Without looking very closely at what is happening I can't suggest anything. It's unlikely to have anything to do with whether the GC is running with one or several threads. That's only likely to have an effect on how long each GC takes.
David
On 24/03/2016 02:31, Michael Norrish wrote:
As per, for example,
https://travis-ci.org/HOL-Theorem-Prover/HOL/jobs/118129741
we are seeing "Out of store" aborts with builds that work fine under 5.5.2. The example above is on a Linux virtual machine. We've also seen them on real (i.e., not virtualised) OSX machines.
Is there something obvious we might be doing that could be causing this?
We're running all our sessions with --gcthreads=1 as per discussion in
http://permalink.gmane.org/gmane.comp.lang.ml.polyml.general/976
Michael