When doing a reduction, we are going to need the whole list anyway. I…

…nstead of chewing it element by element, instead take it in bigger chunks. This reduces the fixed-size overhead per reificication which, in the case of reifying a range, is massively greater than the time to produce a single value. With this and the previous patch [+] 1..1000 runs in 8% of the time it did before; TimToady++ for pointing out this epic performance fail.

src/core/metaops.pm

@@ -64,8 +64,11 @@ sub METAOP_REDUCE(\$op, :$triangle) {
             return $op() unless @values.gimme(1);
             my $result := @values.shift;
             return $op($result) unless @values.gimme(1);
-            $result := $op($result, @values.shift)
-                while @values.gimme(1);
+            while @values.gimme(1000) {
+                my int $i = 0;
+                $result := $op($result, @values.shift)
+                    while ($i = $i + 1) < 1000 && @values.gimme(1);
+            }
             $result;
         })
 }
@@ -87,8 +90,11 @@ sub METAOP_REDUCE_RIGHT(\$op, :$triangle) {
             return $op() unless $list.gimme(1);
             my $result := $list.shift;
             return $op($result) unless $list.gimme(1);
-            $result := $op($list.shift, $result)
-                while $list.gimme(1);
+            while $list.gimme(1000) {
+                my int $i = 0;
+                $result := $op($list.shift, $result)
+                    while ($i = $i + 1) < 1000 && $list.gimme(1);
+            }
             $result;
         }
     }