Josselin - > Is there anything that is possible to help fixing this before the > release?
My reading of header files shows that libc6-dev in etch, lenny, and sid does not suffer from the original problem. The files that include a workaround (ugly, but apparently functional) for possible namespace pollution are /usr/include/sys/kd.h /usr/include/sys/sysctl.h I don't see mention of this problem in the Debian glibc changelog, and I don't know how to pull out older copies to see when this workaround was added. An earlier attempt is part of sarge (libc6-dev version 2.3.2.ds1-22sarge6). The simplest approach to this bug is to close it. Joey mentions apm.h. It turns out this is the _only_ other file (at least on my sid system) in /usr/include that nests to a linux kernel header file. It really should have a workaround like that in glibc. So this bug could be cloned, the copy assigned to apmd, and marked "important". I append a patch that is arguably cleaner than the approach used by glibc, and is somewhat tested (I checked that it didn't break the build of battery-stats, osdsh, sleepd, or wmbattery, and it fixes Joey's test case). My patch depends on a feature (__KERNEL_STRICT_NAMES) of linux/types.h that is at least as old as sarge (linux-kernel-headers version 2.5.999-test7-bk-17). This macro is defined in glibc features.h, so it looks like all the preprocessor gyrations in sys/sysctl.h are not needed, because that file includes <features.h>. - Larry --- apm.h.orig 2003-01-16 13:50:36.000000000 -0800 +++ apm.h 2008-10-13 13:43:47.000000000 -0700 @@ -20,6 +20,13 @@ * $Id: apm.h,v 1.7 1999/07/05 22:31:11 apenwarr Exp $ * */ +#ifndef _APM_H +#define _APM_H 1 + +#ifndef __KERNEL_STRICT_NAMES +#define __KERNEL_STRICT_NAMES +#endif + #include <linux/apm_bios.h> #include <sys/types.h> @@ -93,3 +100,5 @@ #else #define apm_reject(fd) (-EINVAL) #endif + +#endif -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
