Bug 441940

Summary: latest glibc breaks anaconda build
Product: Red Hat Enterprise Linux 5 Reporter: Doug Chapman <dchapman>
Component: anacondaAssignee: Chris Lumens <clumens>
Status: CLOSED ERRATA QA Contact: Alexander Todorov <atodorov>
Severity: urgent Docs Contact:
Priority: urgent    
Version: 5.2CC: anaconda-maint-list, dgregor, dwa, jakub, jlaska, mikem
Target Milestone: rcKeywords: Regression
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: RHBA-2008-0397 Doc Type: Bug Fix
Doc Text:
Story Points: ---
Clone Of: Environment:
Last Closed: 2008-05-21 15:33:52 UTC Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Attachments:
Description Flags
output of rpmbuild -bb anaconda.spec none

Description Doug Chapman 2008-04-10 19:51:59 UTC 
Description of problem:
While trying to debug an unrelated anaconda issue I ran into the problem that I
cannot build anaconda with glibc-2.5-23 (which is new today it appears).

A bunch of _nl_* symbols are now multiply defined or chagned:


/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o):(.rodata+0xa0):
multiple definition of `_nl_category_name_sizes'
../stubs/libunicode-lite.a(unicode-lite.o):(.rodata+0x0): first defined here
/usr/bin/ld: Warning: size of symbol `_nl_category_name_sizes' changed from 104
in ../stubs/libunicode-lite.a(unicode-lite.o) to 13 in
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o)
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o):(.rodata+0x90):
multiple definition of `_nl_category_name_idxs'
../stubs/libunicode-lite.a(unicode-lite.o):/usr/src/redhat/BUILD/anaconda-11.1.2.112/stubs/unicode-lite.c:45:
first defined here
/usr/bin/ld: Warning: size of symbol `_nl_category_name_idxs' changed from 1 in
../stubs/libunicode-lite.a(unicode-lite.o) to 13 in
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o)
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o):(.rodata+0x0):
multiple definition of `_nl_category_names'
../stubs/libunicode-lite.a(unicode-lite.o):(.data.rel.ro.local+0x0): first
defined here
/usr/bin/ld: Warning: size of symbol `_nl_category_names' changed from 104 in
../stubs/libunicode-lite.a(unicode-lite.o) to 137 in
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o)
/usr/lib/gcc/ia64-redhat-linux/4.1.2/../../../libc.a(setlocale.o): In function
`_nl_locale_subfreeres':
(__libc_freeres_fn+0x280): multiple definition of `_nl_locale_subfreeres'



I confirmed if I go back to glibc-2.5-21 the problem goes away.  Also a quick
diff of glibc soruces does indeed show changes in this area.


Version-Release number of selected component (if applicable):
I will upload the full log file of the build also.


How reproducible:
100%

Steps to Reproduce:
1. yum update (to get latest glibc)
2. rpmbild -bb anaconda.spec
3.
  
Actual results:


Expected results:


Additional info:
Comment 1 Doug Chapman 2008-04-10 19:51:59 UTC 
Created attachment 302064 [details]
output of rpmbuild -bb anaconda.spec
Comment 2 Jakub Jelinek 2008-04-10 20:05:37 UTC 
Then anaconda should be changed accordingly.  Does it use libc.a gettext but not
libc.a locale stuff?  That's risky to say at least.  I guess anaconda will have
the same problem in rawhide then...
All that changed in this area is that now __libc_setlocale_lock is a rwlock
rather than mutex.
Comment 3 Jakub Jelinek 2008-04-10 20:07:53 UTC 
-Wl,-M should help figure out why these .o files are now brought in and weren't
before.
Comment 4 Jakub Jelinek 2008-04-11 08:10:56 UTC 
--- anaconda-11.1.2.112/stubs/unicode-lite.c.jj 2008-04-08 18:17:36.000000000 -0400
+++ anaconda-11.1.2.112/stubs/unicode-lite.c    2008-04-11 03:40:02.000000000 -0400
@@ -116,5 +116,8 @@ const size_t _nl_category_name_sizes[] =
 
 /* avoid bringing in glibc's setlocale.o - we want to use our
    fake setlocale() */
-typedef pthread_mutex_t __libc_lock_t;
-__libc_lock_t __libc_setlocale_lock;
+union
+{
+  pthread_mutex_t mtx;
+  pthread_rwlock_t rwlck;
+} __libc_setlocale_lock __attribute__((nocommon));

should fix this (at least loader links).  As the stub doesn't use
__libc_setlocale_lock and so it is there only to satisfy wcsmbsload.o,
the above should make it work both with older and newer glibc (i.e. no matter
if it is actually a mutex, libc one int mutex or rwlock).  All the default
mutex, libc one int mutex and rwlock initializers are all zeros, and nocommon
means the symbol will be preferred over the one in libc.a (setlocale.o) where it
is STT_COMMON.
Comment 5 Chris Lumens 2008-04-11 15:22:50 UTC 
A quick scratch build earlier did not reproduce this issue, though I suppose
that's simply because the latest glibc has not made it into the right tree yet.
 We have at least one more reason to do a rebuild of anaconda, so I'll ACK this.
Comment 10 Chris Lumens 2008-04-15 21:34:32 UTC 
This will be fixed in anaconda-11.1.2.113-1.
Comment 14 errata-xmlrpc 2008-05-21 15:33:52 UTC 
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.

http://rhn.redhat.com/errata/RHBA-2008-0397.html