commented version of a patch suggested by Drago Goricanec
authorJarkko Hietaniemi <jhi@iki.fi>
Fri, 16 Oct 1998 15:24:45 +0000 (18:24 +0300)
committerGurusamy Sarathy <gsar@cpan.org>
Sun, 25 Oct 1998 05:16:41 +0000 (05:16 +0000)
Message-ID: <oeeogrc1ype.fsf@alpha.hut.fi>
Subject: Re: [PATCH 5.005_52] Linux select fails with more than 32 FDs

p4raw-id: //depot/perl@2056

pp_sys.c

index 4439b1c..1cd1cda 100644 (file)
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -868,7 +868,12 @@ PP(pp_sselect)
 /* little endians can use vecs directly */
 #if BYTEORDER == 0x1234 || BYTEORDER == 0x12345678
 #  if SELECT_MIN_BITS > 1
-    growsize = SELECT_MIN_BITS / 8;
+    /* If SELECT_MIN_BITS is greater than one we most probably will want
+     * to align the sizes with SELECT_MIN_BITS/8 because for example
+     * in many little-endian (Intel, Alpha) systems (Linux, OS/2, Digital
+     * UNIX, Solaris, NeXT) the smallest quantum select() operates on
+     * (sets bit) is 32 bits.  */
+    growsize = maxlen + (SELECT_MIN_BITS/8 - (maxlen % (SELECT_MIN_BITS/8)));
 #  else
     growsize = sizeof(fd_set);
 #  endif