

diff --git a/pp_hot.c b/pp_hot.c
index 27e863db9b..1535e4cd8f 100644
--- a/pp_hot.c
+++ b/pp_hot.c
@@ -1266,9 +1266,12 @@ PP(pp_match)
 	    }
 	}
     }
-    /* remove comment to get faster /g but possibly unsafe $1 vars after a
-       match. Test for the unsafe vars will fail as well*/
-    if (( /* !global &&  */ rx->nparens) 
+    /* XXX: comment out !global get safe $1 vars after a
+       match, BUT be aware that this leads to drammatic slowdowns on
+       /g matches against large strings.  So far a solution to this problem
+       appears to be quite tricky.
+       Test for the unsafe vars are TODO for now. */
+    if ((  !global &&  rx->nparens) 
 	    || SvTEMP(TARG) || PL_sawampersand ||
 	    (rx->extflags & (RXf_EVAL_SEEN|RXf_PMf_KEEPCOPY)))
 	r_flags |= REXEC_COPY_STR;