Concurrent polyhedral computations

Use this guide when running polyhedral computations from several Julia tasks or changing code that calls CDD, the polyhedral backend. TamerOp runs its CDD work one operation at a time, while surrounding computations can still run concurrently.

One internal execution lock coordinates this work across PL geometry and incremental rhomboid construction. Exact rational and floating CDD operations use the same protected boundary. There is no user option to disable it.

The boundary includes lazy conversion, point/facet materialization, incidence, volume, and centroid computation. Incremental rhomboid construction copies its facets into ordinary Julia data before continuing outside the boundary. PL caches may retain backend representations; later library accesses are also protected. Session and region cache locks are acquired before the CDD boundary. Code inside that boundary does not acquire a session or region cache lock.

This contract covers calls through TamerOp. Independent calls to CDDLib or CDD-backed Polyhedra operations in other packages do not participate in its lock. Run such external computations in a separate process if they overlap with TamerOp work. Keep CDD global solver, logging, and arithmetic configuration fixed after initialization. Accessing and modifying backend objects through raw fields is outside the supported concurrent API. Input arrays must not be mutated while a computation reads them; clearing a PL geometry cache remains an exclusive maintenance operation after its users finish.