Ids

Every marker has an int64 id that never changes, also when the marker migrates. IdAllocator(rank, size) gives rank r of n the ids r, r + n, r + 2n, ..., so ids are unique across ranks without communication.

from mpiparticles import IdAllocator, ParticleStore

store = ParticleStore(2, 3, id_allocator=IdAllocator(rank=1, size=4))

The rank and size are arguments, not read from the decomposition. When every rank holds its own copy of the grid (a decomposition on COMM_SELF, each rank then loads its own share of the markers), the decomposition has one rank per process, but the ids must still differ from rank to rank: pass the rank and size of the world communicator, or use IdAllocator.from_comm(world). The default allocator is the serial one (rank 0 of 1), which is only right on a single rank.

• allocate(count) returns count new ids as an int64 array.
• advance_past(ids) moves the counter above given ids. ParticleStore.add(..., ids=...) calls it, so arriving markers cannot collide with new ones.
• next_id is the next id to hand out; it is saved by ParticleStore.snapshot() and restored by restore().

Ids are sent as int64 during migration and sorted as int64, so they are exact above 2^53. clear() empties a store without resetting the counter: ids are never reused.