Post #1546605

@shelldozer @regehr assuming Ninja etc. work roughly the way I think they do... any OOM'd process should be rerun once only, in sequence, after the pool is starved of other work. That way, a failed retry would eliminate build parallelism as the cause. I like this idea. I'm on a laptop w. 32GB and 24 threads. I hit this plenty. Bonus: a user friendly error message!