From: Ben Tilly <ben_tilly@operamail.com>
Date: Sat, 15 Jul 2000 17:23:27 +0000 (-0400)
Subject: Add an optimization for map-maps-a-list-element-to-more-list-elements
X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=18ef8beac527ca204cb2b7f96c5a4c585c8ee174;p=p5sagit%2Fp5-mst-13.2.git

Add an optimization for map-maps-a-list-element-to-more-list-elements
case, but add also notes explaining the relationship of this
patch and the earlier notes by Sarathy.

Subject: Map is still slow
Message-ID: <20000715212327.21656.qmail@hotmail.com>

p4raw-id: //depot/perl@6429
---

diff --git a/pp_ctl.c b/pp_ctl.c
index 75673dc..776754e 100644
--- a/pp_ctl.c
+++ b/pp_ctl.c
@@ -742,7 +742,7 @@ PP(pp_mapwhile)
 	     * is repeatedly extended for every set of items.  Is possible
 	     * to do this without any stack extension or copying at all
 	     * by maintaining a separate list over which the map iterates
-	     * (like foreach does). */
+	     * (like foreach does). --gsar */
 
 	    /* everything in the stack after the destination list moves
 	     * towards the end the stack by the amount of room needed */
@@ -750,6 +750,17 @@ PP(pp_mapwhile)
 
 	    /* items to shift up (accounting for the moved source pointer) */
 	    count = (SP - PL_stack_base) - (PL_markstack_ptr[-1] - 1);
+
+	    /* This optimization is by Ben Tilly and it does
+	     * things differently from what Sarathy (gsar)
+	     * is describing.  The downside of this optimization is
+	     * that leaves "holes" (uninitialized and hopefully unused areas)
+	     * to the Perl stack, but on the other hand this
+	     * shouldn't be a problem.  If Sarathy's idea gets
+	     * implemented, this optimization should become
+	     * irrelevant.  --jhi */
+            if (shift < count)
+                shift = count; /* Avoid shifting too often --Ben Tilly */
 	    
 	    EXTEND(SP,shift);
 	    src = SP;