Find a class's lazy counterpart through the base/lazy registry

[trissim]

Fixes #35

What changed

• abbreviation.__call__ no longer builds f"{LAZY_CONFIG_PREFIX}{cls.__name__}" and reads it off sys.modules[cls.__module__]. Nothing in the library looks a lazy class up by name any more.
• The base/lazy relation is owned by register_lazy_type_mapping (the existing owner; it already wrote both directions and already raised on a second counterpart in either direction). It is called once, when _create_lazy_dataclass_unified creates the lazy class. get_base_type_for_lazy and get_lazy_type_for_base are the two reads.
• A class's group abbreviation is declared once, on the class itself (declare_group_abbreviation), and read on demand by get_group_abbreviation, which maps a lazy class to its base through the registry and then walks the base's MRO for the nearest own declaration. @abbreviation(...) and abbreviation= on the global-default decorator write that same declaration, so the order of the two decorators does not matter.
• The lazy class name is produced in one place, the factory (LAZY_CLASS_NAME_PREFIX, which existed but was unused). The three other spellings (f"Lazy{...}" for nested fields, the make_lazy_simple default, LAZY_CONFIG_PREFIX in the decorator) are gone, and the name is no longer carried in _pending_injections. make_lazy_simple(base) with no name returns the lazy class the base already owns, whatever it is called.

Inventory before creation: searched for _lazy_type_registry, _base_to_lazy_registry, register_lazy_type_mapping, get_lazy_type_for_base, LAZY_CONFIG_PREFIX, LAZY_CLASS_NAME_PREFIX, sys.modules, GROUP_ABBREVIATIONS_REGISTRY, _abbreviation. No new registry was added; one was removed.

Acceptance criteria

Criterion	How it is met
No getattr(sys.modules[...], ...) lookup of a lazy class remains	Removed; test_no_lazy_class_is_looked_up_by_constructed_name pins it. The four remaining sys.modules[...] uses in lazy_factory.py are writes that export a generated class or decorator to its module.
A test defines a config class inside a function, applies the decorators in both orders, and gets the abbreviation on both the base and the lazy class	test_class_inside_function_gets_abbreviation_on_base_and_lazy, parametrised over both orders.

Further cases from the task: two same-named classes in one module, a renamed prefix (both in both decorator orders), the explicit abbreviation= parameter, a subclass with and without its own declaration (including through the inherit_as_none rebuild), two different abbreviations for one class raising, and the registry refusing a second counterpart in either direction.

Observed on origin/main (1fcdf12) with the new test file: 5 fail (same-named classes with @abbreviation innermost, renamed prefix in both orders, conflicting declarations, the source check). The function-local case passes on main when the class's module is importable, because the lazy class is still exported to that module; it failed only once a second class reused the name.

Paper results

• [P1] Proposition 3.1: the counterpart is identified by the registered class, not by a name that happens to match a convention. Two same-named classes are distinct keys.
• [P10] Proposition 6.1: the registry entry made at creation carries its source in both directions (lazy -> base, base -> lazy); an abbreviation is read from the declaring class instead of from a copy.
• [P10] Proposition 6.2 and [P9] Theorem 6.2: a second, different counterpart or a second, different abbreviation raises instead of last-writer-wins. Module-attribute names shared by same-named classes are no longer used to answer any query.

Behaviour changes for callers

• objectstate.lazy_factory.GROUP_ABBREVIATIONS_REGISTRY is removed. Read with get_group_abbreviation(config_type); declare with @abbreviation / abbreviation=.
  • formbridge/strategies/preview_formatting.py:26-34 imports it and walks it by hand; it must call objectstate.lazy_factory.get_group_abbreviation at release time.
  • pyqt-reactive/tests/test_preview_formatters.py:10,229,230,279 monkeypatches it; those tests must declare with @abbreviation instead.
• objectstate.lazy_factory.LAZY_CONFIG_PREFIX is removed. No use found in OpenHCS, pyqt-reactive, pycodify or formbridge.
• @abbreviation("x") together with abbreviation="y" on the same class now raises ValueError. Before, which one won depended on decorator order. OpenHCS uses abbreviation= at openhcs/core/config.py:596 and :987; neither class also carries @abbreviation as far as the grep shows.
• @abbreviation applied to a lazy class declares on its base class.
• make_lazy_simple(base) without a name no longer raises when the base already owns a lazy class with a non-default name; it returns that class. An explicit different name still raises.

Sites outside the library that build the lazy name

grep -rn 'LAZY_CONFIG_PREFIX\|f"Lazy{' over openhcs/, external/pyqt-reactive/src, external/pycodify/src: no hits. No caller builds the lazy name itself.

pycodify-generated .py pipelines

Not affected. pycodify spells a class by its __module__ and __qualname__ (pycodify/core.py:61, formatters.py) and has no knowledge of the Lazy prefix. Lazy classes keep the same name, the same __module__, and are still exported to that module under that name, so existing generated source imports the same objects.

Found outside this issue (not changed)

• lazy_factory.py, _inject_all_pending_fields: the global config class is recreated with make_dataclass from an almost empty namespace, so an @abbreviation on the global class itself (OpenHCS core/config.py:210, "gpc") does not reach the recreated class. On main the entry was likewise keyed by the discarded class.
• PREVIEW_LABEL_REGISTRY, FIELD_ABBREVIATIONS_REGISTRY and ALWAYS_VIEWABLE_FIELDS_REGISTRY still copy each base entry to the lazy class at decoration time, the same restated relation this PR removes for group abbreviations.
• The lazy class is still exported to the base's module by name; two same-named classes overwrite each other's export. Nothing in the library reads it back.

Tests

pytest -n 2: 251 passed. ruff check src/ --select E9,F63,F7,F82: clean. scripts/check_ruff_added_lines.py --base origin/main: no diagnostics on added lines. OpenHCS and formbridge were not run against this branch.

🤖 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.

[trissim]

Superseded by #50 (release/3.0.0-context-resolution), which merges this branch unchanged together with #41, #49, #40 and #42 for the 3.0.0 release. Left open; review and merge decisions happen on #50.