Context Ownership and Normalization#

A Context owns three pieces of execution policy:

  • ops - a BackendOps instance such as NumpyOps;

  • dtype - the backend-normalized default representation dtype;

  • check_level - the ordered runtime-validation policy.

It does not own array memory, devices, gradient state, or sparse storage. Those belong to the backend arrays themselves.

Normalization#

spacecore.normalize_context(ctx) accepts a concrete Context, backend name, backend family, or None. Concrete contexts are copied and their dtype is sanitized by their backend. Backend names such as "numpy" and "torch" create a new context with that backend. None returns the global default context.

Context.asarray(x) and Context.assparse(x) convert user input into the context backend and representation dtype. They reject complex-to-real narrowing unless the caller first extracts the real part explicitly. Space constructors keep the normalized context and use it for later array construction and validation. Spaces expose their mathematical real/complex contract separately through space.field.

Accepted as-is versus converted#

User arrays passed to operations such as space.add or A.apply are not silently converted. The context’s check level selects backend, shape, dtype, recursive membership, and strict numerical validation. Under "none", user values are forwarded and backend errors may surface later.

Constructors and explicit conversion are the places where SpaceCore converts: ctx.asarray(...) creates dense arrays, ctx.assparse(...) creates sparse arrays where supported, and obj.convert(new_ctx) rebuilds context-bound objects under a target context.

Guarantees#

A constructed space or operator guarantees that it stores one normalized Context and uses that context for its own helper methods. It does not guarantee that every future user-supplied array has that context unless the selected check level validates it or the user created it through the same context.

Example#

import numpy as np
import spacecore as sc

ctx = sc.Context(sc.NumpyOps(), dtype=np.float64, check_level="standard")
X = sc.DenseCoordinateSpace((2,), ctx)
x = ctx.asarray([1.0, 2.0])
X.check_member(x)