On Wed, 2016-10-12 at 15:40 +0200, Thorsten Behrens wrote:
Dominique Leuenberger / DimStar wrote:
If LO can be split on the build level to a degree where it is not one massive build call, then that would be helpful.
Tough - I see little way to have that supported upstream, so it likely will be one massive hack. Would icecream/distcc be an option? With that, one can have a smallish builder (might need some extra mem), but running it with -j100 & icecream to fan-out the load to idle builders.
At least with the current setup of OBS I don't see a way to achieve this. A worker (an independent VM, no network connection) gets a job assigned to perform a build of a package. I don't know enough about the LO build system / source structure to come up with any reasonable suggestion unfortunately. I'm only bitten by the effects it has :) Anothe rissue we see compared to what 'normal devs' see: we always start from a fresh VM, so there is no 'cache' from earlier builds. The entire code base needs to be done all the time (I think currently it takes between 4 and 6 hours, depending on the worker it ends up on) Dominique