triplets package

triplets.clear_caches() → None

Drop all in-memory engine state (see module docstring).

triplets.cache_scope()

Bound engine-state lifetime to a block: entries created inside are dropped on exit; entries that existed before the block are kept.

triplets.engines() → dict[str, dict]

What each subsystem’s “auto” resolves to, plus the available alternatives.

Returns {kind: {engine, source, policy, auto_order, available, unavailable, aliases}}. policy="input" subsystems (tools, exporter_csv) follow the input object’s flavor per call, so they report engine: None and source: "input" — there is no global pick to show.

triplets.set_engine(**engines_by_kind) → None

Global engine override per subsystem, e.g. set_engine(parser_cimxml="python_lxml_arrow", sparql="rdflib"). None or "auto" restores auto-selection.

Per-call engine= still wins over this. Process-global and intended for startup configuration; loads the engine eagerly so a bad name or missing dependency fails here, not at first use. Note per-call capability constraints still apply (e.g. cimxml datatypes=True requires the python_lxml engine regardless of the override).