Issue with #87 fix: do not constrain resources in remote jobs ?

[pvandyken]

I ran into #87 while dealing with failing tests in snakemake 3421.

I'm a bit confused regarding this: to use memory as an example, if resource constraints are not passed via CLI to a sub-snakemake process, that process will be ignorant regarding potential constraints on memory enforced by the containing job. The group-resource calculator assumes that jobs will be run in linear rather than parallel when insufficient memory is provided, but this assumption will not be enforced by the sub-process if it is not informed of the constraint. This would lead to out-of-memory errors.

Is there some other way resource constraints are provided? If there is uncertainty, I can try to create a test scenario.