3.0.2: resolved values follow the states registered above a state

[trissim]

Defect

An ObjectState built before its ancestor state was registered kept the resolved values it computed at build time (the outer default) after the ancestor registered. The same held for unregistering an ancestor, replacing it at its scope, and clearing the registry.

Cause

A state resolves its inherited fields from the states registered at its ancestor scopes, read from ObjectStateRegistry when the snapshot is computed. Nothing tied the snapshot to that set: register, unregister and unregister_scope_and_descendants told no descendant, and the snapshot did not record what it was computed from. Only edits of an ancestor's values reached descendants (invalidate_by_type_and_scope).

Fix

• ObjectStateRegistry.registered_ancestors(scope_id) is the one definition of the chain; get_ancestor_objects and get_ancestor_objects_with_scopes derive from it (two copies of the loop removed).
• Each resolved snapshot records the registered ancestors it was computed from (ObjectState._ancestors_resolved_from, written in _compute_resolved_snapshot). _ensure_live_resolved and the saved reads compare them with today's and resolve live and saved again in full when they differ, then update dirty fields and notify.
• register, unregister and unregister_scope_and_descendants ask registered states below the changed scope to make that comparison at once (_resolve_again_below), so on_resolved_changed subscribers hear without a read. A state whose ancestors did not change does no work; registering a state does not resolve it again.

No caller invalidates anything and there is no refresh method to remember. The per-read cost is one walk of the scope's prefixes.

Tests

tests/test_resolved_values_follow_registered_ancestors.py: 30 of its 31 passing tests fail on main. Read kinds are asserted separately: stored value, raw field of the built object, uncached attribute access, state.fields, live and saved resolved values and objects, source (scope, class), dirty fields.

Mutations covered: ancestor registered late (descendant never registered / registered before / after it), unregistered (unregister, cascade, clear), registered again with another value, replaced at its scope, registered with unsaved edits (live and saved differ), edit / save / restore of the ancestor, descendant scope_id moved under another ancestor, a context entered after the build, a field invalidation pending together with an ancestor change, notification at registration, no work when ancestors are unchanged.

Open, as strict xfail with reasons (#53): unregistered descendant after an ancestor edit; global config replaced through the setters; a state resolved inside an entered context. Widening the comparison to the ancestors' objects would fix the first two at about 4x the cost per descendant per ancestor edit (measured, in #53).

Precedence order (owner decision: class first)

• context_system.rst states the order once, with a Leaf/Base, plate/global example; the resolver docstring and configuration_framework.rst refer to it. The sentence in configuration_framework.rst that described scope first is corrected.
• The crossed-pair tests from Test that field resolution is one precedence order; characterise the crossed case #52 are the policy now; labels changed, assertions and the resolver unchanged.

Release

3.0.2. Full suite in a fresh venv (Python 3.12): 343 passed, 5 xfailed.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.