Implement simple thread barriers to be used on systems where the POSIX implementation does not provide them. We default to the POSIX ones when detected at configure time.

Fixed the gravity bugs reported in #363. Need to find a good rebuild criterion for gravity-only runs.

Fix to #363. Drifting decision needs to use the same multipole information as the actual tree walk. Also fix some gravity_cache padding values.

There is a problem when the cpu mask is empty and using multiple MPI ranks on a node, in that case all the main threads end up pinned to one core during the reading of the initial condition (and any other tasks done before engine_init is called).

Force compiler to report shadowed variables. Removed all shadowed variables. Added the function scheduler_activate_send() to simplify the unskipping of MPI sends.

Reinstate missing pointer indirection for external gravity dependencies that had disappeared in merge.

Hopefully still quite conservative

Correct use of caches in the self-PP gravity task. Make sure we can run without vectorization. Make sure we can run without hydro.

Correct use of caches in the self-PP gravity task. Make sure we can run without vectorization. Make sure we can run without hydro.

Also note that we don't adjust this value for MPI, in tests the number is just too unpredictable

This is now the default

Lots of refactoring to tidyup code reuse

Previously we had a max of 96, much smaller numbers will work depending on the job

runner_dopair_grav() should abort if the distance between the cells is larger than the mesh smoothing scale.