From: Jarkko Hietaniemi <jhi@iki.fi>
Date: Fri, 16 Oct 1998 15:24:45 +0000 (+0300)
Subject: commented version of a patch suggested by Drago Goricanec
X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=f2da832e840f54013cb015f86420ceda75823b28;p=p5sagit%2Fp5-mst-13.2.git

commented version of a patch suggested by Drago Goricanec
	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
---

diff --git a/pp_sys.c b/pp_sys.c
index 4439b1c..1cd1cda 100644
--- 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