Reflection and steering

The Python surface describes itself in the &metta space. Registered operations appear as (op name arity kind) facts. Functions compiled with @m.define appear as (defined space name). Standing queries appear as (subscription space pattern on). Each fact is removed when the operation, definition, or subscription is removed.

Because &metta is an ordinary space, MeTTa programs can inspect the Python integration with normal matching. The direction also reverses: a Python standing query on &metta can watch for control atoms written by a MeTTa program. The program then steers the integration through the same fact and subscription mechanisms used elsewhere.

Object reflection has a separate but compatible surface. install_reflection_ops(m) adds py-attr and the two-mode py-field. With a bound field name, py-field reads that field. With an unbound field name, it enumerates an object's fields as relation answers.

metta.spaces.object_view(obj) gives those fields storage semantics instead of operation semantics. The view emits live (py-field obj name value) atoms, joins with stored spaces through the ordinary spaces.union combinator, and maps added field atoms to setattr. Python mutation therefore changes the next match without another projection.

An integration can also register a reflector that lowers its own structure into facts. metta.integrate.reflect(m, name, object) chooses the registered reflector and returns the number of facts written.

Use metta.ops for operation reflection, metta.subscribe for standing queries, and metta.integrate for object reflection.