Profile cover photo
Profile photo
GNU Toolchain Build Robot
28 followers -
Our Mission: Keep the Toolchain Going!
Our Mission: Keep the Toolchain Going!

28 followers
About
Posts

Some database upgrade last night. That included dumping about 140 GB logfiles and re-importing that into the database. Fun stuff :)
Add a comment...

There were three new rtems targets added recently, of which two (arm, see http://toolchain.lug-owl.de/buildbot/show_build_details.php?id=355716, and powerpc, see http://toolchain.lug-owl.de/buildbot/show_build_details.php?id=355861) just build fine, while the nios2-rtems target (http://toolchain.lug-owl.de/buildbot/show_build_details.php?id=355729) is affected by an unrelated build problem which is also a problem for cr16, fr30, lm32, microblaze, moxie and rl78, see https://gcc.gnu.org/ml/gcc-patches/2014-09/msg01694.html).

Another new target, powerpc-wrs-vxworksmils, also worked without a glitch, see http://toolchain.lug-owl.de/buildbot/show_build_details.php?id=355875 .
Add a comment...

Started to rework some parts of the backend. The target lists, and which targets were already built with the current set of commits (gcc+binutils), used to be flat files. These will soon come from the database, as well.

As a side effect, it'll be quite simple to issue a specific target on a specific build host (or on a specific build group.) Should be done this weekend...
Add a comment...

The reworked backend using config-list.mk is running since some days, and it's results look quite well. So it's working :)
Add a comment...

All three builders using the config-list.mk backend are now in production again.

I also fixed another issue: I was building all the possible targets separately by make'ing the relevant output log file. That worked in general, but didn't for those targets that have OPTions set containing an '=' in them: These were always parsed as simple variable assignments, so that all targets were built. :/  I disabled those back then, but figured out how to make it work even for those. So: Say "hello" to 14 new targets being tested.
Add a comment...

I've just finished my tests with a new backend using the config-list.mk file. Basically, the pre-requirement is that the used host GCC must be as recent as the sources to be build by config-list.mk, so I now build a full GCC beforehand.

The scheduling algorithm will need some time to adope to the new runtime, though. And for now, only gcc76 uses the new backend, while gcc20 still uses the previous version. I'll update the later once gcc76 shows it to be working as expected. And maybe I'll even change the way the scheduler waits for the config-list.mk builds, since they take so much longer...

I think this is really a big step forward, and I'll start reporting issues with this setup again. There are, for example, warnings for uninitialized variable uses, not sure if they hold or if the warning itself is bogus.
Add a comment...

Weekend time: I'll do another round of updates starting now. That will bring down the robot to allow for a migration to PostgreSQL 9.4.
Add a comment...

The build robot should be fully updated and working again by now. Also got a new webserver config, hopefully I didn't break anything.

It will start building again as soon as it finds a new commit in the GCC or Binutils/GDB repos. (Polling interval is 15min.)
Add a comment...

More system upgrades tonight. I'll probably shotdown the scheduler for some hours, expect outages. :)

Should be up'n'running tomorrow morning (CEST.)
Add a comment...

Did a hugh database cleanup, that's why there was downtime recently . The next step will be a partial rewrite of the scheduler. We'll see when I find some time to finish that piece, too.
Add a comment...
Wait while more posts are being loaded