OS3: callers of ObjectState 3.0.0

docs/refactor/consolidation-20261009/OS3-callers.md

@@ -0,0 +1,180 @@
+# OS3: callers of ObjectState 3.0.0
+
+Base: `origin/refactor/U6-gui-integration` (PR #1219), not `main`.
+Task list: the 3.0.0 caller ledger (`OS-3.0-caller-ledger.md`, kept with the
+integration notes) and the release PR OpenHCSDev/ObjectState#50, head `abe763c`.
+
+## What is wrong
+
+ObjectState 3.0.0 removes public names and makes one resolver answer every
+read. Its callers still use the removed names:
+
+| Caller | Sites that break | Where |
+|---|---|---|
+| formbridge 0.1.1 | 3 | `strategies/preview_formatting.py` walks `GROUP_ABBREVIATIONS_REGISTRY`; version bound |
+| pyqt-reactive 0.7.0 | 10 | `services/widget_service.py` calls `build_context_stack(ancestor_objects=...)`; bound; CI candidates; 7 test lines |
+| pycodify | 0 | |
+| OpenHCS | 10 | version bound; 9 test lines |
+
+The ledger's "no hits" entries were re-checked with one grep over all four
+callers for every removed name: no site beyond the ledger.
+
+## The rule for config reads
+
+A config value has three reads. Each site says which one it needs.
+
+- Raw: only what was set on this object (`state.parameters[path]`, saved `.py`).
+- Resolved: the effective value after inheritance (`state.get_resolved_value(path)`,
+  or attribute access on a lazy object inside its context).
+- Widget preview: the resolved value together with its source
+  (`state.get_provenance(path)`), while the field's own value stays unset.
+
+Both state reads come from the one resolution `resolve_with_provenance`, which
+returns `ResolvedField(value, source_scope, source_type)`.
+
+## Target
+
+Every caller reads through the one owner of each fact, and the three
+repositories release against ObjectState 3.0.0 together.
+
+### formbridge 0.1.2 (OpenHCSDev/formbridge#9, `1e73e6c`)
+
+- `strategies/preview_formatting.py`: the module's own walk of
+  `GROUP_ABBREVIATIONS_REGISTRY` is deleted. `FormattingConfig.container_abbr_func`
+  defaults to `objectstate.lazy_factory.get_group_abbreviation`.
+  Certificate **derived**: ObjectState's function does the lazy-to-base step,
+  the MRO walk and the class-name fallback that the deleted copy did.
+- `objectstate>=3.0,<4`. New `requirements-ci.txt` (pyqt-reactive's convention)
+  installs ObjectState at `abe763c` in CI until 3.0.0 is published.
+
+### pyqt-reactive 0.7.1 (OpenHCSDev/PyQT-reactive#28, `1fa382e`)
+
+0.7.1 and not 0.8.0: no public name is removed.
+
+- `services/widget_service.py` (preview read). The removed
+  `build_context_stack(ancestor_objects=...)` call is not translated; it is
+  deleted. The site rebuilt a context stack by hand, asked for placeholder
+  text, then read the attribute again for the value: a second resolver beside
+  the state. `WidgetService.show_resolved_value_as_placeholder` shows
+  `state.get_resolved_value(path)`; the two refresh paths in
+  `ParameterOpsService`, which each repeated resolve-and-format, call it too.
+  Certificate **derived** (the state's resolved read). One visible change: the
+  value-update path now uses the form's `placeholder_prefix`; it used the
+  default prefix.
+- `widgets/shared/clickable_help_components.py` (source read). The ledger
+  marked this "review: the same-window branch already handles
+  `(None, source_type)`". It did not: the target path was computed first from
+  `ObjectStateRegistry.get_by_scope(None)`, which is the global state. The
+  source state is now the widget's own state when the source scope is its own
+  scope. `get_provenance` is read once per action; the tuple-shape re-check and
+  the blanket `except` are deleted (certificate **contract**: the return type
+  is `tuple[str | None, type] | None`).
+- `widgets/shared/abstract_manager_widget.py`: unused `LiveContextResolver`
+  instance deleted (certificate **unused**).
+- Bounds `objectstate>=3.0,<4`, `formbridge>=0.1.2,<0.2`; `requirements-ci.txt`
+  names both branch commits.
+- Tests: `@abbreviation` instead of writes to the removed registry; removed
+  `_atomic_*` assignments deleted; new `tests/test_provenance_navigation.py`.
+
+The supply map (`--focus abbreviation`, `--focus WidgetService`) reported no
+second provider, copy or forwarder for either class. It sees classes, not
+functions, so the three placeholder copies were found by reading.
+
+### OpenHCS
+
+- `pyproject.toml`: `objectstate>=3.0,<4`, `formbridge>=0.1.2,<0.2`,
+  `pyqt-reactive>=0.7.1,<0.8`. Submodules at `abe763c`, `1e73e6c`, `1fa382e`.
+- `tests/unit/test_time_travel_navigation.py`,
+  `tests/unit/pyqt_gui/test_ui_agent_bridge.py`,
+  `tests/unit/pyqt_gui/test_parameter_form_manager_cross_window.py`: assignments
+  to the removed block-state attributes deleted.
+- No production file in `openhcs/`, `benchmark/` or `scripts/` changes.
+
+### The flagged test
+
+`test_inline_source_bindings_dropdown_edit_flashes_only_changed_cell` is now
+`..._flashes_changed_cell_and_inheriting_cell` and expects two paths.
+The 3.0.0 author's claim holds. Observed with `.claude` probes on this branch:
+
+| Field | Raw | Resolved before the edit | Resolved after | Source |
+|---|---|---|---|---|
+| `source_bindings_config.source_filters` | the clause | `TextEquals` clause | `TextContains` clause | set here |
+| `step_source_bindings_config.source_filters` | `None` | `TextEquals` clause | `TextContains` clause | `SourceBindingsConfig` of the same object |
+
+`StepSourceBindingsConfig` inherits `SourceBindingsConfig`, and the step section
+sets no filters of its own, so the cell it previews changes value with the
+edit. A flash marks a field whose resolved value changed, so both cells flash.
+The result is the same with `scope_id="/plate"` and with none, and an attribute
+read on the object gives the same value. The old expectation held only because
+a state without a scope id did not see its own object as a layer. The test now
+also asserts the step section's raw value stays `None`, its resolved value is
+the edited clause, and its source is `SourceBindingsConfig`.
+
+## New-case experiments
+
+Resolved values, ObjectState 2.1.1 (U6 base) against 3.0.0 (this branch):
+
+- Default `PipelineConfig` (with and without scope id), `GlobalPipelineConfig`
+  and a `FunctionStep` state: 534 paths, no value differs.
+- For every field name held by several configs of `PipelineConfig`, set it on
+  one config and list which other fields change: with a scope id the two
+  versions agree on all 30 probes. Without a scope id, 3.0.0 now agrees with
+  the scoped result; 2.1.1 missed 8 sibling inheritances
+  (`streaming_defaults.{enabled,host,listen_host,persistent}` to the napari and
+  fiji streaming configs, `path_planning_config.output_dir_suffix` to
+  `step_materialization_config`, `napari_display_config.colormap`,
+  `fiji_display_config.{auto_contrast,lut}` to their streaming configs).
+- Production code builds a state without a scope id in two places, neither on
+  a pipeline path: `image_browser.NoDataset.config_state` (one viewer config,
+  nothing loaded) and the synthetic plate generator window. Compiled pipelines
+  use scoped states.
+
+## Tests
+
+Observed locally, CPython 3.12, the three submodules at the commits above.
+
+| Check | Result |
+|---|---|
+| formbridge `pytest tests -n 2` | 107 passed |
+| formbridge `ruff check src tests` | clean |
+| pyqt-reactive `pytest tests -n 2` | 593 passed, 1 failed (`test_published_runtime_is_not_overridden_by_a_source_candidate`: compares the installed `zmqruntime` with its bound; fails the same way at 0.7.0) |
+| pyqt-reactive `ruff check src/ --select E9,F63,F7,F822,F823`, `check_ruff_added_lines.py` | clean |
+| OpenHCS `tests/unit/pyqt_gui -n 2` | 899 passed |
+| OpenHCS 108 unit files that name ObjectState, lazy configs or `PipelineConfig`, with `test_validate_local_release_floors.py` and `agent/test_mcp_tool_contract.py` | 2388 passed, 11 failed, 3 skipped; the same 11 fail on the U6 base with ObjectState 2.1.1 (listed under "Out of scope") |
+| `test_compilation_session.py` (ledger item 9) | passed in that run |
+| `guardrails_local.sh` | `guardrails rc=0` (at `543c56eb8`) |
+| `SMOKE_SKIP_GUI=1 smoke_local.sh` | `smoke rc=0`, 5 passed (at `543c56eb8`) |
+| official30 parity, run 1 | 28/30 equivalent, not a valid run: `cp_tutorial_pixel_based_classification` was compiled by an execution server of another worktree (`agent-a05d025a19446fcd4`, `cannot import name 'LazyAnalysisConsolidationConfig'` from that worktree's file) and `cp_tutorial_translocation_final` was skipped after it |
+| official30 parity, run 2 | 29/30 equivalent, not a clean run either: `ExampleYeastColonies` ended with `OpenHCS ZMQ execution failed: Execution was cancelled` (nothing in this branch cancels; other agents' runs were active on the machine). It was equivalent in run 1, and the two cases that failed in run 1 are equivalent here, so every case was equivalent in one of the two runs and no value mismatch was seen in either |
+
+MCP tool contract: `test_mcp_tool_contract.py` passes with no schema file changed.
+
+## Done when
+
+The three PRs are open, each suite and lint passes, guardrails and smoke end
+`rc=0`, and parity is 30/30.
+
+Release order: ObjectState 3.0.0, formbridge 0.1.2, pyqt-reactive 0.7.1
+(empty its `requirements-ci.txt` first, as formbridge's), then OpenHCS with the
+submodules moved to the release tags.
+
+## Out of scope, found
+
+- `formbridge.forms.parameter_form_service.ParameterFormService.get_placeholder_text`
+  has no caller after the pyqt-reactive change (formbridge; public method, left).
+- `f"{manager.field_id}.{name}" if manager.field_id else name` is written out
+  in 7 pyqt-reactive modules; the form manager should own the field's path.
+- `tests/unit/test_artifact_publication_journey.py` (8),
+  `test_named_checkpoint_publication.py` (2) and
+  `test_bioformats_microscope_handler.py` (1) fail in the 108-file selection
+  below on the U6 base with ObjectState 2.1.1 as well; all but the last pass
+  when their files run alone.
+
+## Done / next
+
+Done: the three library and application changes, the three PRs, every check
+in the table above.
+
+Next: one clean 30/30 parity run on a quiet machine (`.claude/parity.sh` in the
+worktree, under `official30.lock`). Then move the submodules to the release
+tags once the libraries are published.

pyproject.toml

@@ -104,13 +104,13 @@ dependencies = [
     # External dependencies (PyPI versions for production builds)
     "zmqruntime>=0.6.3,<0.7",
     "pycodify>=0.2.1,<0.3",
-    "objectstate>=2.1.1,<3",
+    "objectstate>=3.0,<4",
     "python-introspect>=0.2.8,<0.3",
     "metaclass-registry>=0.3.0,<0.4",
     "arraybridge>=0.4.1,<0.5",
     "polystore>=0.6.1,<0.7",
-    "formbridge>=0.1.1,<0.2",
-    "pyqt-reactive>=0.7,<0.8",
+    "formbridge>=0.1.2,<0.2",
+    "pyqt-reactive>=0.7.1,<0.8",
 ]
 
 [project.optional-dependencies]

tests/unit/pyqt_gui/test_parameter_form_manager_cross_window.py

@@ -175,7 +175,6 @@ def record_invalidation(cls, **kwargs):
         classmethod(record_invalidation),
     )
     ObjectStateRegistry._deferred_invalidations.clear()
-    ObjectStateRegistry._deferred_invalidation_depth = 0
 
     with ObjectStateRegistry.defer_live_invalidations():
         ObjectStateRegistry.invalidate_by_type_and_scope("", str, "name")

tests/unit/pyqt_gui/test_source_bindings_editor.py

@@ -3127,7 +3127,7 @@ def test_inline_source_bindings_edit_flashes_changed_collection() -> None:
     manager.deleteLater()
 
 
-def test_inline_source_bindings_dropdown_edit_flashes_only_changed_cell() -> None:
+def test_inline_source_bindings_dropdown_edit_flashes_changed_cell_and_inheriting_cell() -> None:
     QtApplicationHarness.app()
     state = ObjectState(PipelineConfig())
     manager = ParameterFormManager(
@@ -3173,16 +3173,33 @@ def test_inline_source_bindings_dropdown_edit_flashes_only_changed_cell() -> Non
         )
         QApplication.processEvents()
 
-        expected_path = "source_bindings_config.source_filters[0].match_type"
-        assert [entry[0] for entry in registered] == [expected_path]
-        assert queued == [expected_path]
-        assert state.parameters["source_bindings_config.source_filters"] == (
+        # The edited cell, and the same cell of the step section: that section
+        # sets no filters of its own and shows the ones it inherits from
+        # source_bindings_config of the same PipelineConfig, so the value it
+        # resolves to changes with the edit while its own value stays unset.
+        edited_clauses = (
             SourceFilterClause(
                 SourceFileName,
                 TextContains,
                 "DNA",
             ),
         )
+        expected_paths = [
+            "source_bindings_config.source_filters[0].match_type",
+            "step_source_bindings_config.source_filters[0].match_type",
+        ]
+        assert [entry[0] for entry in registered] == expected_paths
+        assert queued == expected_paths
+        assert state.parameters["source_bindings_config.source_filters"] == edited_clauses
+        assert state.parameters["step_source_bindings_config.source_filters"] is None
+        assert (
+            state.get_resolved_value("step_source_bindings_config.source_filters")
+            == edited_clauses
+        )
+        assert state.get_provenance("step_source_bindings_config.source_filters") == (
+            state.scope_id,
+            SourceBindingsConfig,
+        )
         queued.clear()
         registered.clear()
         combo = widget.table("source_filters").cellWidget(

tests/unit/pyqt_gui/test_ui_agent_bridge.py

@@ -3735,9 +3735,6 @@ def reset_object_state_registry():
     ObjectStateRegistry._current_timeline = "main"
     ObjectStateRegistry._current_head = None
     ObjectStateRegistry._in_time_travel = False
-    ObjectStateRegistry._atomic_depth = 0
-    ObjectStateRegistry._atomic_label = None
-    ObjectStateRegistry._atomic_triggering_scope = None
     ObjectStateRegistry._token = 0
     yield
     BridgeSessions.close()

tests/unit/test_time_travel_navigation.py

@@ -26,9 +26,6 @@ def _reset_registry() -> None:
     ObjectStateRegistry._current_timeline = "main"
     ObjectStateRegistry._current_head = None
     ObjectStateRegistry._in_time_travel = False
-    ObjectStateRegistry._atomic_depth = 0
-    ObjectStateRegistry._atomic_label = None
-    ObjectStateRegistry._atomic_triggering_scope = None
 
 
 def test_time_travel_navigation_filters_unrelated_dirty_scopes() -> None: