[trissim]

What was wrong

tests/unit/agent/test_declaration_selected_discovery.py::test_examplehuman_selected_source_uses_same_declaration_owner fails on main. The test is not stale. 91f05c26b (#763) made CellProfilerModule a CompilerPreparedAutoRegisterFamily whose hook reads every registry member, and _scan_module_registries also counted the members of every prepared family. Pipeline import prepares callables, so retrieving one knowledge-base example imported all 91 CellProfiler declarations.

Change

• CallablePreparation.prepare_for_serving (new, openhcs/core/processing_preparation.py) prepares every callable and then completes the lazy registries their modules use, once, through AutoRegisterRegistryPreparation.discover_module_registries (previously uncalled). RegistryService.prepare_in_current_process calls it. This keeps Complete CellProfiler measurement registry discovery before READY #763's requirement that registries are discovered before READY.
• Preparing one callable now only runs family hooks; it never reads a registry's members.
• Guard: tests/unit/test_callable_contract.py::test_only_serving_readiness_reads_the_members_of_a_registry.

Deletions

• CellProfilerModule.prepare_registered_family and its CompilerPreparedAutoRegisterFamily base: derived (serving readiness now completes the registry).
• AutoRegisterRegistryPreparationReport and its counts, _scan_module_registries: unused (only a test read prepared_family_count).

Verification (local, Linux, merged with origin/main ae6ddace5, head 6b9398566; the later commit changes the receipt only)

• The fixed test file with test_callable_contract.py and test_library_registry_discovery.py: 62 passed in 91.34s.
• Twelve preparation and registry neighbour files (test_processing_preparation, test_execution_server_startup_preparation, test_fork_compiler_preparation, test_preparation_admission, test_callable_abi_preparation, test_registry_owner_iteration, test_registry_library_boundary, test_colocalization_preparation_readiness, test_cellprofiler_kernel_preparation, test_cellprofiler_callable_kernel_preparation, test_cellprofiler_strategy_registries, test_func_registry_projection): 114 passed in 188.37s.
• guardrails_local.sh: guardrails rc=0. SMOKE_SKIP_GUI=1 smoke_local.sh: 5 passed, smoke rc=0.
• Official30 parity under flock: the full run did not pass as one run. It ended 1 failed with 27 of 30 cases equivalent (0 differences each). ExampleYeastPatches, cp_tutorial_advanced_segmentation_final and cp_tutorial_beginner_segmentation_final ended Execution was cancelled (two at compilation, one at execution), with no value comparison. Two other worktrees started execution servers on the same port (7849) 59 s and 72 s before the first two failures, and one of those servers logged a cancellation 5 s before the second. Those three cases, rerun alone under the lock through the test's own helper on a different port, were equivalent with 0 differences. So all 30 cases were observed equivalent, across two runs; a single clean 30/30 run was not observed.
• Not checked: GPU libraries installed (readiness now completes about 60 lazy registries; CPU-only locally); macOS and Windows; the full unit suite.

Out of scope, found

The parity lock covers parity runs only. _free_zmq_port_pair in tests/integration/test_cellprofiler_official30_zmq.py picks the same first free port as other ZMQ tests on the machine, so an unlocked ZMQ test in another worktree can cancel a parity case.

Note: PR #1213 adds the same receipt file; whichever merges second needs a trivial merge of that document.

R1 (item 4) · PR #1217 · guards 1/1 · production −52 +47 · tests −17 +60

🤖 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.