Hi Dominik, On 31-10-14 15:32, Paul Gevers wrote: > dbconfig-generate-include should be called by dpkg/common on line 547 > with the -U option, so the path should be the ucf path. While I was > thinking about it I believe that indeed all the configuration files > which are not conffiles should be handled by ucf, because that is > supposed to do-the-right-thing.
The more I look at the code, the less I am convinced that your patch in dbconfig-common actually solves your problem. I added "set -x" and "set +x" to dbconfig-generate-include around the block that we are discussing, and when I reconfigure roundcube-core I see: (pbuild29710)root@wollumbin:/# dpkg-reconfigure roundcube-core dbconfig-common: writing config to /etc/dbconfig-common/roundcube.conf + [ /etc/roundcube/debian-db.php ] + ls -al /tmp/dbconfig-generate-include.7z8TTm -rw------- 1 root root 546 Nov 1 12:04 /tmp/dbconfig-generate-include.7z8TTm + ls -al /etc/roundcube/debian-db.php -rw-r----- 1 root www-data 546 Nov 1 10:51 /etc/roundcube/debian-db.php + [ 1 ] + ucf --debconf-ok /tmp/dbconfig-generate-include.7z8TTm /etc/roundcube/debian-db.php Replacing config file /etc/roundcube/debian-db.php with new version + rm -f /tmp/dbconfig-generate-include.7z8TTm + ls -al /etc/roundcube/debian-db.php -rw------- 1 root root 546 Nov 1 12:04 /etc/roundcube/debian-db.php + check_permissions + local line + dpkg-statoverride --list /etc/roundcube/debian-db.php + [ ] + [ ] + ls -al /etc/roundcube/debian-db.php -rw------- 1 root root 546 Nov 1 12:04 /etc/roundcube/debian-db.php + set +x apache2_invoke roundcube.conf: already enabled invoke-rc.d: policy-rc.d denied execution of reload. invoke-rc.d: policy-rc.d denied execution of reload. (pbuild29710)root@wollumbin:/# Your patch in the "else" statement is completely irrelevant in this sequence. So my question, how did you test your patch? [1] Furthermore, I have been reading up on the use of cp. It seems, but if you are using ACL's I like you to test, that cp opens the file for writing, so it should keep existing ACL's AFAICT. Paul [1] I think I now understand why my fix for bug 720517 was wrong and I didn't catch that during my testing. I had a local change, and then ucf leaves the original file completely alone. I didn't test better, because my expectation from ucf was just wrong.
signature.asc
Description: OpenPGP digital signature
