Michel Hidalgo
Michel Hidalgo
> @hidmic should we should reactivate or drop this. > > I tend lean on the second option, since as long as we publish results, and replicability is possible through...
> I meant, having a 12 months old paused PR is not great marketing. Regardless of what we decide to do about benchmarking, we should tilt this PR one way...
It's starting to happen on AMD64 too, see https://build.ros2.org/job/Hbin_uJ64__beluga_amcl__ubuntu_jammy_amd64__binary/3.
I did some minimal digging, reducing `-ftemplate-depth` (maximum number of template expansions). IIUC `gcc` defaults it to 900. I can take it down to 100. Compilation breaks at 50, on...
I've so far failed to reduce template nesting. I think we'll have to go the explicit instantiation route instead, as you suggested @nahueespinosa, and hope we can cut enough bloat...
@glpuga it's the templates. `clang` does a better job than `gcc`, memory consumption-wise. My attempts to optimize it failed. Explicit template instantiation may help.
Got another death on the farm https://build.ros2.org/job/Jbin_uN64__beluga_amcl__ubuntu_noble_amd64__binary. We gotta fix this.
I've been following that thread. It basically confirms our suspicion: the allocation build jobs get on buildfarm VMs is small-ish for today's standards. We need the compiler to use less...
What's going on with cluster based estimation? :eyes:
I haven't received any build failure notifications from the buildfarm in a good while. Either @nahueespinosa's efforts paid off or the ROS buildfarm team bumped compute resources for their VMs....