On Tue, 26 Feb 2002, Paul Eggert wrote:
From: Peter Breitenlohner [mailto:peb@mppmu.mpg.de] Sent: Monday, February 25, 2002 4:56 AM
Bad timing between two instances of zic, where one tried to link to a file which the other had not yet created.
Thanks for the bug report. Here's an untested patch to glibc/timezone/Makefile to fix that timing problem. (Your symlink fix to zic.c is also welcome, but that's a separate problem.)
Can you please test this patch? If it works for you, we can submit it to bug-glibc as "the consensus of the tz experts" (all 2 of us :-).
2002-02-25 Paul Eggert <eggert@twinsun.com>
* timezone/Makefile (tzbases): Move systemv from here... (tzlinks): to here, since the entries in systemv were all changed from Zone to Link lines in tz-2001a.
--- glibc-2.2.5/timezone/Makefile 2001-08-29 15:45:25.000000000 -0700 +++ glibc-2.2.5-fix/timezone//Makefile 2002-02-25 23:50:22.373587963 -0800 @@ -31,9 +31,9 @@ others := zdump zic tests := test-tz tst-timezone
tzbases := africa antarctica asia australasia europe northamerica \ - southamerica etcetera factory systemv \ + southamerica etcetera factory \ solar87 solar88 solar89 -tzlinks := backward +tzlinks := backward systemv tzfiles := $(tzbases) $(tzlinks) # pacificnew doesn't compile; if it is to be used, it should be included in # northamerica.
Hi Paul, I'll test the patch, but this will take a few days. Right now I am upgrading all our systems linux-2.2.18 => 2.4.17, glibc-2.1.2 => 2.2.5 and this will take another two or three days. Once that is finished I intend to recompile glibc-2.2.5 and update all systems with the recompiled version (hoping that this will not require another reboot!). This way we will then take advantage of the 32 bit uid's and gid's in glibc. Until now we had a nasty NIS-gid problem (supplementary group 4294967294 for nobody from our AIX NIS master), which forced us to patch all occurences of initgroups("nobody",...) (in su, ftpd, apache, ...) -- quite a pain. BUT: As I said in my original mail, the actual bug was due to unfortunate timing which is hard to reproduce. I can only test the basic Makefile-functionality (the patch looks OK to me and should certainly delay the creation of the systemv timezone files). I'll inform you when I have recompiled glibc-2.2.5 regards Peter Breitenlohner <peb@mppmu.mpg.de>