Release 3.0.0: one field resolution, loud context stacks, per-context blocks, registry-owned lazy counterparts

[trissim]

Fixes #29, #30, #31, #33, #34, #35

Release branch for 3.0.0. Public names are removed, so the major version changes (__version__ in src/objectstate/__init__.py, the one place it is declared).

Supersedes

• Context stacks: with-only build_context_stack, scoped ancestors only, no layer skipped or substituted (#31, #34 items 1-2) #41 fix/34-31-context-stack-loud (Building a context continues after failures with a layer missing or substituted #31, build_context_stack returns an open ExitStack; remove the deprecated ancestor format and legacy_entries #34 items 1-2)
• One field resolution with its source; caches owned by the context (#29, #33) #49 fix/29-33-one-resolver (Attribute reads and displayed provenance can resolve the same field to different values #29, Resolution caches are process-wide but depend on the per-thread context #33), which was stacked on Context stacks: with-only build_context_stack, scoped ancestors only, no layer skipped or substituted (#31, #34 items 1-2) #41
• Keep atomic and deferral block state per context (#30); rename legacy_entries (#34 item 3) #40 fix/30-34-atomic-block-context (atomic and defer_live_invalidations share one counter across threads #30, build_context_stack returns an open ExitStack; remove the deprecated ancestor format and legacy_entries #34 item 3)
• Find a class's lazy counterpart through the base/lazy registry #42 fix/35-lazy-registry (The lazy counterpart of a class is looked up by constructed name #35)

Merged in that order with merge commits. Git reported no textual conflict. #49 and #40 both edit object_state_registry.py in different places (#49 removes three calls to the deleted cache-clear functions; #40 moves the block state into ContextVars) and #41, #49 and #42 all edit lazy_factory.py; each merged cleanly and every PR's tests pass together.

Added on this branch

get_provenance for a state without a scope id. With #49, an ObjectState without a scope id has its own object as a layer, so an unset field inherits from another config of the same object, as it already did for a state with a scope id and for an attribute read on the object. The resolved value was right, but get_provenance still returned None for it (its scope_id is None branch predates #49), which callers read as "set on this field". It now returns (None, source_type): the state's own scope id, and the type that supplied the value. tests/test_scopeless_state_resolution.py runs each case with and without a scope id:

Read	Value	Source
nothing set on the object	global value	("", WellFilterConfig)
sibling config of the same object sets it	sibling's value	(state.scope_id, WellFilterConfig); (None, WellFilterConfig) without a scope id
the field itself is set	its own value	None (not inherited)

Two of these cases fail without the fix, both for the state without a scope id.

The third id()-keyed cache. LiveContextResolver keyed answers by (id(config_obj), attr_name, ids of the context objects, token). Each answer is now kept together with the config object and context objects it was computed from and found again with is; holding the objects means an address cannot be reused while its answer is kept. Answers belong to one cache token and are dropped when a new token arrives (before, every token's answers were kept for the life of the resolver). tests/test_live_context_resolver_answers.py; the reused-address test fails on the old code.

Not changed: object_state_registry.py:1054 puts id(state) in an atomic block's entry signature. That is a comparison within one block while the states are held, not a cache key.

Checks run locally (CPython 3.12)

• pytest -n 2: 297 passed.
• ruff check src/ --select E9,F63,F7,F82: clean.
• scripts/check_ruff_added_lines.py --base origin/main: no diagnostics on added lines.
• black --check src/ (non-blocking in CI): 23 files would be reformatted.
• Docs build and mypy: not run (Sphinx is not installed in the environment used).

What callers must change

Full ledger with file:line for every site: OS-3.0-caller-ledger.md (kept with the OpenHCS integration notes). Summary:

pyqt-reactive (10 sites)

• services/widget_service.py:310-316: build_context_stack(..., ancestor_objects=...) raises TypeError. Use ObjectStateRegistry.get_ancestor_objects_with_scopes(...) and ancestor_objects_with_scopes=.
• pyproject.toml:32: objectstate>=2.0,<3 excludes this release. requirements-ci.txt:2 names 2.1.1.
• tests/test_preview_formatters.py:10,229,230,279: GROUP_ABBREVIATIONS_REGISTRY is removed; declare with @abbreviation.
• tests/test_function_list_editor_tokens.py:32-34: assigns removed _atomic_* attributes.
• Review: clickable_help_components.py:534,707,735 can now receive (None, source_type) from get_provenance; the same-window branch already handles it.

formbridge (3 sites)

• strategies/preview_formatting.py:28,31: imports GROUP_ABBREVIATIONS_REGISTRY; call objectstate.lazy_factory.get_group_abbreviation.
• pyproject.toml:29: version pin.

pycodify: none.

OpenHCS (10 sites, none in openhcs/, benchmark/ or scripts/)

• pyproject.toml:104: version pin.
• tests/unit/pyqt_gui/test_source_bindings_editor.py:3180-3181: asserts that only the edited cell flashes. It pinned the old disagreement between the two resolvers: with a scope id the same edit already flashed two cells on main, because step_source_bindings_config.source_filters inherits from source_bindings_config of the same object. Assert both paths.
• tests/unit/test_time_travel_navigation.py:53-55, tests/unit/pyqt_gui/test_ui_agent_bridge.py:3738-3740, tests/unit/pyqt_gui/test_parameter_form_manager_cross_window.py:178: assign removed block-state attributes.

Removed names: ancestor_objects=, resolve_field_inheritance, clear_mro_resolution_cache, invalidate_mro_cache_for_field, clear_extract_all_configs_cache, set_current_temp_global, _has_concrete_field_override, GROUP_ABBREVIATIONS_REGISTRY, LAZY_CONFIG_PREFIX, TimeTravelChangeSet.legacy_entries (now states_needing_navigation), ObjectStateRegistry._atomic_depth / _atomic_label / _atomic_triggering_scope / _atomic_snapshot_requested / _atomic_entry_signature / _deferred_invalidation_depth (read current_atomic_block()), and six uncalled private functions listed in #41.

Changed behaviour: failures that were swallowed now raise (NoActiveContextError, constructor errors in config_context and merge_configs, get_base_global_config); get_context_layer_stack() returns a ContextLayerStack, not a tuple of pairs; build_context_stack sets nothing until entered; atomic blocks belong to the thread or task that opened them. The four PR bodies list every item.

Open point

tests/unit/test_compilation_session.py::test_compiler_streaming_config_snapshot_preserves_inherited_port (OpenHCS) failed once in a full -n 4 unit run against this branch and not in the matching run against main. It passes alone and with its file. Not reproduced and not explained; first thing to check is that build_context_stack(use_live=False) now reads only the saved global config.

Not merged, tagged or released.

🤖 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.