When
%attr(-,mysql,mysql) %{_datadir}/mysql-test
is used,the path, *and everything on it* is added
Only when prepended with '%dir', it won't added anything else than the directory itself
The mistake resulted in all files under
%attr(-,mysql,mysql) %{_datadir}/mysql-test
to be present in both subpackages 'mysql-test' and 'mysql-test-data'.
Explanation of the options per:
https://dev.mysql.com/doc/dev/mysql-server/latest/PAGE_MYSQL_TEST_RUN_PL.html
The MTR_MAX_PARALLEL environment variable, if set, specifies the maximum number
of parallel workers that can be spawned when the --parallel=auto option is specified.
If --parallel=auto is not specified, MTR_MAX_PARALLEL variable has no effect.
Explanation of the options per:
https://dev.mysql.com/doc/dev/mysql-server/latest/PAGE_MYSQL_TEST_RUN_PL.html
--port-base=P
The value may also be given with the environment variable MTR_PORT_BASE.
The variable MTR_PORT_BASE is a more logical replacement for the original variable MTR_BUILD_THREAD.
It gives the actual port number directly (will be rounded down to a multiple of 10).
If you use MTR_BUILD_THREAD, the port number is found by multiplying this by 10 and adding 10000.
Specify base of port numbers to be used; a block of 10 will be allocated. P should be divisible by 10;
if it is not, it will be rounded down. If running with more than one parallel test thread, thread 2
will use the next block of 10 and so on.
If the port number is given as auto, which is also the default,
mysql-test-run.pl will allocate a number unique to this host.
--port-base was added as a more logical alternative to --build-thread.
If both are used, --port-base takes precedence.
Explanation of the options per:
https://dev.mysql.com/doc/dev/mysql-server/latest/PAGE_MYSQL_TEST_RUN_PL.html
--max-test-fail=N
Stop execution after the specified number of tests have failed, to avoid
using up resources (and time) in case of massive failures. retries are not
counted. Defaults to 10, set to 0 for no limit.
--retry=N
If a test fails, it is retried up to a maximum of N runs, but will terminate
after 2 failures. Default is 3, set to 1 or 0 for no retries. This option has
no effect unless --force is also used; without it, test execution will terminate
after the first failure.
Small sub-package 'mysql-test' will remain architecture dependent
and will continue to carry all architecture dependent parts of the testsuite.
New huge sub-package 'mysql-test-data' will hold all of the
achitecture independent data and will be built as 'noarch'.
This will save over 1GB of redundant data per each package build.
Two reasons:
1) This license is included due to the Perl module 'Expect'. However I wasn't able to actually
locate it's sources in the MySQL source tarball. I believe it's a remnant of a code that was
removed in the past.
I've asked the MySQL upstream to verify my claim and either disprove it or fix the LICENSE
file in wihch they mention it.
For this reason I believe the 'GPL-1.0-or-later' also does not apply, so I'm removing it too.
I'll put the 'GPL-1.0-or-later' should upstream disprove my findings.
2) As per clarification on the Fedora License mailing list:
https://lists.fedoraproject.org/archives/list/legal@lists.fedoraproject.org/message/LBAZXYBAA4YHJ2GEWEP5S7QXVOB76MUR/
"
If your package is licensed under a known choice of two licenses and
one is a "good" license and one is a "bad" license, then the License:
field must reflect the "good" license only. This is highly uncommon in
Fedora packages apart from the case of Perl modules dual licensed
under the GPL and the Artistic License 1.0. In that case you must pick
the appropriate identifier for the GPL side (which in Perl modules
will typically map to SPDX "GPL-1.0-or-later"). You are encouraged to
include a comment memorializing this, for example:
\# Upstream project is dual licensed GPL | Artistic 1.0
"
After discussion with change owner, I've prepared the correct solution.
There is a difference between Fedora and CentOS Stream 10 / RHEL 10:
- in CentOS Stream 10 / RHEL 10, the change is self-contained by the
'openssl-devel' package itself, which now defines OPENSSL_NO_ENGINE
in the headers so new application builds don't use engine at all
- in Fedora, the headers containing the OPENSSL_NO_ENGINE definition
are put into a separate sub-package 'openssl-devel-engine', instead
of being put into the 'openssl-devel'
mysql-c99.patch
- directory "plugin/innodb_memcached" no longer exists
boost-1.76.0-fix_multiprecision_issue_419-ppc64le.patch
- upstreamed
ExcludeArch: %{ix86}
- MySQL 8.4 is not supported on 32-bit architectures
%ifnarch aarch64 s390x
BuildRequires: libquadmath-devel
%endif
- sub-package of gcc
- Only available on some architectures
%{_bindir}/mysql_upgrade
- was deprecated in 8.0
- now no longer available
%global runselftest 0
- disable testsuite during development time,
takes ages to complete on s390x and produces random failures on ppc64le
- I've checked that the result is sane
- testsuite enabled
- testcase checking disabled, as it produces huge amount of errors before each test even begin
these errors seems to be fully non-deterministic (random)
These packages are either not needed, or are brought in anyway by their *-devel sub-packgages.
It is possible to check whether the MySQL miss something by searching through the CMake checks, for "EXECUTABLE-NOTFOUND" keyword.
Some are expected to be missing. e.g. Git.
We don't want the MySQL to try to additional sources from web from git during compilation (not like it can anyway, but without the git, CMake won't even try)
List of currently not found executables, for future reference:
GIT_EXECUTABLE:FILEPATH=GIT_EXECUTABLE-NOTFOUND
DOXYGEN_EXECUTABLE:FILEPATH=DOXYGEN_EXECUTABLE-NOTFOUND
DOXYGEN_DOT_EXECUTABLE:FILEPATH=DOXYGEN_DOT_EXECUTABLE-NOTFOUND
CTAGS_EXECUTABLE:FILEPATH=CTAGS_EXECUTABLE-NOTFOUND
PATCHELF_EXECUTABLE:FILEPATH=PATCHELF_EXECUTABLE-NOTFOUND
When you try to use system version of the library, following error emerges:
| -- Performing Test HAVE_RAPIDJSON_WITH_STD_REGEX
| -- Performing Test HAVE_RAPIDJSON_WITH_STD_REGEX - Failed
| CMake Error at cmake/rapidjson.cmake:114 (MESSAGE):
| System rapidjson lacks some fixes required for support of regular
| expressions. See extra/RAPIDJSON-README for details.
| Call Stack (most recent call first):
| CMakeLists.txt:1952 (MYSQL_CHECK_RAPIDJSON)
| -- Configuring incomplete, errors occurred!
I specifically chose not to specify version in the bundled(rapidjson),
to avoid confusion that it's the same version as of the system variant of the library from package 'rapidjson'
Both are set to 'bundled' by default !
We really want to avoid bundling zlib.
On top of that, there is currently a problem with the bundled zlib, which results in a linker error during compilation.
We want to un-bundle rapidjson too, but:
| -- Performing Test HAVE_RAPIDJSON_WITH_STD_REGEX
| -- Performing Test HAVE_RAPIDJSON_WITH_STD_REGEX - Failed
| CMake Error at cmake/rapidjson.cmake:114 (MESSAGE):
| System rapidjson lacks some fixes required for support of regular
| expressions. See extra/RAPIDJSON-README for details.
| Call Stack (most recent call first):
| CMakeLists.txt:1952 (MYSQL_CHECK_RAPIDJSON)
| -- Configuring incomplete, errors occurred!
needs further investigation.
Installation path macros are not allowed in file dependencies.
Starting with F40, these break flatpak builds entirely because
dnf5 only recognizes file deps in /etc, /usr/bin, and /usr/sbin,
not /app/etc. At this point, mariadb-connector-c-config is the
sole provider of my.cnf.
https://docs.fedoraproject.org/en-US/packaging-guidelines/#_file_and_directory_dependencies
--
Based on commit 45b40b2 in rpms/mariadb10.11 authored by Yaakov Selkowitz on Apr 8 2024
45b40b2e9c
When building the package as alternative (non default) version
for a distro, we should not have provides that are virtual and
without version. Otherwise, packages that depend on these
virtual provides (like build requires of perl-DBD-MySQL),
could install the non-default version by accident.
