Optional Backend Routing

This showcase models a common scenario: your library supports multiple engines, but only some users activate optional ones.

Problem

You want a fast default path while still supporting an optional backend like OpenMM.

Pattern

Use a conditional dependency check and keep backend import lazy:

from depdigest import dep_digest

@dep_digest("openmm", when={"backend": "openmm"})
def run(job, backend="native"):
    if backend == "openmm":
        import openmm
        return run_openmm(job)
    return run_native(job)

Why this works

• Users on default backend pay zero optional-import cost.

• Optional dependency is enforced only on the path that needs it.

• Missing dependency errors remain explicit and actionable.

Where to apply

• computational engines;

• export backends;

• optional acceleration providers.