The powl(3) function is not actually needed here. Replace it with

constant number and make R-bigmemory compiled back again on systems
where double long mathlib functions missed.

R-bigmemory/distinfo

@@ -1,5 +1,6 @@
-$NetBSD: distinfo,v 1.1.1.1 2011/08/16 02:50:08 bubuchka Exp $
+$NetBSD: distinfo,v 1.2 2012/07/19 13:12:01 bubuchka Exp $
 
 SHA1 (R/bigmemory_4.2.11.tar.gz) = 3dd3c3a02dbb6ca742b81bf501e598bb6e0bdc1f
 RMD160 (R/bigmemory_4.2.11.tar.gz) = 7407fa28852f1f7981b45c36a8b7a3b254fb450a
 Size (R/bigmemory_4.2.11.tar.gz) = 1434562 bytes
+SHA1 (patch-aa) = 287e1117dcd23ba21b9b1d3c6f7c7c11fea9f596

R-bigmemory/patches/patch-aa

@@ -0,0 +1,15 @@
+$NetBSD: patch-aa,v 1.1 2012/07/19 13:12:02 bubuchka Exp $
+
+The powl(3) actually can be avoided in this particular case.
+
+--- src/bigmemory.cpp.orig	2011-07-26 16:35:38.000000000 +0000
++++ src/bigmemory.cpp
+@@ -39,7 +39,7 @@ string ttos<char>(char i)
+ 
+ bool TooManyRIndices( index_type val )
+ {
+-  return val > (powl(2, 31)-1);
++  return val > (double long)0xffffffff;	/* (powl(2, 31)-1) */
+ }
+ 
+ template<typename CType, typename RType, typename BMAccessorType>