| View previous topic :: View next topic |
| Author |
Message |
halfgaar
l33t
Joined: 22 Feb 2004
Posts: 781
Location: Netherlands
|
 Posted: Sat Mar 04, 2006 1:06 pm Post subject: Does upgrading GCC result in orphaned files? |
 |
|
When you upgrade GCC, it often runs by all the .la files on your system to change hardcoded filepaths contained in them. You see messages like:
| Code: | | FIXING: /usr/kde/3.4/lib/libkdeui.la ...[v] |
Now, what I'd like to know is, does emerge/portage correct the mtime and md5sum info in the DB which contains the installed packages of your system? It would appear it doesn't:
| Code: |
# equery check =kdelibs-3.4.1-r1
!!! /usr/kde/3.4/lib/kde3/kabc_net.la has incorrect md5sum
!!! /usr/kde/3.4/lib/kde3/kimg_xcf.la has incorrect md5sum
!!! /usr/kde/3.4/lib/kde3/kio_ftp.la has incorrect md5sum
!!! /usr/kde/3.4/lib/kde3/kio_http.la has incorrect md5sum
!!! /usr/kde/3.4/lib/libknewstuff.la has incorrect md5sum
!!! /usr/kde/3.4/lib/kde3/kdeprint_ext.la has incorrect md5sum
!!! /usr/kde/3.4/lib/kde3/kded_kdeprintd.la has incorrect md5sum
etc |
But doesn't this mess up your entire system, because now you have hundreds of orphaned files, which don't get uninstalled when you remove kdelibs.
In my opinion, this is a rather serious bug. Any thoughts? |
|
| Back to top |
|
 |
drwook
Veteran
Joined: 30 Mar 2005
Posts: 1324
Location: London
|
| Posted: Sat Mar 04, 2006 3:32 pm Post subject: |
|
|
hmm... I haven't looked at it but it sounds a reasonable assumption from what you say... If you've an affected system you can afford to fiddle with, why not try unmerging kdelibs to confirm the files remain? (qpkg kdelibs or something first for quick reinstall  ) Then if so open a bug on bugzilla. |
|
| Back to top |
|
|
sg313
n00b
Joined: 25 Dec 2005
Posts: 34
|
| Posted: Sat Mar 04, 2006 4:03 pm Post subject: |
|
|
hi! i've written a script that can fix the db. It overwrites the time/md5 entries with the values of the existing file in the filesystem.
| Code: |
#!/bin/bash
scripttitle="fixdb for Portage Package Database, v0.3 (4MAR06)"
pkgdb=/var/db/pkg/
declare -a packagenames
regex=0
ask=0
arg_vec=($@)
if [ $# -eq 0 ]; then arg_vec="--help"; fi
declare -i index
for (( index=0; $index <= $#; index++ )); do
cur_arg="${arg_vec[$index]}"
if [ "$cur_arg" == "--pkgdb" ]; then
let "index=index+1";
pkgdb="${arg_vec[index]}";
if [[ $pkgdb =~ '.*[^/]$' ]]; then
pkgdb="$pkgdb/"
fi
elif [ "$cur_arg" == "--ask" ]; then ask=1
elif [ "$cur_arg" == "--regex" ]; then regex=1
elif [ "$cur_arg" == "--help" ]; then
echo -e "$scripttitle
Updates md5sums and mtimes of given packages and backups old CONTENTS file as CONTENTS_OLD.
ATTENTION: Sequential runs will overwrite CONTENTS_OLD!
This script was written by sg313@gmail.com. It is distibuted under the GPLv2.\n
Usage : fixdb [Options] packagenames
Options: --pkgdb <loc>/ : use database at <loc> instead of '/var/db/pkg
--ask : if multiple versions are installed, ask which should be fixed
--regex : treate packagenames as regular expressions. all matching packages
will befixed. use --ask if you want a list to choose from.
--help : display this text.\n
Script will bail out if unknown option is given."
exit
elif [[ $cur_arg =~ '\-.*' ]]; then
echo "Unknown Option: "$cur_arg". Try --help."
exit
else
packagenames=("$packagenames $cur_arg")
fi
done
if [ ${#packagenames[@]} -eq 0 ]; then
echo "No package to fix specified; try --help"
exit
fi
echo "$scripttitle"
# generate an array of directories from imput
fixes=0; nonexistent=0;
for packagename in $packagenames; do
declare -a pkgs
if [ $regex -eq 1 ]; then
package=( $(find $pkgdb -maxdepth 2 -regex ".*/$packagename" -type d) )
else
package=( $(find $pkgdb -maxdepth 2 -regex ".*/$packagename\-.*" -type d) )
fi
pkg_count=${#package[@]}
if [ "$pkg_count" -eq "0" ]; then
echo "No package found named $packagename Exiting."
exit
elif [ "$pkg_count" -eq "1" ]; then
pkg=${package[0]}
elif [[ "$pkg_count" -gt "1" && $ask -eq 1 ]]; then
echo "The following versions of $packagename were found"
index=0;
while [ "$index" -lt "$pkg_count" ]; do
pkg_name=`echo ${package[$index]} | sed '#$pkgdb#d'`
echo "$index. $pkg_name"
let "index = $index + 1"
done
echo -n "Choose wisely: "
read package_number
pkgs=( $pkgs "${package[$package_number]}" )
else
pkgs=( $pkgs "${package[@]}" )
fi
done
# now check all the directories in array 'pkgs'
declare -i pkg_i
for ((pkg_i=0; $pkg_i<${#pkgs[@]}; pkg_i++)); do
pkg=${pkgs[$pkg_i]}
pkg_name=`echo $pkg | sed '#$pkgdb#d'`
let "pkg_i_out=pgk_i+1"
echo "($pkg_i_out/${#pkgs[@]}) fixing $pkg_name..."
cont=$pkg/CONTENTS_NEW
rm -rf $cont
touch $cont
while read -a col
do
if [ "${col[0]}" == "dir" ];
then
echo ${col[0]} ${col[1]} >> $cont
elif [ -e ${col[1]} ];
then
nmt=`stat --format=%Y ${col[1]}`
if [ "${col[0]}" == "obj" ];
then
if [ "${col[3]}" != "$nmt" ];
then
echo " fixing mtime for" ${col[1]}
((fixes++))
fi
nmd=`md5sum ${col[1]} | sed 's/^\(.*\)[ ].*/\1/'`
if [ ${col[2]} != $nmd ]; then
echo " fixing md5 for" ${col[1]}
((fixes++))
fi
echo ${col[0]} ${col[1]} $nmd $nmt >> $cont
elif [ "${col[0]}" == "sym" ];
then
if [ "${col[3]}" == "$nmt" ];
then
echo " fixing mtime for" ${col[1]}
((fixes++))
fi
echo ${col[0]} ${col[1]} ${col[2]} ${col[3]} $nmt >> $cont
fi
else
echo ${col[@]} >> $cont
echo $pkg_name "(!exist)" ${col[0]} ${col[1]}
let "nonexistent=nonexistent+1"
fi
done < $pkg/CONTENTS
#rm -rf -- $pkg/CONTENTS_NEW
mv $pkg/CONTENTS $pkg/CONTENTS_OLD
mv $pkg/CONTENTS_NEW $pkg/CONTENTS
done
if [ $fixes -gt 0 ]; then
echo "$fixes entries fixed."
elif [ $nonexistent -eq 0 ]; then
echo "Nothing to fix."
fi
if [ $nonexistent -gt 0 ]; then
echo "!!! There are nonexistent files. These where not fixed. Take a look at the output above."
fi
|
|
|
| Back to top |
|
|
halfgaar
l33t
Joined: 22 Feb 2004
Posts: 781
Location: Netherlands
|
| Posted: Sat Mar 04, 2006 4:06 pm Post subject: |
|
|
I've confirmed it with "atlantik" just now. "equery check" showed 4 invalid files, and they all remained after uninstall.
During unmerge, they are marked:
| Code: | | --- !mtime obj /usr/kde/3.4/lib/kde3/kio_atlantik.la |
Where do I report this? I mean, which category? I think the bug applies to much more packages than GCC.
edit:
g313,
Your fix is very dangerous, because the mtime check during unmerge, is to prevent files which are overwritten legitimatly, to be uninstalled when you unmerge a package. Your fix would simply be equivalent to removing the mtime check from the unmerge procedure. |
|
| Back to top |
|
|
sg313
n00b
Joined: 25 Dec 2005
Posts: 34
|
| Posted: Sat Mar 04, 2006 4:28 pm Post subject: |
|
|
| halfgaar, your right. so one has to be verry careful running that script. i needed it once I had to recover my root partition and thus had lost all filetimes. |
|
| Back to top |
|
|
halfgaar
l33t
Joined: 22 Feb 2004
Posts: 781
Location: Netherlands
|
| Posted: Sat Mar 04, 2006 4:32 pm Post subject: |
|
|
| What do you mean, recover your root partition? Restoring a backup? Good backup tools set mtimes and such. And, when you use cp, just use cp -a. If you don't, you loose a lot more information, like hardlinks. |
|
| Back to top |
|
|
sg313
n00b
Joined: 25 Dec 2005
Posts: 34
|
| Posted: Sat Mar 04, 2006 5:14 pm Post subject: |
|
|
reiserfsck /dev/hda3 --rebuild-tree --scan-whole-partition
Took about 20 hours, but almost everything was there again. To validate my system, i checked all packages, reemerged those that had lost files and used fixdb for those that had all files, but incorrect mtimes and/or md5sums.
Definitly an obscure way, but it was a lot of fun 
And thanks for the cp -a, didn't know about it. |
|
| Back to top |
|
|
halfgaar
l33t
Joined: 22 Feb 2004
Posts: 781
Location: Netherlands
|
| Posted: Sat Mar 04, 2006 5:56 pm Post subject: |
|
|
| Are you saying that the reiserfsck changed mtimes of files? Is that appropriate behaviour? |
|
| Back to top |
|
|
halfgaar
l33t
Joined: 22 Feb 2004
Posts: 781
Location: Netherlands
|
| Posted: Sat Mar 04, 2006 6:25 pm Post subject: |
|
|
| OK, posted a bugreport. |
|
| Back to top |
|
|
irondog
l33t
Joined: 07 Jul 2003
Posts: 715
Location: Voor mijn TV. Achter mijn pc.
|
|
| Back to top |
|
|
|
Portage & Programming
