Note: this change is needed only in compat levels < 13

This fixes build for some reverse-dependencies when using mysql-8.0 backend, such as boinc

Specify contributions from different people in a more standard way

MySQL's has been changed to use standard atomics handling.
Part of the patched code has been removed. It was a
compilation test, so should not be needed.

The test checks the year of a timestamp, expecting it to be 2020.
This is fixed by upstream in an upcoming release.

Ref https://bugs.mysql.com/bug.php?id=102243

0 was used for infinity in older versions of systemd, but newer versions
should have it set explicitly.

For large databases the system might hit the timeout limit,
causing systemd to kill the process, which could potentially
lead to a corrupted database. Set to 0 (infinity)

Resolves LP#1882527

The flag is used to default to system for any dependency where MySQL
can use either bundled or system package.

Some new plugins have been added by upstream, and a sample file has been moved.

The issues with expired test certificates and client library
segfaulting for certain custom charset definitions have both
been fixed upstream.

Update to upstream version '8.0.22'
with Debian dir 39a949ffabf2c079e9d759d0d055ff2862c27a87

These are required to use the protobuf library provided by the system,
as enabled in the previous commit.

Using the bundled version caused problems with debug symbols,
since both router and server bundled the same so-file.

The test doesn't correctly use the --client-bindir flag to see where it should
look for testing binaries, so it fails because it can't find mysqltest_safe_process.
Ref: https://bugs.mysql.com/bug.php?id=100532

debian/patches/revert_faster_tls_model.patch: revert an upstream commit
that switches the tls-model to initial-exec. This is causing "cannot
allocate memory in static TLS block" errors when attempting to load the
library in certain scenarios.

LP: #1889851

d/p/use-largest-lock-free-type-selector-on-riscv.patch: Force the use of
Largest_lock_free_type_selector instead of Lock_free_type_selector when
compiling for RISC-V, since the latter will cause a compilation failure
due to RISC-V's inability to provide the always-lock-free property for
some specific types.

Update protobuf lib to 3.11.4, add new protobuf plugin for router and
test component for test suite.

Adjusted by Robie Basak: use wildcards for versions to avoid having to
keep bumping this. Taken from Ubuntu with thanks to Marc Deslauriers.

Ref: https://bugs.mysql.com/bug.php?id=100274

Some certificates used in mtr expired 2020-06-30, causing test failures.

The LICENSE file merely contains an explanation of why the timezone
database is not under BCP 78

Ref: https://github.com/mysql/mysql-server/blob/8.0/LICENSE#L1929

Written by Lars - thanks!

This happens because mariadb installs charset files to the same place MySQL does, so if
mariadb is combined with libmysqlclient (and possibly vice-versa), the client library
may attempt to use those charset files.

A bug in the parser combined with the differences in charset data can lead to the client
library segfaulting. The patch simply prevents the segfault by checking that the collation
pointer is not NULL, until the underlying issue in the parser is fixed upstream.

Ref: https://bugs.launchpad.net/ubuntu/+source/mysql-8.0/+bug/1877504

Seems a &p=23 parameter is now needed for the gpg signature url

The notes explain that MySQL bundles header files from these projects
because using system libraries doesn't work.