[debian-mysql] Bug#838914: Bug#838914: mariadb-10.0: testsuite failures on mips/mipsel

[Kristian Nielsen]

FYI,

James Cowgill <jcowgill at debian.org> writes:

> mips-machine.patch
> Fix the value of DEFAULT_MACHINE on 32-bit mips platforms.

I've upstreamed this patch, it should appear in MariaDB 10.0.28.

> mips-groonga-atomic.patch
> Link groonga against libatomic as it uses 64-bit atomic operations not
> supported on mips without that helper library.
>
> mips-connect-unaligned.patch
> This is a rather large patch attempting to fix some unaligned memory
> accesses in the CONNECT storage engine which cause bus errors on mips
> (and probably other platforms which prohibit unaligned memory access).

Mroonga and CONNECT are storage engines that I think are developed
independently from MariaDB - ie. MariaDB is not the real upstream. So I've
sent mails on the maria-developers@ list with the patches, asking how these
can be properly upstreamed.

  https://lists.launchpad.net/maria-developers/msg10020.html
  https://lists.launchpad.net/maria-developers/msg10021.html
  
> mips-unstable-tests.patch
> Remove various tests from mysql-test/unstable-tests which now pass on mips.
>
> debian-mips-unstable-tests.patch
> Updates debian/unstable-tests.mips*
> I've added the 'connect.mysql_index' test to the mipsel unstable tests.
> It seems to fail sporadically on mipsel. It usually takes about 5
> retries to fail, but sometimes less. I'm not sure what the cause is here.

I chose not to upstream these two. I'm frankly not too impressed with these
"unstable-tests". Anyway, I think it's fine to carry these patches
separately in Debian for now.

Hope this helps,

 - Kristian.