When generating ninja, the -j will control the local pools depth. This is not intuitive, and gets obscured into the ninja file. At least, Ninja should be setting local pools to the available cores, and ignoring -j passed to scons during generation.

Possibly another interface could be made for better control, like an SCon Var such as NINJA_LOCAL_JOBS, which limits local jobs in the generated ninja.

Other things to consider regarding this aspect, is that custom pools could also be made allowing for more local concurrency. Should a generalized ninja tool try to handle this by setting pools to a proportion of the NINJA_LOCAL_JOBS or NINJA_MAX_JOBS?

Also when ninja is limiting its jobs, its not obvious what is the current limit. Should a new target be made to easily extract this info? A build edge which echos the current pools and limits?

ninja list_pools