This is a possibility. I will look into it.

I've added a kvm-deb-jessie-ppc64le builder to buildbot. It's in the experimental section for now, but it is building successfully, though there are test failures:

• http://buildbot.askmonty.org/buildbot/grid?category=experimental&branch=10.1
• http://buildbot.askmonty.org/buildbot/grid?category=experimental&branch=10.2

The builder is setup similarly to our amd64 and x86 kvm builders, but runs on the POWER8 server we have on loan from IBM.

The one thing the builder doesn't have that our other POWER8 builders do have is the ATC compilers and libraries. I see that the unicamp repository does have packages for jessie at http://ftp.unicamp.br/pub/linuxpatch/toolchain/at/debian/dists/jessie/at10.0/ so my next task is to install them on the build VM and figure out a configuration that will work for this builder without breaking (or negatively impacting) the x86 and amd64 builders (they share build configurations in buildbot). After that this task should be done.

danblack: Am I right in thinking that at10.0 is what we should target now? Or should I go with at9.0? Thanks.

Yes at10.0 (based on gcc-6) is recommended (see MDEV-10807)

I have a patch for MTR to port from MySQL that does does rounding of numbers that seem to be different in the GIS results causing test failures.

Happy to advice as required.

We are building on Debian 8 Jessie ppc64le and we have repositories for 10.0, 10.1, and 10.2 available at https://downloads.mariadb.org/mariadb/repositories/ , so closing.

thanks dbart