Ubuntu 14.04 packages (MariaDB 5.5.37) contain libmysqld., but both libmysqld.so and libmysqld.a are in 'libmariadbd-dev' package while in RPM (for example CentOS 6) libmysqld.a is in MariaDB-devel package and libmysqld.so* are in MariaDB-server package.
As a result everything that is build under Ubuntu 14.04 depends on 'libmariadbd-dev' (on development package!)
Are we going to merge this packaging ito MariaDB?

About "Fix Version: 10.0" - please note that the merge request is for the 5.5 series only. I'll do a separate merge request for 10.0 and open a new JIRA issue for that, as the contents in debian/* for 10.0 is different enough to justify a separate issue. But before I push the 10.0 branch I'd like to see 5.5 merged first so we can learn from possible issues that arise in 5.5 work and fix not repeat them for 10.0.

serg This hasn't seen much progress because I've been busy with constantly updating and improving the Debian/Ubuntu packaging. Also the turnaround for request/review has been slow. What if we would meet and have a 1-2 day sprint to get this merge reviewed and pushed in one go?

What if we also changed the strategy, and not merge the latest packaging into 10.0 and 5.5 but rather only on 10.1 and keep downstream and upstream Debian packaging in sync from 10.1 onwards?

serg Note that current https://github.com/ottok/mariadb-10.0/compare/upstream...master.patch is 45488 lines long..

Attached a complete review of the above: ottok.diff

Recent changes done to downstream packaging based on Serg's review:

commit ada2c8c84271283eb0f604820d5994c9a057784e

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 22:40:16 2015 +0300

    Recover mysql-upgrade dir/link handlig wrongly removed in f7caa041db

    The preinst is supposed to save possible mysql datadir link

    and the postinst is supposed to recover it.

    Also unify namig of the variables that reference the data

    and log directories.

commit d190985e7ba7621c99f877a9451943c6e2ae5d54

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 22:02:54 2015 +0300

    mysqld config: rename key_buffer -> key_buffer_size

    Short name version is deprecated. Use full name instead.

    https://mariadb.com/kb/en/mariadb/myisam-system-variables/#key_buffer_size

commit d8fb0f21e7e835336b37d9d3533d7276fe7a99b3

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 21:12:18 2015 +0300

    Clean up old cruft from rules file after review by Sergei Golubchik

     * Don't set or use MYSQL_BUILD_CC as they were hard-coded anyway

     * Remove unnecessary comment about how make translates -j option to --jobserver

     * Assembler functions are long gone and references to them should be removed

     * C and C++ should have same optimization levels, that is -O3

     * BIG-JOINS are not used anywhere, remove them

     * Make compilation comment nicer (visible via "show variables like 'version%';")

     * Make -DDEB contain actual distro name, might be of later use and anyway better

       than just simple '1' (true) value

     * Not needed to have TAOCRYPT_OPT in CFLAGS, taocrypt is C++

     * Libwrap is disabled by default

     * INFO_BIN and INFO_SRC files are not used anymore, no need to install them.

     * Use INSTALL_MYSQLTESTDIR instead of manual mv line

commit a9bbb0c5dac55aefcad7131dc29a8b2a809cc3ed

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 18:23:25 2015 +0300

    Unified config file layout with upstream .cnf layout

    Now the packages mariadb-common, mariadb-client and mariadb-server each

    install their own cnf file. A system will not have extra cnf files around.

    Also the .cnf files list all possible stanzas that can be used to

    compartmentalize settings for different binaries.

    Role model was upstream

    https://github.com/MariaDB/server/tree/83ba48b7c670f6dba465325cafd808c91f551544/support-files/rpm

commit 59c96c74e098767b2e01c0fa6b2e70b8bce592cc

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 18:00:27 2015 +0300

    Simplyfy overkill utf8mb4 settings

    From https://dev.mysql.com/doc/refman/5.6/en/charset-connection.html

      A SET NAMES 'charset_name' statement is equivalent to these three statements:

      SET character_set_client = charset_name;

      SET character_set_results = charset_name;

      SET character_set_connection = charset_name;

    Therefore it is probably overkill to have 'SET NAMES'.

    Also using both - and _ in variable names in useless as they translate

    to the same variable.

commit 2f71f6834d93d9591cc187d6cce6e3dd7268bd33

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 17:46:24 2015 +0300

    Update upstream info in d/copyright

commit df02e679e29dabd1926f477c70d984cc168fc159

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 17:27:29 2015 +0300

    The test suite must depend on the exact MariaDB version as tests are version specific.

    Depending on latest or any newer version is wrong, as tests might not be compatible

    with some never program versions.

commit e3fe9539f010270aac5aa6e70a2c0e561c749df0

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 17:16:11 2015 +0300

    Wrap and sort

commit ed276550168e9933f9e804bcf5f10da0cf4fe648

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 16:24:30 2015 +0300

    Created package libmariadbd and ship libmysqld.so.18 in it.

commit 0ed651b0cd8bafed1d43be339d499cc0214e54e5

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 15:39:47 2015 +0300

    Add libdbd-mysql-perl as Depends instead of Suggests as libdbi-perl needs it.

    The package libdbi-perl is a dependency and it is useless unless libdbd-mysql-perl

    is also installed. Thus both should be in Depends. These lines are now also

    identical to those in the mysql-5.6 control file.

commit be79bfaf31b9839205cb34f6bb1244c2f9b36ef5

Author: Otto Kekäläinen <otto@seravo.fi>

Date:   Tue Jul 21 08:39:45 2015 +0300

    Updated d/changelog

(full log at https://github.com/ottok/mariadb-10.0/commits/master)

Preliminary branch for upstream merge is at https://github.com/MariaDB/server/compare/10.1...ottok:ok-debpkg

serg Pull request filed at https://github.com/MariaDB/server/pull/97

It fixes at least MDEV-8667 and MDEV-6326 right now, and the new control file will prevent lots of future problems.

As the control file logic is now almost in sync with that of the official Debian version, this pull request will prevent problems people might have who try to upgrade from 10.0 in Debian/Ubuntu to 10.1 from mariadb.org repos.

Once this PR is accepted, I'll continue on the merge for other parts.

merged

sergratzpo I think I should continue to do another round of backporting fixes to 10.1 to close multiple issues linked above so that 10.1 would ship with less broken packaging. However the current 10.1 master head fails in multiple tests and I don't have a clean starting point: https://buildbot.askmonty.org/buildbot/grid?branch=ok-debpkg-10.1&category=main

None of those fails are due to Debian packaging. I'll wait until 10.1 master is all green before I start working on my branch.

https://buildbot.askmonty.org/buildbot/grid?branch=ok-debpkg-10.1&category=main green again.

I've been working on the next step for 5 days now, should soon be done. WIP at https://github.com/ottok/mariadb/commits/ok-debpkg-10.2

PR available at https://github.com/MariaDB/server/pull/251

There is still quite a lot to do to close this issue and all the issues linked to it, but this is a big milestone in that process.

Changing fixVersion. Everything that was supposed to be done for 10.2 has been done. The rest will go into 10.3.

Work continues now on 10.3 branch by me and oerdnj

otto, may I ask you if it's normal that this issue that is not closed is blocking so many issues that are closed?

julien.fritsch Yes, this issue is still open. We need to fix e.g. https://jira.mariadb.org/browse/CONC-304 and https://jira.mariadb.org/browse/CONC-456 and then we are closer to closing this one. I expect this issue to be closed during the 10.5 cycle (assuming I get some help with a couple of things that should be changed upstream).

otto my question is more my question is more: how can these issues that have a link "blocking by" this one, can be closed it this one is not closed already?

Issues marked "blocked by" are things that need to be fixed upstream before we can finalize the merge.

Issues marked "blocks" are bugs that are already fixed in Debian downstream and will automatically be fixed upstream at MariaDB.org when this issue is finalized.

Hi otto would the "upstream" help then, maybe here to explain that indeed, this issue?

I am both upstream and downstream. Currently things holding back this are my PRs pending to be reviewed/accepted and the issues that are still open that need help from somebody else. Most pressingly https://jira.mariadb.org/browse/CONC-304 and https://jira.mariadb.org/browse/CONC-456 as stated above. I'll try to book 2h of time with Georg to get those CONC done, they are fairly simple path and filename changes, but just need to be done in a way that is backwards compatible.

OK, clear thank you

Side notice: seems mtr depends on patch and libtool-bin, need to add that as a run-time dependency:

nano +2175 mysql-test-run.pl

sub executable_setup () {

  $exe_patch='patch' if `patch -v`;

  #

  # Check if libtool is available in this distribution/clone

  # we need it when valgrinding or debugging non installed binary

  # Otherwise valgrind will valgrind the libtool wrapper or bash

  # and gdb will not find the real executable to debug

  #

  if ( -x "../libtool")

  {

    $exe_libtool= "../libtool";

    if ($opt_valgrind or $glob_debugger or $opt_strace)

    {

      mtr_report("Using \"$exe_libtool\" when running valgrind, strace or debugger");

    }

  }

Also, for all tests to run, they should be run as user 'mysql' with access to pam_auth_tool, and again with another user that is not in user.mysql by default. Also some tests only run when mtr is invoked with --big-test, there is IPv6 available, there is a Cassandra back-end running etc. I wonder how big tests should be run to validate that there are no testable regressions in this MDEV once all patches are done (it is going to be huge).

This MDEV can be closed when all of these are green:

https://salsa.debian.org/mariadb-team/mariadb-server/pipelines/124140

patch can be "suggested" or "recommended", because tests won't fail if it's not present, they'll be automatically skipped

> Side notice: seems mtr depends on patch and libtool-bin, need to add that as a run-time dependency:

Tested libtool-bin, but it did not provide any extended traces, on the contrary the trace lost the filename paths:

Tested by running mtr (without --force and without --parallel) and then issuing in another windows `killall -s 6 mariadbd` comparing baseline and situation after running `apt install libtool-bin`.

`patch` is already a suggests, no need to research it.

All but https://jira.mariadb.org/browse/MDEV-5484 and merging of patches (https://jira.mariadb.org/browse/MDEV-14900) is done, but I deem this issue now closed enough.