В Tue, 9 Oct 2012 12:32:07 +0200 Julien Cristau <jcristau@debian.org> пишет: > On Tue, Oct 9, 2012 at 13:17:23 +0400, Alexander Golovko wrote: > > > 1. Build packages for all database types in the same time, not a > > separate process for sqlite3, mysql and pgsql. > > > > http://anonscm.debian.org/gitweb/?p=pkg-bacula/bacula.git;a=commit;h=1ca440fc3758a28fdcd17c05aa24f724934dbc5f > > > nak. I try to convince you again, if you don't mind. Benefits for our team: Package building is faster. It was "start, wait, make something, look, wait, look, wait, forget, remember only after some time" :( It was really annoying and significantly slow my work. Package building process is much clear. Anybody other can help us if he has basic knowledge about debhelper. With old variant d/rules is more hard for understanding. Also this change i test very-very deep. I compare packages, built from sources with and without this change. Comparison include non-elf files by content, where point symlinks, list of linked libraries for elf files and buildlogs for anomaly differences in flags. There was one of my most tested changes. I really can promise, that this change not affect anything in binary packages. I hope, if this arguments will enough for allow exclusion and accept this change. > > 3. fix daemons user/group on systems with systemd (Closes: #679958). > > delegate daemons uid/gid changing to start-stop-daemon or systemd, > > thanks to Matija Nalis (Closes: #556207). > > > > This changes are related and intersected. First commit reverted by > > last. > > > > http://anonscm.debian.org/gitweb/?p=pkg-bacula/bacula.git;a=commit;h=d077cd3c71734828b635f8605c8411f6cd86b6f6 > > http://anonscm.debian.org/gitweb/?p=pkg-bacula/bacula.git;a=commit;h=cf9eb640182f4adfd83d05954dc35a20b60170c1 > > http://anonscm.debian.org/gitweb/?p=pkg-bacula/bacula.git;a=commit;h=694b788e0f82a734ca98bb0930a97432240c7fe8 > > > > Upstream already much time (at least since 2010, Oct) use > > start-stop-daemon for change daemons uid/gid in their variant of > > init scripts. Our patch for uid/gid changing by systemd was already > > accepted into upstream. > > > > The big problem, that without this change user can't simply get > > backtraces on daemons crashes. This backtraces required for some > > bugreport to upstream. > > > Meh. Can't the systemd files be simply removed? If this doesn't > affect sysvinit systems I would prefer not to bother. Hmm. But systemd service files was added to package in 5.2.6+dfsg-1 via bug #624532. And patch for fix them is trivial. > > 5. Add build-depends for read-all capability support (Closes: > > #683080). > > > A bit nervous about this one. How much has this been tested, and how > likely is it to affect unrelated features? There are only one difference between packages built with and without capabilities support in src/lib/priv.c on lines 110-132. http://anonscm.debian.org/gitweb/?p=pkg-bacula/bacula.git;a=blob;f=src/lib/priv.c;h=cfa4ec768970df40c3baea3f9d87831e761b7735;hb=52609e79da8df27cb68918acd074bff4af7cceb0 If binaries built without capabilities, than bacula-fd will return error "Keep readall caps not implemented" when user try to use it. If user start bacula without flags for enabling cap usage (default), then the same code will be executed in both binaries - with or without cap support. This change should not affect to anything other. > > 8. make package purging more careful about users files. > > > Don't really see the point of this one. Purging means purging. actually, this is a difference from squeeze behavior for bacula-director packages. Revert anyway? > > 10. force /etc/defaults/bacula-dir reregistration in ucf when > > changing bacula-director database type, fix purging after this > > (Closes: #680051). > > > Not sure about this, would appreciate advice from somebody who knows > ucf. > > > 11. switch to /run directory > > > nak. Ok. i revert it. > > 12. Fix impossibility to run out-of-box scripts make_mysql_tables > > and update_mysql_tables scripts, shipped with package (#679855). > > > Are these scripts ever run automatically by the package? No, scripts (dbconfig-common) use other files, generated on build time. This files intended for users, who don't want or can't use dbconfig for database population. > > 13. switch from usermod to more debian-policy friendly adduser. > > > Unless this is fixing an actual bug, nak. Ok, i revert it. > > 14. fix hostname substitution (Closes: #682966). > > > Doesn't seem critical? It is not critical for package installation. But this changes will do all users immediately after first package installation. > > 15. add bacula into cdrom group (Closes: #520508). > > > Same here, I'd say defer to wheezy+1. Ok, i revert it. > > 16. Cleanup list of linked libraries. > > > nak. Maybe i incorrect worded this change. This change remove unused libraries (no symbols from library used) from list of linked libraries. From ld documentation: --as-needed causes DT_NEEDED tags to only be emitted for libraries that satisfy some symbol reference from regular objects which is undefined at the point that the library was linked. This fix next lines in build log: dpkg-shlibdeps: warning: package could avoid a useless dependency if debian/bacula-fd/usr/sbin/bacula-fd was not linked against libwrap.so.0 (it uses none of the library's symbols). > > 17. Add build-depends for LZO support. > > > nak. Ok. i revert it. And the next question. Currently bacula can't built on hurd platforms. Should i include fix for this problem into next upload, intended for wheezy or not? -- with best regards, Alexander Golovko email: alexandro@ankalagon.ru xmpp: alexandro@ankalagon.ru
Attachment:
signature.asc
Description: PGP signature